Merge branch 'erew123:alltalkbeta' into alltalkbeta
This commit is contained in:
9 files changed
+1245
-1056
No files matched your search
@@ -1,25 +1,25 @@
|
||||
coqui-tts>=0.24.1
|
||||
faster-whisper>=1.0.*
|
||||
faster-whisper>=1.0.0
|
||||
fuzzywuzzy
|
||||
gradio>=4.26.0
|
||||
importlib_metadata>=7.1.*
|
||||
importlib_metadata>=7.1.0
|
||||
inputimeout>=1.0.4
|
||||
Jinja2>=3.1.*
|
||||
Jinja2>=3.1.0
|
||||
librosa>=0.10.2.post1
|
||||
local_attention>=1.9.*
|
||||
local_attention>=1.9.0
|
||||
nvidia-cublas-cu11==11.11.3.6
|
||||
nvidia-cudnn-cu11==9.1.1.17
|
||||
onnxruntime-gpu>=1.18.*
|
||||
pydantic>=2.7.*
|
||||
python-ffmpeg>=2.0.*
|
||||
onnxruntime-gpu>=1.18.0
|
||||
pydantic>=2.7.0
|
||||
python-ffmpeg>=2.0.0
|
||||
python-Levenshtein>=0.25.1
|
||||
praat-parselmouth>=0.4.*
|
||||
pyworld>=0.3.*
|
||||
praat-parselmouth>=0.4.0
|
||||
pyworld>=0.3.0
|
||||
sounddevice>=0.4.7
|
||||
soundfile>=0.12.*
|
||||
soundfile>=0.12.0
|
||||
spacy>=3.7.1
|
||||
torchcrepe>=0.0.2
|
||||
tqdm>=4.66.*
|
||||
tqdm>=4.66.0
|
||||
unidic-lite>=1.0.8
|
||||
uvicorn>=0.29.0
|
||||
pillow==10.3.0
|
||||
@@ -43,4 +43,4 @@ hangul_romanize
|
||||
word2number
|
||||
git+https://github.com/huggingface/parler-tts.git
|
||||
piper-tts; sys_platform == "linux"
|
||||
trainer
|
||||
trainer
|
||||
@@ -1,25 +1,25 @@
|
||||
coqui-tts>=0.24.1
|
||||
faster-whisper>=1.0.*
|
||||
TTS
|
||||
faster-whisper>=1.0.0
|
||||
fuzzywuzzy
|
||||
gradio>=4.26.0
|
||||
importlib_metadata>=7.1.*
|
||||
importlib_metadata>=7.1.0
|
||||
inputimeout>=1.0.4
|
||||
Jinja2>=3.1.*
|
||||
Jinja2>=3.1.0
|
||||
librosa>=0.10.2.post1
|
||||
local_attention>=1.9.*
|
||||
local_attention>=1.9.0
|
||||
nvidia-cublas-cu11>=11.11.3.6
|
||||
nvidia-cudnn-cu11>=9.1.1.17
|
||||
onnxruntime-gpu>=1.18.*
|
||||
pydantic>=2.7.*
|
||||
python-ffmpeg>=2.0.*
|
||||
onnxruntime-gpu>=1.18.0
|
||||
pydantic>=2.7.0
|
||||
python-ffmpeg>=2.0.0
|
||||
python-Levenshtein>=0.25.1
|
||||
praat-parselmouth>=0.4.*
|
||||
pyworld>=0.3.*
|
||||
praat-parselmouth>=0.4.0
|
||||
pyworld>=0.3.0
|
||||
sounddevice>=0.4.7
|
||||
soundfile>=0.12.*
|
||||
soundfile>=0.12.0
|
||||
spacy>=3.7.1
|
||||
torchcrepe>=0.0.2
|
||||
tqdm>=4.66.*
|
||||
tqdm>=4.66.0
|
||||
unidic-lite>=1.0.8
|
||||
uvicorn>=0.29.0
|
||||
pillow==10.3.0
|
||||
@@ -32,4 +32,4 @@ git+https://github.com/huggingface/parler-tts.git
|
||||
piper-tts; sys_platform == "linux"
|
||||
piper-tts; sys_platform == "darwin"
|
||||
piper-phonemize==1.1.0; sys_platform == "darwin"
|
||||
trainer
|
||||
trainer
|
||||
@@ -1,25 +1,25 @@
|
||||
coqui-tts>=0.24.1
|
||||
faster-whisper>=1.0.*
|
||||
faster-whisper>=1.0.0
|
||||
fuzzywuzzy
|
||||
gradio>=4.26.0
|
||||
importlib_metadata>=7.1.*
|
||||
importlib_metadata>=7.1.0
|
||||
inputimeout>=1.0.4
|
||||
Jinja2>=3.1.*
|
||||
Jinja2>=3.1.0
|
||||
librosa>=0.10.2.post1
|
||||
local_attention>=1.9.*
|
||||
local_attention>=1.9.0
|
||||
nvidia-cublas-cu11>=11.11.3.6
|
||||
nvidia-cudnn-cu11>=9.1.1.17
|
||||
onnxruntime-gpu>=1.18.*
|
||||
pydantic>=2.7.*
|
||||
python-ffmpeg>=2.0.*
|
||||
onnxruntime-gpu>=1.18.0
|
||||
pydantic>=2.7.0
|
||||
python-ffmpeg>=2.0.0
|
||||
python-Levenshtein>=0.25.1
|
||||
praat-parselmouth>=0.4.*
|
||||
pyworld>=0.3.*
|
||||
praat-parselmouth>=0.4.0
|
||||
pyworld>=0.3.0
|
||||
sounddevice>=0.4.7
|
||||
soundfile>=0.12.*
|
||||
soundfile>=0.12.0
|
||||
spacy>=3.7.1
|
||||
torchcrepe>=0.0.2
|
||||
tqdm>=4.66.*
|
||||
tqdm>=4.66.0
|
||||
unidic-lite>=1.0.8
|
||||
uvicorn>=0.29.0
|
||||
pypinyin
|
||||
@@ -31,4 +31,4 @@ git+https://github.com/huggingface/parler-tts.git
|
||||
piper-tts; sys_platform == "linux"
|
||||
piper-tts; sys_platform == "darwin"
|
||||
piper-phonemize==1.1.0; sys_platform == "darwin"
|
||||
trainer
|
||||
trainer
|
||||
@@ -1,18 +1,33 @@
|
||||
{
|
||||
"first_start_model": "parler_tts_mini_v0.1",
|
||||
"models": [
|
||||
{
|
||||
"model_name": "parler_tts_mini_v0.1",
|
||||
"files_to_download": [
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/generation_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/preprocessor_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/special_tokens_map.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/spiece.model?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/tokenizer.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/tokenizer_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler_tts_mini_v0.1/resolve/main/model.safetensors?download=true"
|
||||
]
|
||||
}
|
||||
]
|
||||
{
|
||||
"first_start_model": "parler-tts-mini-v1",
|
||||
"models": [
|
||||
{
|
||||
"model_name": "parler-tts-mini-v1",
|
||||
"files_to_download": [
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/generation_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/preprocessor_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/special_tokens_map.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/spiece.model?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/tokenizer.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/tokenizer_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-mini-v1/resolve/main/model.safetensors?download=true"
|
||||
]
|
||||
},
|
||||
{
|
||||
"model_name": "parler-tts-large-v1",
|
||||
"files_to_download": [
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/generation_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/preprocessor_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/special_tokens_map.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/spiece.model?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/tokenizer.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/tokenizer_config.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/model.safetensors.index.json?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/model-00001-of-00002.safetensors?download=true",
|
||||
"https://huggingface.co/parler-tts/parler-tts-large-v1/resolve/main/model-00002-of-00002.safetensors?download=true"
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
@@ -1,463 +1,470 @@
|
||||
###############################################
|
||||
# DONT CHANGE # These are base imports needed #
|
||||
###############################################
|
||||
import os
|
||||
import sys
|
||||
import json
|
||||
import time
|
||||
import torch
|
||||
import logging
|
||||
from pathlib import Path
|
||||
from fastapi import (HTTPException)
|
||||
logging.disable(logging.WARNING)
|
||||
###############################################
|
||||
# DONT CHANGE # Get Pytorch & Python versions #
|
||||
###############################################
|
||||
pytorch_version = torch.__version__
|
||||
cuda_version = torch.version.cuda
|
||||
major, minor, micro = sys.version_info[:3]
|
||||
python_version = f"{major}.{minor}.{micro}"
|
||||
try:
|
||||
import deepspeed
|
||||
deepspeed_available = True
|
||||
except ImportError:
|
||||
deepspeed_available = False
|
||||
pass
|
||||
|
||||
#############################################################################################################
|
||||
#############################################################################################################
|
||||
# CHANGE ME # Run any specifc imports, requirements or setup any global vaiables needed for this TTS Engine #
|
||||
#############################################################################################################
|
||||
#############################################################################################################
|
||||
# In this section you will import any imports that your specific TTS Engine will use. You will provide any
|
||||
# start-up errors for those bits, as if you were starting up a normal Python script. Note the logging.disable
|
||||
# a few lines up from here, you may want to # that out while debugging!
|
||||
from parler_tts import ParlerTTSForConditionalGeneration
|
||||
from transformers import AutoTokenizer
|
||||
import soundfile as sf
|
||||
|
||||
#################################################################################################################################
|
||||
# DONT CHANGE # Do not change the Class name from tts_class as this is what will be imported into the main tts_server.py script #
|
||||
#################################################################################################################################
|
||||
class tts_class:
|
||||
def __init__(self):
|
||||
########################################################################
|
||||
# DONT CHANGE # Sets up the base variables required for any tts engine #
|
||||
########################################################################
|
||||
self.branding = None
|
||||
self.this_dir = Path(__file__).parent.resolve() # Sets up self.this_dir as a variable for the folder THIS script is running in.
|
||||
self.main_dir = Path(__file__).parent.parent.parent.parent.resolve() # Sets up self.main_dir as a variable for the folder AllTalk is running in
|
||||
self.device = "cuda" if torch.cuda.is_available() else "cpu" # Sets up self.device to cuda if torch exists with Nvidia/CUDA, otherwise sets to cpu
|
||||
self.cuda_is_available = torch.cuda.is_available() # Sets up cuda_is_available as a True/False to track if Nvidia/CUDA was found on the system
|
||||
self.tts_generating_lock = False # Used to lock and unlock the tts generation process at the start/end of tts generation.
|
||||
self.tts_stop_generation = False # Used in conjunction with tts_generating_lock to call for a stop to the current generation. If called (set True) it needs to be set back to False when generation has been stopped.
|
||||
self.tts_narrator_generatingtts = False # Used to track if the current tts processes is narrator based. This can be used in conjunction with lowvram and device to avoid moving model between GPU(CUDA)<>RAM(CPU) each chunk of narrated text generated.
|
||||
self.model = None # If loading a model into CUDA/VRAM/RAM "model" is used as the variable name to load and interact with (see the XTTS model_engine script for examples.)
|
||||
self.is_tts_model_loaded = False # Used to track if a model is actually loaded in and error/fail things like TTS generation if its False
|
||||
self.current_model_loaded = None # Stores the name of the currenly loaded in model
|
||||
self.available_models = None # List of available models found by "def scan_models_folder"
|
||||
self.setup_has_run = False # Tracks if async def setup(self) has run, by setting to True, so that the /api/ready endpoint can provide a "Ready" status
|
||||
##############################################################################################
|
||||
# DONT CHANGE # Load in a list of the available TTS engines and the currently set TTS engine #
|
||||
##############################################################################################
|
||||
tts_engines_file = os.path.join(self.main_dir, "system", "tts_engines", "tts_engines.json")
|
||||
with open(tts_engines_file, "r") as f:
|
||||
tts_engines_data = json.load(f)
|
||||
self.engines_available = [engine["name"] for engine in tts_engines_data["engines_available"]] # A list of ALL the TTS engines available to be loaded by AllTalk
|
||||
self.engine_loaded = tts_engines_data["engine_loaded"] # In "tts_engines.json" what is the currently set TTS engine loading into AllTalk
|
||||
self.selected_model = tts_engines_data["selected_model"] # In "tts_engines.json" what is the currently set TTS model loading into AllTalk
|
||||
############################################################################
|
||||
# DONT CHANGE # Pull out all the settings for the currently set TTS engine #
|
||||
############################################################################
|
||||
with open(os.path.join(self.this_dir, "model_settings.json"), "r") as f:
|
||||
tts_model_loaded = json.load(f)
|
||||
# Access the model details
|
||||
self.manufacturer_name = tts_model_loaded["model_details"]["manufacturer_name"] # The company/person/body that generated the TTS engine/models etc
|
||||
self.manufacturer_website = tts_model_loaded["model_details"]["manufacturer_website"] # The website of the company/person/body where people can find more information
|
||||
# Access the features the model is capable of:
|
||||
self.audio_format = tts_model_loaded["model_capabilties"]["audio_format"] # This details the audio format your TTS engine is set to generate TTS in e.g. wav, mp3, flac, opus, acc, pcm. Please use only 1x format.
|
||||
self.deepspeed_capable = tts_model_loaded["model_capabilties"]["deepspeed_capable"] # Is your model capable of DeepSpeed
|
||||
self.deepspeed_available = 'deepspeed' in globals() # When we did the import earlier, at the top of this script, was DeepSpeed available for use
|
||||
self.generationspeed_capable = tts_model_loaded["model_capabilties"]["generationspeed_capable"] # Does this TTS engine support changing the speed of the generated TTS
|
||||
self.languages_capable = tts_model_loaded["model_capabilties"]["languages_capable"] # Are the actual models themselves capable of generating in multiple languages OR is each model language specific
|
||||
self.lowvram_capable = tts_model_loaded["model_capabilties"]["lowvram_capable"] # Is this engine capable of using low VRAM (moving the model between CPU And GPU memory)
|
||||
self.multimodel_capable = tts_model_loaded["model_capabilties"]["multimodel_capable"] # Is there just the one model or are there multiple models this engine supports.
|
||||
self.repetitionpenalty_capable = tts_model_loaded["model_capabilties"]["repetitionpenalty_capable"] # Is this TTS engine capable of changing the repititon penalty
|
||||
self.streaming_capable = tts_model_loaded["model_capabilties"]["streaming_capable"] # Is this TTS engine capabale of generating streaming audio
|
||||
self.temperature_capable = tts_model_loaded["model_capabilties"]["temperature_capable"] # Is this TTS engine capable of changing the temperature of the models
|
||||
self.multivoice_capable = tts_model_loaded["model_capabilties"]["multivoice_capable"] # Are the models multi-voice or single vocice models
|
||||
self.pitch_capable = tts_model_loaded["model_capabilties"]["pitch_capable"] # Is this TTS engine capable of changing the pitch of the genrated TTS
|
||||
# Access the current enginesettings
|
||||
self.def_character_voice = tts_model_loaded["settings"]["def_character_voice"] # What is the current default main/character voice that will be used if no voice specified.
|
||||
self.def_narrator_voice = tts_model_loaded["settings"]["def_narrator_voice"] # What is the current default narrator voice that will be used if no voice specified.
|
||||
self.deepspeed_enabled = tts_model_loaded["settings"]["deepspeed_enabled"] # If its available, is DeepSpeed enabled for the TTS engine
|
||||
self.engine_installed = tts_model_loaded["settings"]["engine_installed"] # Has the TTS engine been setup/installed (not curently used)
|
||||
self.generationspeed_set = tts_model_loaded["settings"]["generationspeed_set"] # What is the set/stored speed for generation.
|
||||
self.lowvram_enabled = tts_model_loaded["settings"]["lowvram_enabled"] # If its available, is LowVRAM enabled for the TTS engine
|
||||
# Check if someone has enabled lowvram on a system that's not CUDA enabled
|
||||
self.lowvram_enabled = False if not torch.cuda.is_available() else self.lowvram_enabled # If LowVRAM is mistakenly set and CUDA is not available, this will force it back off
|
||||
self.repetitionpenalty_set = tts_model_loaded["settings"]["repetitionpenalty_set"] # What is the currenly set repitition policy of the model (If it support repetition)
|
||||
self.temperature_set = tts_model_loaded["settings"]["temperature_set"] # What is the currenly set temperature of the model (If it support temp)
|
||||
self.pitch_set = tts_model_loaded["settings"]["pitch_set"] # What is the currenly set pitch of the model (If it support temp)
|
||||
# Gather the OpenAI API Voice Mappings
|
||||
self.openai_alloy = tts_model_loaded["openai_voices"]["alloy"] # The TTS engine voice that will be mapped to Open AI Alloy voice
|
||||
self.openai_echo = tts_model_loaded["openai_voices"]["echo"] # The TTS engine voice that will be mapped to Open AI Echo voice
|
||||
self.openai_fable = tts_model_loaded["openai_voices"]["fable"] # The TTS engine voice that will be mapped to Open AI Fable voice
|
||||
self.openai_nova = tts_model_loaded["openai_voices"]["nova"] # The TTS engine voice that will be mapped to Open AI Nova voice
|
||||
self.openai_onyx = tts_model_loaded["openai_voices"]["onyx"] # The TTS engine voice that will be mapped to Open AI Onyx voice
|
||||
self.openai_shimmer = tts_model_loaded["openai_voices"]["shimmer"] # The TTS engine voice that will be mapped to Open AI Shimmer voice
|
||||
###################################################################
|
||||
# DONT CHANGE # Load params and api_defaults from confignew.json #
|
||||
###################################################################
|
||||
# Define the path to the confignew.json file
|
||||
configfile_path = self.main_dir / "confignew.json"
|
||||
# Load config file and get settings
|
||||
with open(configfile_path, "r") as configfile:
|
||||
configfile_data = json.load(configfile)
|
||||
self.branding = configfile_data.get("branding", "") # Sets up self.branding for outputting the name stored in the "confgnew.json" file, as used in print statements.
|
||||
self.params = configfile_data # Loads in the curent "confgnew.json" file to self.params.
|
||||
self.debug_tts = configfile_data.get("debugging").get("debug_tts") # Can be used within this script as a True/False flag for generally debugging the TTS generation process.
|
||||
self.debug_tts_variables = configfile_data.get("debugging").get("debug_tts_variables") # Can be used within this script as a True/False flag for generally debugging variables (if you wish).
|
||||
################################################################
|
||||
# DONT CHANGE # Print out Python, CUDA, DeepSpeed versions ####
|
||||
################################################################
|
||||
def printout_versions(self):
|
||||
if deepspeed_available:
|
||||
print(f"[{self.branding}ENG] \033[92mDeepSpeed version :\033[93m",deepspeed.__version__,"\033[0m")
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[92mDeepSpeed version :\033[93m Not available\033[0m")
|
||||
print(f"[{self.branding}ENG] \033[92mPython Version :\033[93m {python_version}\033[0m")
|
||||
print(f"[{self.branding}ENG] \033[92mPyTorch Version :\033[93m {pytorch_version}\033[0m")
|
||||
if cuda_version is None:
|
||||
print(f"[{self.branding}ENG] \033[92mCUDA Version :\033[91m Not available\033[0m")
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[92mCUDA Version :\033[93m {cuda_version}\033[0m")
|
||||
print(f"[{self.branding}ENG]")
|
||||
return
|
||||
|
||||
###################################################################################
|
||||
###################################################################################
|
||||
# CHANGE ME # Inital setup of the model and engine. Called when the script starts #
|
||||
###################################################################################
|
||||
###################################################################################
|
||||
# In here you will add code to load in your model and do its inital setup. So you
|
||||
# may be calling your model loader via handle_tts_method_change or if your TTS
|
||||
# engine doesnt actually load a model into CUDA or System RAM, you may be doing
|
||||
# Something to fake its start-up.
|
||||
async def setup(self):
|
||||
self.printout_versions()
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
self.available_models = self.scan_models_folder()
|
||||
if self.selected_model:
|
||||
tts_model = f"{self.selected_model}"
|
||||
if tts_model in self.available_models:
|
||||
await self.handle_tts_method_change(tts_model)
|
||||
self.current_model_loaded = tts_model
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
else:
|
||||
self.current_model_loaded = "No Models Available"
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: Selected model '{self.selected_model}' not found in the models folder.")
|
||||
self.setup_has_run = True # Set to True, so that the /api/ready endpoint can provide a "Ready" status
|
||||
|
||||
##################################
|
||||
##################################
|
||||
# CHANGE ME # Low VRAM Swapping #
|
||||
##################################
|
||||
##################################
|
||||
# If your model does load into CUDA and you want to support LowVRAM, aka moving the model
|
||||
# on the fly between CUDA and System RAM on each generation request, you will be adding
|
||||
# The code in here to do it. See the XTTS tts engine for an example. Piper however, doesnt
|
||||
# Load into VRAM or System RAM, also low vram is set globally disabled by the model_settings.JSON
|
||||
# file, so this would never get called anywyay, however, we still need to keep an empty
|
||||
# function in place. Piper TTS uses "pass" telling function to just exit out cleanly if called.
|
||||
# However, its quite a simple check along the lines of "if CUDA is available and model is
|
||||
# in X place, then send it to Y place (or Y to X).
|
||||
async def handle_lowvram_change(self):
|
||||
if torch.cuda.is_available():
|
||||
if self.device == "cuda":
|
||||
self.device = "cpu"
|
||||
self.model.to(self.device)
|
||||
torch.cuda.empty_cache()
|
||||
else:
|
||||
self.device == "cpu"
|
||||
self.device = "cuda"
|
||||
self.model.to(self.device)
|
||||
|
||||
########################################
|
||||
########################################
|
||||
# CHANGE ME # DeepSpeed model loading #
|
||||
########################################
|
||||
########################################
|
||||
# If the model supports CUDA and DeepSpeed, this is where you can handle re-loading
|
||||
# the model as/when DeepSpeed is enabled/selected in the user interface. If your
|
||||
# TTS model doesnt support DeepSpeed, then it should be globally set in your
|
||||
# model_settings.JSON and this function will never get called, however it still needs
|
||||
# to exist as a function.
|
||||
async def handle_deepspeed_change(self, value):
|
||||
if value:
|
||||
# DeepSpeed enabled
|
||||
print(f"[{self.branding}ENG] \033[93mDeepSpeed Activating\033[0m")
|
||||
await self.unload_model()
|
||||
self.params["tts_method_api_local"] = False
|
||||
self.params["tts_method_xtts_local"] = True
|
||||
self.deepspeed_enabled = True
|
||||
await self.setup()
|
||||
else:
|
||||
# DeepSpeed disabled
|
||||
print(f"[{self.branding}ENG] \033[93mDeepSpeed De-Activating\033[0m")
|
||||
self.deepspeed_enabled = False
|
||||
await self.unload_model()
|
||||
await self.setup()
|
||||
return value # Return new checkbox value
|
||||
|
||||
##############################################################################################################################################
|
||||
##############################################################################################################################################
|
||||
# CHANGE ME # scan for available models/voices that are relevant to this TTS engine # XTTS is very unusal as it has 2x model loading methods #
|
||||
##############################################################################################################################################
|
||||
##############################################################################################################################################
|
||||
# This function looks and reports back the list of possible models your TTS engine can
|
||||
# load in. Some TTS engines have multiple models they can load and you will want to use
|
||||
# code for checking if the models are in the correct location/placement within the disk
|
||||
# the correct files exst per model etc (check XTTS for an example of this). Some models
|
||||
# like Piper, the actual models are the voices, so in the Piper scan_models_folder
|
||||
# function, we fake the only model being available as {'piper': 'piper'} aka, model name
|
||||
# then engine name, then we use the voices_file_list to populate the models as available
|
||||
# voices that can be selected in the interface.
|
||||
# If no models are found, we return "No Models Available" and continue on with the script.
|
||||
def scan_models_folder(self):
|
||||
models_folder = self.main_dir / "models" / "parler" # Edit to match the name of your folder where voices are stored
|
||||
print("models_folder is:", models_folder) if self.debug_tts else None
|
||||
self.available_models = {}
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# Check if model_path exists
|
||||
if not os.path.exists(models_folder):
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: The specified model path '{models_folder}' does not exist.")
|
||||
self.available_models = {'No Models Available': 'parler'} # Change the name in here to your TTS engine name
|
||||
return self.available_models
|
||||
required_files = ["config.json", "generation_config.json", "model.safetensors", "preprocessor_config.json", "special_tokens_map.json", "spiece.model", "tokenizer.json", "tokenizer_config.json"]
|
||||
for subfolder in models_folder.iterdir():
|
||||
if subfolder.is_dir():
|
||||
model_name = subfolder.name
|
||||
print("model_name is:", model_name) if self.debug_tts else None
|
||||
if all(subfolder.joinpath(file).exists() for file in required_files):
|
||||
self.available_models[f"parler - {model_name}"] = "parler"
|
||||
print("self.available_models is:", self.available_models) if self.debug_tts else None
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
else:
|
||||
self.available_models = {'No Models Available': 'parler'} # Change the name in here to your TTS engine name
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: Model folder '{model_name}' is missing required")
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: files or the folder does not exist.")
|
||||
return self.available_models
|
||||
|
||||
#############################################################
|
||||
#############################################################
|
||||
# CHANGE ME # POPULATE FILES LIST FROM VOICES DIRECTORY ####
|
||||
#############################################################
|
||||
#############################################################
|
||||
# This function looks and reports back the list of possible voice your TTS engine can
|
||||
# load in. Some TTS engines the voices are wav file samples (XTTS), some are models
|
||||
# (Piper) and some are text (Parler) thats stored in a JSON file. We just need to
|
||||
# populate the "voices" variable somehow and if no voices are found, we return
|
||||
# "No Voices Found" back to the interface/api.
|
||||
def voices_file_list(self):
|
||||
try:
|
||||
voices = []
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
voices_file = os.path.join(self.main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices = json.load(f)
|
||||
return [voice["voice_name"] for voice in voices["voices"]]
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
if not voices:
|
||||
return ["No Voices Found"] # Return a list with {'No Voices Found'} if there are no voices/voice models.
|
||||
return voices # Return a list of models in the format {'engine name': 'model 1', 'engine name': 'model 2", etc}
|
||||
except Exception as e:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: Voices/Voice Models not found. Cannot load a list of voices.")
|
||||
print(f"[{self.branding}ENG]")
|
||||
return ["No Voices Found"]
|
||||
|
||||
#############################
|
||||
#############################
|
||||
# CHANGE ME # Model loading #
|
||||
#############################
|
||||
#############################
|
||||
# This function will handle the loading of your model, into VRAM/CUDA, System RAM or whatever.
|
||||
# In XTTS, which has 2x model loader types, there are 2x loaders. They are called by "def handle_tts_method_change"
|
||||
# In Piper we fake a model loader as Piper doesnt actually load a model into CUDA/System RAM as such. So, in that
|
||||
# situation, api_manual_load_model is kind of a blank function. Though we do set self.is_tts_model_loaded = True
|
||||
# as this is used elsewhere in the scripts to confirm that a model is available to be used for TTS generation.
|
||||
# We always check for "No Models Available" being sent as that means we are trying to load in a model that
|
||||
# doesnt exist/wasnt found on script start-up e.g. someone deleted the model from the folder or something.
|
||||
async def api_manual_load_model(self, model_name):
|
||||
if "No Models Available" in self.available_models:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: No models for this TTS engine were found to load. Please download a model.")
|
||||
return
|
||||
model_path = self.main_dir / "models" / "parler" / model_name
|
||||
print("model_path is:", model_path) if self.debug_tts_variables else None
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
self.model = ParlerTTSForConditionalGeneration.from_pretrained(model_path).to(self.device, dtype=torch.float32)
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(model_path)
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
self.is_tts_model_loaded = True
|
||||
return self.model
|
||||
|
||||
###############################
|
||||
###############################
|
||||
# CHANGE ME # Model unloading #
|
||||
###############################
|
||||
###############################
|
||||
# This function will handle the UN-loading of your model, from VRAM/CUDA, System RAM or whatever.
|
||||
# In XTTS, that model loads into CUDA/System Ram, so when we swap models, we want to unload the current model
|
||||
# free up the memory and then load in the new model to VRAM/CUDA. On the flip side of that, Piper doesnt
|
||||
# doesnt load into memory, so we just need to put a fake function here that doesnt really do anything
|
||||
# other than set "self.is_tts_model_loaded = False", which would be set back to true by the model loader.
|
||||
# So look at the Piper model_engine.py if you DONT need to unload models.
|
||||
async def unload_model(self):
|
||||
self.is_tts_model_loaded = False
|
||||
if not self.current_model_loaded == None:
|
||||
print(f"[{self.branding}ENG] \033[94mUnloading model \033[0m") if self.debug_tts else None
|
||||
if hasattr(self, 'model'):
|
||||
del self.model
|
||||
if torch.cuda.is_available():
|
||||
torch.cuda.empty_cache()
|
||||
return None
|
||||
|
||||
###################################################################################################################################
|
||||
###################################################################################################################################
|
||||
# CHANGE ME # Model changing. Unload out old model and load in a new one # XTTS is very unusal as it has 2x model loading methods #
|
||||
###################################################################################################################################
|
||||
###################################################################################################################################
|
||||
# This function is your central model loading/unloading handler that deals with the above functions as necesary, to call loading, unloading,
|
||||
# swappng DeepSpeed, Low vram etc. This function gets called with a "engine name - model name" type call. In XTTS, because there are 2x
|
||||
# model loader types, (XTTS and APILocal), we take tts_method and split the "engine name - model name" into a loader type and the model
|
||||
# that it needs to load in and then we call the correct loader function. Whereas in Piper, which doesnt load models into memory at all,
|
||||
# we just have a fake function that doesnt really do anything. We always check to see if the model name has "No Models Available" in the
|
||||
# name thats sent over, just to catch any potential errors. We display the start load time and end load time. Thats about it.
|
||||
async def handle_tts_method_change(self, tts_method):
|
||||
generate_start_time = time.time() # Record the start time of loading the model
|
||||
if "No Models Available" in self.available_models:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: No models for this TTS engine were found to load. Please download a model.")
|
||||
return False
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
await self.unload_model()
|
||||
if tts_method.startswith("parler"):
|
||||
model_name = tts_method.split(" - ")[1]
|
||||
print(f"[{self.branding}ENG]\033[94m Model/Engine :\033[93m {model_name}\033[94m loading into\033[93m", self.device,"\033[0m")
|
||||
self.model = await self.api_manual_load_model(model_name)
|
||||
self.current_model_loaded = f"parler - {model_name}"
|
||||
else:
|
||||
self.current_model_loaded = None
|
||||
return False
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
generate_end_time = time.time() # Create an end timer for calculating load times
|
||||
generate_elapsed_time = generate_end_time - generate_start_time # Calculate start time minus end time
|
||||
print(f"[{self.branding}ENG] \033[94mLoad time :\033[93m {generate_elapsed_time:.2f} seconds.\033[0m") # Print out the result of the load time
|
||||
return True
|
||||
|
||||
##########################################################################################################################################
|
||||
##########################################################################################################################################
|
||||
# CHANGE ME # Model changing. Unload out old model and load in a new one # XTTS is very unusal as it has 2x model TTS generation methods #
|
||||
##########################################################################################################################################
|
||||
##########################################################################################################################################
|
||||
# In here all the possible options are sent over (text, voice to use, lanugage, speed etc etc) and its up to you how you use them, or not.
|
||||
# obviously if your TTS engine doesnt support speed for example, generationspeed_capable should be set False in your model_settings.JSON file
|
||||
# and a fake "generationspeed_set" value should be set. This allows a fake value to be sent over from the main script, even though it
|
||||
# wouldnt actually ever be used in the generation below. Nonethless all these values, real or just whats inside the configuration file
|
||||
# will be sent over for use.
|
||||
# Setting the xxxxxxx_capabale in the model_settings.JSON file, will enable/disable them being selectable by the user. For example, if you
|
||||
# set "generationspeed_capable" as false in the model_settings.JSON file, a user will not be able to select OR set the setting for
|
||||
# generation speed.
|
||||
# One thing to note is that we HAVE to have something in this generation request that is synchronous from the way its called, which means
|
||||
# we have to have an option for Streaming, even if our TTS engine doesnt support streaming. So in that case, we would set streaming_capable
|
||||
# as false in our model_settings.JSON file, meaning streaming will never be called. However, we have to put a fake streaming routine in our
|
||||
# function below (or a real function if it does support streaming of course). Parler has an example of a fake streaming function, which is
|
||||
# very clearly highlighted in its model_engine.py script.
|
||||
# Piper TTS, which uses command line based calls and therefore has different ones for Windows and Linux/Mac, has an example of doing this
|
||||
# within its model_engine.py file.
|
||||
async def generate_tts(self, text, voice, language, temperature, repetition_penalty, speed, pitch, output_file, streaming):
|
||||
print(f"[{self.branding}Debug] Entered model_engine.py generate_tts function") if self.debug_tts else None
|
||||
if not self.is_tts_model_loaded: # Check if a model is loaded and error out if not.
|
||||
error_message = f"[{self.branding}ENG] \033[91mError\033[0m: You currently have no TTS model loaded."
|
||||
print(error_message)
|
||||
raise HTTPException(status_code=400, detail="You currently have no TTS model loaded.") # Raise an exception with a meaningful HTTP status code
|
||||
self.tts_generating_lock = True # Set the tts_generating lock to True, which stops other generation requests being sent into the pipeline
|
||||
print(f"[{self.branding}Debug] Checking low VRAM") if self.debug_tts else None
|
||||
if self.lowvram_enabled and self.device == "cpu": # If necessary, move the model out of System Ram to VRAM
|
||||
print(f"[{self.branding}Debug] Switching device") if self.debug_tts else None
|
||||
await self.handle_lowvram_change()
|
||||
print(f"[{self.branding}Debug] Setting a generate time") if self.debug_tts else None
|
||||
generate_start_time = time.time() # Record the start time of generating TTS
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
voices_file = os.path.join(self.main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices_data = json.load(f)
|
||||
for voice_data in voices_data["voices"]:
|
||||
if voice_data["voice_name"] == voice:
|
||||
description = voice_data["description"]
|
||||
break
|
||||
else:
|
||||
description = f"A {voice} voice speaking in {language}."
|
||||
else:
|
||||
description = f"A {voice} voice speaking in {language}."
|
||||
input_ids = self.tokenizer(description, return_tensors="pt").input_ids.to(self.device)
|
||||
prompt_input_ids = self.tokenizer(text, return_tensors="pt").input_ids.to(self.device)
|
||||
generation = self.model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids).to(torch.float32)
|
||||
audio_arr = generation.cpu().numpy().squeeze()
|
||||
sf.write(output_file, audio_arr, self.model.config.sampling_rate)
|
||||
|
||||
# Fake Streaming function here
|
||||
wavs = None # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
if streaming and wavs is not None: # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
# Streaming-specific operations # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
for wav_chunk in wavs: # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
yield wav_chunk.numpy().tobytes() # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
generate_end_time = time.time() # Record the end time to generate TTS
|
||||
generate_elapsed_time = generate_end_time - generate_start_time
|
||||
print(f"[{self.branding}GEN] \033[94mTTS Generate: \033[93m{generate_elapsed_time:.2f} seconds. \033[94mLowVRAM: \033[33m{self.lowvram_enabled} \033[94mDeepSpeed: \033[33m{self.deepspeed_enabled}\033[0m")
|
||||
if self.lowvram_enabled and self.device == "cuda" and self.tts_narrator_generatingtts == False:
|
||||
await self.handle_lowvram_change()
|
||||
self.tts_generating_lock = False # Unlock the TTS generation queue to allow TTS generation requests to come in again.
|
||||
print(f"[{self.branding}Debug] generate_tts function completed") if self.debug_tts else None
|
||||
return
|
||||
|
||||
|
||||
|
||||
|
||||
###############################################
|
||||
# DONT CHANGE # These are base imports needed #
|
||||
###############################################
|
||||
import os
|
||||
import sys
|
||||
import json
|
||||
import time
|
||||
import torch
|
||||
import logging
|
||||
from pathlib import Path
|
||||
from fastapi import (HTTPException)
|
||||
logging.disable(logging.WARNING)
|
||||
###############################################
|
||||
# DONT CHANGE # Get Pytorch & Python versions #
|
||||
###############################################
|
||||
pytorch_version = torch.__version__
|
||||
cuda_version = torch.version.cuda
|
||||
major, minor, micro = sys.version_info[:3]
|
||||
python_version = f"{major}.{minor}.{micro}"
|
||||
try:
|
||||
import deepspeed
|
||||
deepspeed_available = True
|
||||
except ImportError:
|
||||
deepspeed_available = False
|
||||
pass
|
||||
|
||||
#############################################################################################################
|
||||
#############################################################################################################
|
||||
# CHANGE ME # Run any specifc imports, requirements or setup any global vaiables needed for this TTS Engine #
|
||||
#############################################################################################################
|
||||
#############################################################################################################
|
||||
# In this section you will import any imports that your specific TTS Engine will use. You will provide any
|
||||
# start-up errors for those bits, as if you were starting up a normal Python script. Note the logging.disable
|
||||
# a few lines up from here, you may want to # that out while debugging!
|
||||
from parler_tts import ParlerTTSForConditionalGeneration
|
||||
from transformers import AutoTokenizer
|
||||
import soundfile as sf
|
||||
|
||||
#################################################################################################################################
|
||||
# DONT CHANGE # Do not change the Class name from tts_class as this is what will be imported into the main tts_server.py script #
|
||||
#################################################################################################################################
|
||||
class tts_class:
|
||||
def __init__(self):
|
||||
########################################################################
|
||||
# DONT CHANGE # Sets up the base variables required for any tts engine #
|
||||
########################################################################
|
||||
self.branding = None
|
||||
self.this_dir = Path(__file__).parent.resolve() # Sets up self.this_dir as a variable for the folder THIS script is running in.
|
||||
self.main_dir = Path(__file__).parent.parent.parent.parent.resolve() # Sets up self.main_dir as a variable for the folder AllTalk is running in
|
||||
self.device = "cuda" if torch.cuda.is_available() else "cpu" # Sets up self.device to cuda if torch exists with Nvidia/CUDA, otherwise sets to cpu
|
||||
self.cuda_is_available = torch.cuda.is_available() # Sets up cuda_is_available as a True/False to track if Nvidia/CUDA was found on the system
|
||||
self.tts_generating_lock = False # Used to lock and unlock the tts generation process at the start/end of tts generation.
|
||||
self.tts_stop_generation = False # Used in conjunction with tts_generating_lock to call for a stop to the current generation. If called (set True) it needs to be set back to False when generation has been stopped.
|
||||
self.tts_narrator_generatingtts = False # Used to track if the current tts processes is narrator based. This can be used in conjunction with lowvram and device to avoid moving model between GPU(CUDA)<>RAM(CPU) each chunk of narrated text generated.
|
||||
self.model = None # If loading a model into CUDA/VRAM/RAM "model" is used as the variable name to load and interact with (see the XTTS model_engine script for examples.)
|
||||
self.is_tts_model_loaded = False # Used to track if a model is actually loaded in and error/fail things like TTS generation if its False
|
||||
self.current_model_loaded = None # Stores the name of the currenly loaded in model
|
||||
self.available_models = None # List of available models found by "def scan_models_folder"
|
||||
self.setup_has_run = False # Tracks if async def setup(self) has run, by setting to True, so that the /api/ready endpoint can provide a "Ready" status
|
||||
##############################################################################################
|
||||
# DONT CHANGE # Load in a list of the available TTS engines and the currently set TTS engine #
|
||||
##############################################################################################
|
||||
tts_engines_file = os.path.join(self.main_dir, "system", "tts_engines", "tts_engines.json")
|
||||
with open(tts_engines_file, "r") as f:
|
||||
tts_engines_data = json.load(f)
|
||||
self.engines_available = [engine["name"] for engine in tts_engines_data["engines_available"]] # A list of ALL the TTS engines available to be loaded by AllTalk
|
||||
self.engine_loaded = tts_engines_data["engine_loaded"] # In "tts_engines.json" what is the currently set TTS engine loading into AllTalk
|
||||
self.selected_model = tts_engines_data["selected_model"] # In "tts_engines.json" what is the currently set TTS model loading into AllTalk
|
||||
############################################################################
|
||||
# DONT CHANGE # Pull out all the settings for the currently set TTS engine #
|
||||
############################################################################
|
||||
with open(os.path.join(self.this_dir, "model_settings.json"), "r") as f:
|
||||
tts_model_loaded = json.load(f)
|
||||
# Access the model details
|
||||
self.manufacturer_name = tts_model_loaded["model_details"]["manufacturer_name"] # The company/person/body that generated the TTS engine/models etc
|
||||
self.manufacturer_website = tts_model_loaded["model_details"]["manufacturer_website"] # The website of the company/person/body where people can find more information
|
||||
# Access the features the model is capable of:
|
||||
self.audio_format = tts_model_loaded["model_capabilties"]["audio_format"] # This details the audio format your TTS engine is set to generate TTS in e.g. wav, mp3, flac, opus, acc, pcm. Please use only 1x format.
|
||||
self.deepspeed_capable = tts_model_loaded["model_capabilties"]["deepspeed_capable"] # Is your model capable of DeepSpeed
|
||||
self.deepspeed_available = 'deepspeed' in globals() # When we did the import earlier, at the top of this script, was DeepSpeed available for use
|
||||
self.generationspeed_capable = tts_model_loaded["model_capabilties"]["generationspeed_capable"] # Does this TTS engine support changing the speed of the generated TTS
|
||||
self.languages_capable = tts_model_loaded["model_capabilties"]["languages_capable"] # Are the actual models themselves capable of generating in multiple languages OR is each model language specific
|
||||
self.lowvram_capable = tts_model_loaded["model_capabilties"]["lowvram_capable"] # Is this engine capable of using low VRAM (moving the model between CPU And GPU memory)
|
||||
self.multimodel_capable = tts_model_loaded["model_capabilties"]["multimodel_capable"] # Is there just the one model or are there multiple models this engine supports.
|
||||
self.repetitionpenalty_capable = tts_model_loaded["model_capabilties"]["repetitionpenalty_capable"] # Is this TTS engine capable of changing the repititon penalty
|
||||
self.streaming_capable = tts_model_loaded["model_capabilties"]["streaming_capable"] # Is this TTS engine capabale of generating streaming audio
|
||||
self.temperature_capable = tts_model_loaded["model_capabilties"]["temperature_capable"] # Is this TTS engine capable of changing the temperature of the models
|
||||
self.multivoice_capable = tts_model_loaded["model_capabilties"]["multivoice_capable"] # Are the models multi-voice or single vocice models
|
||||
self.pitch_capable = tts_model_loaded["model_capabilties"]["pitch_capable"] # Is this TTS engine capable of changing the pitch of the genrated TTS
|
||||
# Access the current enginesettings
|
||||
self.def_character_voice = tts_model_loaded["settings"]["def_character_voice"] # What is the current default main/character voice that will be used if no voice specified.
|
||||
self.def_narrator_voice = tts_model_loaded["settings"]["def_narrator_voice"] # What is the current default narrator voice that will be used if no voice specified.
|
||||
self.deepspeed_enabled = tts_model_loaded["settings"]["deepspeed_enabled"] # If its available, is DeepSpeed enabled for the TTS engine
|
||||
self.engine_installed = tts_model_loaded["settings"]["engine_installed"] # Has the TTS engine been setup/installed (not curently used)
|
||||
self.generationspeed_set = tts_model_loaded["settings"]["generationspeed_set"] # What is the set/stored speed for generation.
|
||||
self.lowvram_enabled = tts_model_loaded["settings"]["lowvram_enabled"] # If its available, is LowVRAM enabled for the TTS engine
|
||||
# Check if someone has enabled lowvram on a system that's not CUDA enabled
|
||||
self.lowvram_enabled = False if not torch.cuda.is_available() else self.lowvram_enabled # If LowVRAM is mistakenly set and CUDA is not available, this will force it back off
|
||||
self.repetitionpenalty_set = tts_model_loaded["settings"]["repetitionpenalty_set"] # What is the currenly set repitition policy of the model (If it support repetition)
|
||||
self.temperature_set = tts_model_loaded["settings"]["temperature_set"] # What is the currenly set temperature of the model (If it support temp)
|
||||
self.pitch_set = tts_model_loaded["settings"]["pitch_set"] # What is the currenly set pitch of the model (If it support temp)
|
||||
# Gather the OpenAI API Voice Mappings
|
||||
self.openai_alloy = tts_model_loaded["openai_voices"]["alloy"] # The TTS engine voice that will be mapped to Open AI Alloy voice
|
||||
self.openai_echo = tts_model_loaded["openai_voices"]["echo"] # The TTS engine voice that will be mapped to Open AI Echo voice
|
||||
self.openai_fable = tts_model_loaded["openai_voices"]["fable"] # The TTS engine voice that will be mapped to Open AI Fable voice
|
||||
self.openai_nova = tts_model_loaded["openai_voices"]["nova"] # The TTS engine voice that will be mapped to Open AI Nova voice
|
||||
self.openai_onyx = tts_model_loaded["openai_voices"]["onyx"] # The TTS engine voice that will be mapped to Open AI Onyx voice
|
||||
self.openai_shimmer = tts_model_loaded["openai_voices"]["shimmer"] # The TTS engine voice that will be mapped to Open AI Shimmer voice
|
||||
###################################################################
|
||||
# DONT CHANGE # Load params and api_defaults from confignew.json #
|
||||
###################################################################
|
||||
# Define the path to the confignew.json file
|
||||
configfile_path = self.main_dir / "confignew.json"
|
||||
# Load config file and get settings
|
||||
with open(configfile_path, "r") as configfile:
|
||||
configfile_data = json.load(configfile)
|
||||
self.branding = configfile_data.get("branding", "") # Sets up self.branding for outputting the name stored in the "confgnew.json" file, as used in print statements.
|
||||
self.params = configfile_data # Loads in the curent "confgnew.json" file to self.params.
|
||||
self.debug_tts = configfile_data.get("debugging").get("debug_tts") # Can be used within this script as a True/False flag for generally debugging the TTS generation process.
|
||||
self.debug_tts_variables = configfile_data.get("debugging").get("debug_tts_variables") # Can be used within this script as a True/False flag for generally debugging variables (if you wish).
|
||||
################################################################
|
||||
# DONT CHANGE # Print out Python, CUDA, DeepSpeed versions ####
|
||||
################################################################
|
||||
def printout_versions(self):
|
||||
if deepspeed_available:
|
||||
print(f"[{self.branding}ENG] \033[92mDeepSpeed version :\033[93m",deepspeed.__version__,"\033[0m")
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[92mDeepSpeed version :\033[93m Not available\033[0m")
|
||||
print(f"[{self.branding}ENG] \033[92mPython Version :\033[93m {python_version}\033[0m")
|
||||
print(f"[{self.branding}ENG] \033[92mPyTorch Version :\033[93m {pytorch_version}\033[0m")
|
||||
if cuda_version is None:
|
||||
print(f"[{self.branding}ENG] \033[92mCUDA Version :\033[91m Not available\033[0m")
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[92mCUDA Version :\033[93m {cuda_version}\033[0m")
|
||||
print(f"[{self.branding}ENG]")
|
||||
return
|
||||
|
||||
###################################################################################
|
||||
###################################################################################
|
||||
# CHANGE ME # Inital setup of the model and engine. Called when the script starts #
|
||||
###################################################################################
|
||||
###################################################################################
|
||||
# In here you will add code to load in your model and do its inital setup. So you
|
||||
# may be calling your model loader via handle_tts_method_change or if your TTS
|
||||
# engine doesnt actually load a model into CUDA or System RAM, you may be doing
|
||||
# Something to fake its start-up.
|
||||
async def setup(self):
|
||||
self.printout_versions()
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
self.available_models = self.scan_models_folder()
|
||||
if self.selected_model:
|
||||
tts_model = f"{self.selected_model}"
|
||||
if tts_model in self.available_models:
|
||||
await self.handle_tts_method_change(tts_model)
|
||||
self.current_model_loaded = tts_model
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
else:
|
||||
self.current_model_loaded = "No Models Available"
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: Selected model '{self.selected_model}' not found in the models folder.")
|
||||
self.setup_has_run = True # Set to True, so that the /api/ready endpoint can provide a "Ready" status
|
||||
|
||||
##################################
|
||||
##################################
|
||||
# CHANGE ME # Low VRAM Swapping #
|
||||
##################################
|
||||
##################################
|
||||
# If your model does load into CUDA and you want to support LowVRAM, aka moving the model
|
||||
# on the fly between CUDA and System RAM on each generation request, you will be adding
|
||||
# The code in here to do it. See the XTTS tts engine for an example.
|
||||
async def handle_lowvram_change(self):
|
||||
if torch.cuda.is_available():
|
||||
if self.device == "cuda":
|
||||
self.device = "cpu"
|
||||
self.model.to(self.device)
|
||||
torch.cuda.empty_cache()
|
||||
else:
|
||||
self.device == "cpu"
|
||||
self.device = "cuda"
|
||||
self.model.to(self.device)
|
||||
|
||||
########################################
|
||||
########################################
|
||||
# CHANGE ME # DeepSpeed model loading #
|
||||
########################################
|
||||
########################################
|
||||
# If the model supports CUDA and DeepSpeed, this is where you can handle re-loading
|
||||
# the model as/when DeepSpeed is enabled/selected in the user interface. If your
|
||||
# TTS model doesnt support DeepSpeed, then it should be globally set in your
|
||||
# model_settings.JSON and this function will never get called, however it still needs
|
||||
# to exist as a function.
|
||||
async def handle_deepspeed_change(self, value):
|
||||
if value:
|
||||
# DeepSpeed enabled
|
||||
print(f"[{self.branding}ENG] \033[93mDeepSpeed Activating\033[0m")
|
||||
await self.unload_model()
|
||||
self.params["tts_method_api_local"] = False
|
||||
self.params["tts_method_xtts_local"] = True
|
||||
self.deepspeed_enabled = True
|
||||
await self.setup()
|
||||
else:
|
||||
# DeepSpeed disabled
|
||||
print(f"[{self.branding}ENG] \033[93mDeepSpeed De-Activating\033[0m")
|
||||
self.deepspeed_enabled = False
|
||||
await self.unload_model()
|
||||
await self.setup()
|
||||
return value # Return new checkbox value
|
||||
|
||||
##############################################################################################################################################
|
||||
##############################################################################################################################################
|
||||
# CHANGE ME # scan for available models/voices that are relevant to this TTS engine # XTTS is very unusal as it has 2x model loading methods #
|
||||
##############################################################################################################################################
|
||||
##############################################################################################################################################
|
||||
# This function looks and reports back the list of possible models your TTS engine can
|
||||
# load in. Some TTS engines have multiple models they can load and you will want to use
|
||||
# code for checking if the models are in the correct location/placement within the disk
|
||||
# the correct files exst per model etc (check XTTS for an example of this). Some models
|
||||
# like Piper, the actual models are the voices, so in the Piper scan_models_folder
|
||||
# function, we fake the only model being available as {'piper': 'piper'} aka, model name
|
||||
# then engine name, then we use the voices_file_list to populate the models as available
|
||||
# voices that can be selected in the interface.
|
||||
# If no models are found, we return "No Models Available" and continue on with the script.
|
||||
def scan_models_folder(self):
|
||||
models_folder = self.main_dir / "models" / "parler" # Edit to match the name of your folder where voices are stored
|
||||
print("models_folder is:", models_folder) if self.debug_tts else None
|
||||
self.available_models = {}
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# Check if model_path exists
|
||||
common_required_files = [
|
||||
"config.json", "generation_config.json", "preprocessor_config.json",
|
||||
"special_tokens_map.json", "spiece.model", "tokenizer.json", "tokenizer_config.json"
|
||||
]
|
||||
|
||||
for subfolder in models_folder.iterdir():
|
||||
if subfolder.is_dir():
|
||||
model_name = subfolder.name
|
||||
print("model_name is:", model_name) if self.debug_tts else None
|
||||
|
||||
# Check for common required files
|
||||
if all(subfolder.joinpath(file).exists() for file in common_required_files):
|
||||
# Check for mini model (single safetensors file)
|
||||
if subfolder.joinpath("model.safetensors").exists():
|
||||
self.available_models[f"parler - {model_name}"] = "parler"
|
||||
print("self.available_models is:", self.available_models) if self.debug_tts else None
|
||||
# Check for large model (split safetensors files and index)
|
||||
elif (subfolder.joinpath("model.safetensors.index.json").exists() and
|
||||
subfolder.joinpath("model-00001-of-00002.safetensors").exists() and
|
||||
subfolder.joinpath("model-00002-of-00002.safetensors").exists()):
|
||||
self.available_models[f"parler - {model_name}"] = "parler"
|
||||
print("self.available_models is:", self.available_models) if self.debug_tts else None
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: Model folder '{model_name}' is missing required model files.")
|
||||
else:
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: Model folder '{model_name}' is missing common required files.")
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
else:
|
||||
self.available_models = {'No Models Available': 'parler'} # Change the name in here to your TTS engine name
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: Model folder '{model_name}' is missing required")
|
||||
print(f"[{self.branding}ENG] \033[91mWarning\033[0m: files or the folder does not exist.")
|
||||
return self.available_models
|
||||
|
||||
#############################################################
|
||||
#############################################################
|
||||
# CHANGE ME # POPULATE FILES LIST FROM VOICES DIRECTORY ####
|
||||
#############################################################
|
||||
#############################################################
|
||||
# This function looks and reports back the list of possible voice your TTS engine can
|
||||
# load in. Some TTS engines the voices are wav file samples (XTTS), some are models
|
||||
# (Piper) and some are text (Parler) thats stored in a JSON file. We just need to
|
||||
# populate the "voices" variable somehow and if no voices are found, we return
|
||||
# "No Voices Found" back to the interface/api.
|
||||
def voices_file_list(self):
|
||||
try:
|
||||
voices = []
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
voices_file = os.path.join(self.main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices = json.load(f)
|
||||
return [voice["voice_name"] for voice in voices["voices"]]
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
if not voices:
|
||||
return ["No Voices Found"] # Return a list with {'No Voices Found'} if there are no voices/voice models.
|
||||
return voices # Return a list of models in the format {'engine name': 'model 1', 'engine name': 'model 2", etc}
|
||||
except Exception as e:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: Voices/Voice Models not found. Cannot load a list of voices.")
|
||||
print(f"[{self.branding}ENG]")
|
||||
return ["No Voices Found"]
|
||||
|
||||
#############################
|
||||
#############################
|
||||
# CHANGE ME # Model loading #
|
||||
#############################
|
||||
#############################
|
||||
# This function will handle the loading of your model, into VRAM/CUDA, System RAM or whatever.
|
||||
# In XTTS, which has 2x model loader types, there are 2x loaders. They are called by "def handle_tts_method_change"
|
||||
# In Parler we fake a model loader as Parler doesnt actually load a model into CUDA/System RAM as such. So, in that
|
||||
# situation, api_manual_load_model is kind of a blank function. Though we do set self.is_tts_model_loaded = True
|
||||
# as this is used elsewhere in the scripts to confirm that a model is available to be used for TTS generation.
|
||||
# We always check for "No Models Available" being sent as that means we are trying to load in a model that
|
||||
# doesnt exist/wasnt found on script start-up e.g. someone deleted the model from the folder or something.
|
||||
async def api_manual_load_model(self, model_name):
|
||||
if "No Models Available" in self.available_models:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: No models for this TTS engine were found to load. Please download a model.")
|
||||
return
|
||||
model_path = self.main_dir / "models" / "parler" / model_name
|
||||
print("model_path is:", model_path) if self.debug_tts_variables else None
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
self.model = ParlerTTSForConditionalGeneration.from_pretrained(model_path).to(self.device, dtype=torch.float32)
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(model_path)
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
self.is_tts_model_loaded = True
|
||||
return self.model
|
||||
|
||||
###############################
|
||||
###############################
|
||||
# CHANGE ME # Model unloading #
|
||||
###############################
|
||||
###############################
|
||||
# This function will handle the UN-loading of your model, from VRAM/CUDA, System RAM or whatever.
|
||||
# In XTTS, that model loads into CUDA/System Ram, so when we swap models, we want to unload the current model
|
||||
# free up the memory and then load in the new model to VRAM/CUDA. On the flip side of that, Parler doesnt
|
||||
# doesnt load into memory, so we just need to put a fake function here that doesnt really do anything
|
||||
# other than set "self.is_tts_model_loaded = False", which would be set back to true by the model loader.
|
||||
# So look at the Parler model_engine.py if you DONT need to unload models.
|
||||
async def unload_model(self):
|
||||
self.is_tts_model_loaded = False
|
||||
if not self.current_model_loaded == None:
|
||||
print(f"[{self.branding}ENG] \033[94mUnloading model \033[0m") if self.debug_tts else None
|
||||
if hasattr(self, 'model'):
|
||||
del self.model
|
||||
if torch.cuda.is_available():
|
||||
torch.cuda.empty_cache()
|
||||
return None
|
||||
|
||||
###################################################################################################################################
|
||||
###################################################################################################################################
|
||||
# CHANGE ME # Model changing. Unload out old model and load in a new one # XTTS is very unusal as it has 2x model loading methods #
|
||||
###################################################################################################################################
|
||||
###################################################################################################################################
|
||||
# This function is your central model loading/unloading handler that deals with the above functions as necesary, to call loading, unloading,
|
||||
# swappng DeepSpeed, Low vram etc. This function gets called with a "engine name - model name" type call. In XTTS, because there are 2x
|
||||
# model loader types, (XTTS and APILocal), we take tts_method and split the "engine name - model name" into a loader type and the model
|
||||
# that it needs to load in and then we call the correct loader function. Whereas in Parler, which doesnt load models into memory at all,
|
||||
# we just have a fake function that doesnt really do anything. We always check to see if the model name has "No Models Available" in the
|
||||
# name thats sent over, just to catch any potential errors. We display the start load time and end load time. Thats about it.
|
||||
async def handle_tts_method_change(self, tts_method):
|
||||
generate_start_time = time.time() # Record the start time of loading the model
|
||||
if "No Models Available" in self.available_models:
|
||||
print(f"[{self.branding}ENG] \033[91mError\033[0m: No models for this TTS engine were found to load. Please download a model.")
|
||||
return False
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
await self.unload_model()
|
||||
if tts_method.startswith("parler"):
|
||||
model_name = tts_method.split(" - ")[1]
|
||||
print(f"[{self.branding}ENG]\033[94m Model/Engine :\033[93m {model_name}\033[94m loading into\033[93m", self.device,"\033[0m")
|
||||
self.model = await self.api_manual_load_model(model_name)
|
||||
self.current_model_loaded = f"parler - {model_name}"
|
||||
else:
|
||||
self.current_model_loaded = None
|
||||
return False
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
generate_end_time = time.time() # Create an end timer for calculating load times
|
||||
generate_elapsed_time = generate_end_time - generate_start_time # Calculate start time minus end time
|
||||
print(f"[{self.branding}ENG] \033[94mLoad time :\033[93m {generate_elapsed_time:.2f} seconds.\033[0m") # Print out the result of the load time
|
||||
return True
|
||||
|
||||
##########################################################################################################################################
|
||||
##########################################################################################################################################
|
||||
# CHANGE ME # Model changing. Unload out old model and load in a new one # XTTS is very unusal as it has 2x model TTS generation methods #
|
||||
##########################################################################################################################################
|
||||
##########################################################################################################################################
|
||||
# In here all the possible options are sent over (text, voice to use, lanugage, speed etc etc) and its up to you how you use them, or not.
|
||||
# obviously if your TTS engine doesnt support speed for example, generationspeed_capable should be set False in your model_settings.JSON file
|
||||
# and a fake "generationspeed_set" value should be set. This allows a fake value to be sent over from the main script, even though it
|
||||
# wouldnt actually ever be used in the generation below. Nonethless all these values, real or just whats inside the configuration file
|
||||
# will be sent over for use.
|
||||
# Setting the xxxxxxx_capabale in the model_settings.JSON file, will enable/disable them being selectable by the user. For example, if you
|
||||
# set "generationspeed_capable" as false in the model_settings.JSON file, a user will not be able to select OR set the setting for
|
||||
# generation speed.
|
||||
# One thing to note is that we HAVE to have something in this generation request that is synchronous from the way its called, which means
|
||||
# we have to have an option for Streaming, even if our TTS engine doesnt support streaming. So in that case, we would set streaming_capable
|
||||
# as false in our model_settings.JSON file, meaning streaming will never be called. However, we have to put a fake streaming routine in our
|
||||
# function below (or a real function if it does support streaming of course). Parler has an example of a fake streaming function, which is
|
||||
# very clearly highlighted in its model_engine.py script.
|
||||
async def generate_tts(self, text, voice, language, temperature, repetition_penalty, speed, pitch, output_file, streaming):
|
||||
print(f"[{self.branding}Debug] Entered model_engine.py generate_tts function") if self.debug_tts else None
|
||||
if not self.is_tts_model_loaded: # Check if a model is loaded and error out if not.
|
||||
error_message = f"[{self.branding}ENG] \033[91mError\033[0m: You currently have no TTS model loaded."
|
||||
print(error_message)
|
||||
raise HTTPException(status_code=400, detail="You currently have no TTS model loaded.") # Raise an exception with a meaningful HTTP status code
|
||||
self.tts_generating_lock = True # Set the tts_generating lock to True, which stops other generation requests being sent into the pipeline
|
||||
print(f"[{self.branding}Debug] Checking low VRAM") if self.debug_tts else None
|
||||
if self.lowvram_enabled and self.device == "cpu": # If necessary, move the model out of System Ram to VRAM
|
||||
print(f"[{self.branding}Debug] Switching device") if self.debug_tts else None
|
||||
await self.handle_lowvram_change()
|
||||
print(f"[{self.branding}Debug] Setting a generate time") if self.debug_tts else None
|
||||
generate_start_time = time.time() # Record the start time of generating TTS
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
# ↑↑↑ Keep everything above this line ↑↑↑
|
||||
# ↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑↑
|
||||
|
||||
voices_file = os.path.join(self.main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices_data = json.load(f)
|
||||
for voice_data in voices_data["voices"]:
|
||||
if voice_data["voice_name"] == voice:
|
||||
description = voice_data["description"]
|
||||
break
|
||||
else:
|
||||
description = f"A {voice} voice speaking in {language}."
|
||||
else:
|
||||
description = f"A {voice} voice speaking in {language}."
|
||||
input_ids = self.tokenizer(description, return_tensors="pt").input_ids.to(self.device)
|
||||
prompt_input_ids = self.tokenizer(text, return_tensors="pt").input_ids.to(self.device)
|
||||
generation = self.model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids).to(torch.float32)
|
||||
audio_arr = generation.cpu().numpy().squeeze()
|
||||
sf.write(output_file, audio_arr, self.model.config.sampling_rate)
|
||||
|
||||
# Fake Streaming function here
|
||||
wavs = None # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
if streaming and wavs is not None: # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
# Streaming-specific operations # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
for wav_chunk in wavs: # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
yield wav_chunk.numpy().tobytes() # This is a fake function as this model doesnt stream but Python demands it otherwise it complains about await functions in the main script
|
||||
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
# ↓↓↓ Keep everything below this line ↓↓↓
|
||||
# ↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓
|
||||
generate_end_time = time.time() # Record the end time to generate TTS
|
||||
generate_elapsed_time = generate_end_time - generate_start_time
|
||||
print(f"[{self.branding}GEN] \033[94mTTS Generate: \033[93m{generate_elapsed_time:.2f} seconds. \033[94mLowVRAM: \033[33m{self.lowvram_enabled} \033[94mDeepSpeed: \033[33m{self.deepspeed_enabled}\033[0m")
|
||||
if self.lowvram_enabled and self.device == "cuda" and self.tts_narrator_generatingtts == False:
|
||||
await self.handle_lowvram_change()
|
||||
self.tts_generating_lock = False # Unlock the TTS generation queue to allow TTS generation requests to come in again.
|
||||
print(f"[{self.branding}Debug] generate_tts function completed") if self.debug_tts else None
|
||||
return
|
||||
|
||||
|
||||
|
||||
|
||||
@@ -1,42 +1,42 @@
|
||||
{
|
||||
"model_details": {
|
||||
"manufacturer_name": "Parler-TTS",
|
||||
"manufacturer_website": "https://github.com/huggingface/parler-tts",
|
||||
"model_description": "Parler-TTS is a lightweight text-to-speech (TTS) model that can generate high-quality, natural sounding speech in the style of a given speaker (gender, pitch, speaking style, etc). It is a reproduction of work from the paper Natural language guidance of high-fidelity text-to-speech with synthetic annotations by Dan Lyth and Simon King, from Stability AI and Edinburgh University respectively. Contrarily to other TTS models, Parler-TTS is a fully open-source release. All of the datasets, pre-processing, training code and weights are released publicly under permissive license, enabling the community to build on our work and develop their own powerful TTS models. \n\nParler's unique party trick is that like Stable Diffusion, you can describe the voice you want in a text entry; however, this means you are unlikely to get the same voice reproduced each time. Please see the Voice Editor tab."
|
||||
},
|
||||
"model_capabilties": {
|
||||
"audio_format": "wav",
|
||||
"deepspeed_capable": false,
|
||||
"generationspeed_capable": true,
|
||||
"languages_capable": false,
|
||||
"lowvram_capable": true,
|
||||
"multimodel_capable": true,
|
||||
"repetitionpenalty_capable": false,
|
||||
"streaming_capable": false,
|
||||
"temperature_capable": false,
|
||||
"multivoice_capable": true,
|
||||
"pitch_capable": false,
|
||||
"windows_capable": true,
|
||||
"linux_capable": true,
|
||||
"mac_capable": true
|
||||
},
|
||||
"settings": {
|
||||
"def_character_voice": "enthusiastic_female",
|
||||
"def_narrator_voice": "enthusiastic_female",
|
||||
"deepspeed_enabled": false,
|
||||
"engine_installed": true,
|
||||
"generationspeed_set": 1,
|
||||
"lowvram_enabled": true,
|
||||
"pitch_set": 0,
|
||||
"repetitionpenalty_set": 10,
|
||||
"temperature_set": 0.75
|
||||
},
|
||||
"openai_voices": {
|
||||
"alloy": "enthusiastic_female",
|
||||
"echo": "enthusiastic_female",
|
||||
"fable": "enthusiastic_female",
|
||||
"nova": "enthusiastic_female",
|
||||
"onyx": "enthusiastic_female",
|
||||
"shimmer": "enthusiastic_female"
|
||||
}
|
||||
{
|
||||
"model_details": {
|
||||
"manufacturer_name": "Parler-TTS",
|
||||
"manufacturer_website": "https://github.com/huggingface/parler-tts",
|
||||
"model_description": "Parler-TTS is a lightweight text-to-speech (TTS) model that can generate high-quality, natural sounding speech in the style of a given speaker (gender, pitch, speaking style, etc). It is a reproduction of work from the paper Natural language guidance of high-fidelity text-to-speech with synthetic annotations by Dan Lyth and Simon King, from Stability AI and Edinburgh University respectively. Contrarily to other TTS models, Parler-TTS is a fully open-source release. All of the datasets, pre-processing, training code and weights are released publicly under permissive license, enabling the community to build on our work and develop their own powerful TTS models. \n\nParler's unique party trick is that like Stable Diffusion, you can describe the voice you want in a text entry; however, this means you are unlikely to get the same voice reproduced each time. Please see the Voice Editor tab."
|
||||
},
|
||||
"model_capabilties": {
|
||||
"audio_format": "wav",
|
||||
"deepspeed_capable": false,
|
||||
"generationspeed_capable": false,
|
||||
"languages_capable": false,
|
||||
"lowvram_capable": true,
|
||||
"multimodel_capable": true,
|
||||
"repetitionpenalty_capable": false,
|
||||
"streaming_capable": false,
|
||||
"temperature_capable": false,
|
||||
"multivoice_capable": true,
|
||||
"pitch_capable": false,
|
||||
"windows_capable": true,
|
||||
"linux_capable": true,
|
||||
"mac_capable": true
|
||||
},
|
||||
"settings": {
|
||||
"def_character_voice": "enthusiastic_female",
|
||||
"def_narrator_voice": "enthusiastic_female",
|
||||
"deepspeed_enabled": false,
|
||||
"engine_installed": true,
|
||||
"generationspeed_set": 1,
|
||||
"lowvram_enabled": false,
|
||||
"pitch_set": 0,
|
||||
"repetitionpenalty_set": 10,
|
||||
"temperature_set": 0.75
|
||||
},
|
||||
"openai_voices": {
|
||||
"alloy": "enthusiastic_female",
|
||||
"echo": "enthusiastic_female",
|
||||
"fable": "enthusiastic_female",
|
||||
"nova": "enthusiastic_female",
|
||||
"onyx": "enthusiastic_female",
|
||||
"shimmer": "enthusiastic_female"
|
||||
}
|
||||
}
|
||||
@@ -1,402 +1,432 @@
|
||||
import os
|
||||
import json
|
||||
import requests
|
||||
import gradio as gr
|
||||
from tqdm import tqdm
|
||||
from pathlib import Path
|
||||
this_dir = Path(__file__).parent.resolve() # Sets up self.this_dir as a variable for the folder THIS script is running in.
|
||||
main_dir = Path(__file__).parent.parent.parent.parent.resolve() # Sets up self.main_dir as a variable for the folder AllTalk is running in
|
||||
|
||||
##########################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Populate the voices list, using the method specific to your TTS engine #
|
||||
##########################################################################
|
||||
# This function is responsible for populating the list of available voices for your TTS engine.
|
||||
# You need to modify this function to use the appropriate method for your engine to retrieve the voice list.
|
||||
#
|
||||
# The current implementation lists all the WAV files in a "voices" directory, which may not be suitable for your engine.
|
||||
# You should replace the `xxxx_voices_file_list` function name to match your engine name. For example, if your engine
|
||||
# is named "mytts", the function should be named `mytts_voices_file_list`.
|
||||
#
|
||||
# You will also neef to update the code with your own implementation that retrieves the voice list according to your
|
||||
# engine's specific requirements. Typically this is the same code as will be in your model_engine.py file.
|
||||
#
|
||||
# For example, if your engine has a dedicated API or configuration file for managing voices, you should modify this
|
||||
# function to interact with that API or read from that configuration file.
|
||||
#
|
||||
# After making the necessary changes, this function should return a list of available voices that can be used
|
||||
# in your TTS engine's settings page.
|
||||
def parler_voices_file_list():
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices_data = json.load(f)
|
||||
voices_list = [voice["voice_name"] for voice in voices_data["voices"]]
|
||||
return sorted(voices_list)
|
||||
else:
|
||||
return []
|
||||
|
||||
|
||||
######################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Imports and saves the TTS engine-specific settings #
|
||||
######################################################
|
||||
# This function is responsible for importing and saving the settings specific to your TTS engine.
|
||||
# You need to make the following change:
|
||||
#
|
||||
# 1. Change the name of the function `xxxx_model_update_settings` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_model_update_settings`.
|
||||
#
|
||||
# After making this change, the function will load the model settings from a JSON file, update the settings and voice
|
||||
# dictionaries with the values provided as arguments, and save the updated settings back to the JSON file.
|
||||
#
|
||||
# You do not need to modify the function's logic or any other part of the code.
|
||||
def parler_model_update_settings(def_character_voice_gr, def_narrator_voice_gr, lowvram_enabled_gr, deepspeed_enabled_gr, temperature_set_gr, repetitionpenalty_set_gr, pitch_set_gr, generationspeed_set_gr, alloy_gr, echo_gr, fable_gr, nova_gr, onyx_gr, shimmer_gr):
|
||||
# Load the model_config_data from the JSON file
|
||||
with open(os.path.join(this_dir, "model_settings.json"), "r") as f:
|
||||
model_config_data = json.load(f)
|
||||
# Update the settings and openai_voices dictionaries with the new values
|
||||
model_config_data["settings"]["def_character_voice"] = def_character_voice_gr
|
||||
model_config_data["settings"]["def_narrator_voice"] = def_narrator_voice_gr
|
||||
model_config_data["openai_voices"]["alloy"] = alloy_gr
|
||||
model_config_data["openai_voices"]["echo"] = echo_gr
|
||||
model_config_data["openai_voices"]["fable"] = fable_gr
|
||||
model_config_data["openai_voices"]["nova"] = nova_gr
|
||||
model_config_data["openai_voices"]["onyx"] = onyx_gr
|
||||
model_config_data["openai_voices"]["shimmer"] = shimmer_gr
|
||||
model_config_data["settings"]["lowvram_enabled"] = lowvram_enabled_gr == "Enabled"
|
||||
model_config_data["settings"]["deepspeed_enabled"] = deepspeed_enabled_gr == "Enabled"
|
||||
model_config_data["settings"]["temperature_set"] = temperature_set_gr
|
||||
model_config_data["settings"]["repetitionpenalty_set"] = repetitionpenalty_set_gr
|
||||
model_config_data["settings"]["pitch_set"] = pitch_set_gr
|
||||
model_config_data["settings"]["generationspeed_set"] = generationspeed_set_gr
|
||||
# Save the updated model_config_data to the JSON file
|
||||
with open(os.path.join(this_dir, "model_settings.json"), "w") as f:
|
||||
json.dump(model_config_data, f, indent=4)
|
||||
return "Settings updated successfully!"
|
||||
|
||||
#######################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Sets up the engine-specific settings page in Gradio #
|
||||
#######################################################
|
||||
# This function sets up the Gradio interface for the settings page specific to your TTS engine.
|
||||
# You need to make the following changes:
|
||||
#
|
||||
# 1. Change the name of the function `xxxx_model_alltalk_settings` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_model_alltalk_settings`.
|
||||
#
|
||||
# 2. Change the name of the `submit_button.click` function call to match the name you gave to the function
|
||||
# that imports and saves your engine's settings (the function you modified above).
|
||||
#
|
||||
# 3. Change the name of the `voice_list` function call to match the name of the function that lists
|
||||
# the available voices for your TTS engine.
|
||||
#
|
||||
# 4. Change the 'title' of the `gr.Blocks` to match your engine's name e.g. title="mytts TTS"
|
||||
#
|
||||
# After making these changes, this function will create and return the Gradio interface for your TTS engine's
|
||||
# settings page, allowing users to configure various options and voice selections.
|
||||
def parler_model_alltalk_settings(model_config_data):
|
||||
features_list = model_config_data['model_capabilties']
|
||||
voice_list = parler_voices_file_list()
|
||||
with gr.Blocks(title="parler TTS") as app:
|
||||
with gr.Tab("Default Settings"):
|
||||
with gr.Row():
|
||||
lowvram_enabled_gr = gr.Radio(choices={"Enabled": "true", "Disabled": "false"}, label="Low VRAM" if model_config_data["model_capabilties"]["lowvram_capable"] else "Low VRAM N/A", value="Enabled" if model_config_data["settings"]["lowvram_enabled"] else "Disabled", interactive=model_config_data["model_capabilties"]["lowvram_capable"])
|
||||
deepspeed_enabled_gr = gr.Radio(choices={"Enabled": "true", "Disabled": "false"}, label="DeepSpeed Activate" if model_config_data["model_capabilties"]["deepspeed_capable"] else "DeepSpeed N/A", value="Enabled" if model_config_data["settings"]["deepspeed_enabled"] else "Disabled", interactive=model_config_data["model_capabilties"]["deepspeed_capable"])
|
||||
temperature_set_gr = gr.Slider(value=float(model_config_data["settings"]["temperature_set"]), minimum=0, maximum=1, step=0.05, label="Temperature" if model_config_data["model_capabilties"]["temperature_capable"] else "Temperature N/A", interactive=model_config_data["model_capabilties"]["temperature_capable"])
|
||||
repetitionpenalty_set_gr = gr.Slider(value=float(model_config_data["settings"]["repetitionpenalty_set"]), minimum=1, maximum=20, step=1, label="Repetition Penalty" if model_config_data["model_capabilties"]["repetitionpenalty_capable"] else "Repetition N/A", interactive=model_config_data["model_capabilties"]["repetitionpenalty_capable"])
|
||||
pitch_set_gr = gr.Slider(value=float(model_config_data["settings"]["pitch_set"]), minimum=-10, maximum=10, step=1, label="Pitch" if model_config_data["model_capabilties"]["pitch_capable"] else "Pitch N/A", interactive=model_config_data["model_capabilties"]["pitch_capable"])
|
||||
generationspeed_set_gr = gr.Slider(value=float(model_config_data["settings"]["generationspeed_set"]), minimum=0.25, maximum=2.00, step=0.25, label="Speed" if model_config_data["model_capabilties"]["generationspeed_capable"] else "Speed N/A", interactive=model_config_data["model_capabilties"]["generationspeed_capable"])
|
||||
with gr.Row():
|
||||
with gr.Column():
|
||||
gr.Markdown("### OpenAI Voice Mappings")
|
||||
with gr.Group():
|
||||
with gr.Row():
|
||||
alloy_gr = gr.Dropdown(value=model_config_data["openai_voices"]["alloy"], label="Alloy", choices=voice_list, allow_custom_value=True)
|
||||
echo_gr = gr.Dropdown(value=model_config_data["openai_voices"]["echo"], label="Echo", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Row():
|
||||
fable_gr = gr.Dropdown(value=model_config_data["openai_voices"]["fable"], label="Fable", choices=voice_list, allow_custom_value=True)
|
||||
nova_gr = gr.Dropdown(value=model_config_data["openai_voices"]["nova"], label="Nova", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Row():
|
||||
onyx_gr = gr.Dropdown(value=model_config_data["openai_voices"]["onyx"], label="Onyx", choices=voice_list, allow_custom_value=True)
|
||||
shimmer_gr = gr.Dropdown(value=model_config_data["openai_voices"]["shimmer"], label="Shimmer", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Column():
|
||||
gr.Markdown("### Default Voices")
|
||||
with gr.Row():
|
||||
def_character_voice_gr = gr.Dropdown(value=model_config_data["settings"]["def_character_voice"], label="Default/Character Voice", choices=voice_list, allow_custom_value=True)
|
||||
def_narrator_voice_gr = gr.Dropdown(value=model_config_data["settings"]["def_narrator_voice"], label="Narrator Voice", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Group():
|
||||
with gr.Row():
|
||||
details_text = gr.Textbox(label="Details", show_label=False, lines=5, interactive=False, value="In this section, you can set the default settings for this TTS engine. Settings that are not supported by the current engine will be greyed out and cannot be selected. Default voices specified here will be used when no specific voice is provided in the TTS generation request. If a voice is specified in the request, it will override these default settings. When using the OpenAI API compatable API with this TTS engine, the voice mappings will be applied. As the OpenAI API has a limited set of 6 voices, these mappings ensure compatibility by mapping the OpenAI voices to the available voices in this TTS engine.")
|
||||
with gr.Row():
|
||||
submit_button = gr.Button("Update Settings")
|
||||
output_message = gr.Textbox(label="Output Message", interactive=False, show_label=False)
|
||||
submit_button.click(parler_model_update_settings, inputs=[def_character_voice_gr, def_narrator_voice_gr, lowvram_enabled_gr, deepspeed_enabled_gr, temperature_set_gr, repetitionpenalty_set_gr, pitch_set_gr, generationspeed_set_gr, alloy_gr, echo_gr, fable_gr, nova_gr, onyx_gr, shimmer_gr], outputs=output_message)
|
||||
|
||||
###########################################################################################
|
||||
# Do not change this section apart from "TTS Engine Name" value to match your engine name #
|
||||
###########################################################################################
|
||||
with gr.Tab("Engine Information"):
|
||||
with gr.Row():
|
||||
with gr.Group():
|
||||
gr.Textbox(label="Manufacturer Name", value=model_config_data['model_details']['manufacturer_name'], interactive=False)
|
||||
gr.Textbox(label="Manufacturer Website/TTS Engine Support", value=model_config_data['model_details']['manufacturer_website'], interactive=False)
|
||||
gr.Textbox(label="Engine/Model Description", value=model_config_data['model_details']['model_description'], interactive=False, lines=13)
|
||||
with gr.Column():
|
||||
with gr.Row():
|
||||
gr.Textbox(label="DeepSpeed Capable", value='Yes' if features_list['deepspeed_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Pitch Capable", value='Yes' if features_list['pitch_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Generation Speed Capable", value='Yes' if features_list['generationspeed_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Repetition Penalty Capable", value='Yes' if features_list['repetitionpenalty_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Multi Languages Capable", value='Yes' if features_list['languages_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Streaming Capable", value='Yes' if features_list['streaming_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Low VRAM Capable", value='Yes' if features_list['lowvram_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Temperature Capable", value='Yes' if features_list['temperature_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Multi Model Capable Engine", value='Yes' if features_list['multimodel_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Multi Voice Capable Models", value='Yes' if features_list['multivoice_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Default Audio output format", value=model_config_data['model_capabilties']['audio_format'], interactive=False)
|
||||
gr.Textbox(label="TTS Engine Name", value="parler", interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Windows Support", value='Yes' if features_list['windows_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Linux Support", value='Yes' if features_list['linux_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Mac Support", value='Yes' if features_list['mac_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Markdown("""
|
||||
#### 🟧 DeepSpeed Capable
|
||||
DeepSpeed is a deep learning optimization library that can significantly speed up model training and inference. If a model is DeepSpeed capable, it means it can utilize DeepSpeed to accelerate the generation of text-to-speech output. This requires the model to be loaded into CUDA/VRAM on an Nvidia GPU and the model's inference method to support DeepSpeed.
|
||||
#### 🟧 Pitch Capable
|
||||
Pitch refers to the highness or lowness of a sound. If a model is pitch capable, it can adjust the pitch of the generated speech, allowing for more expressive and varied output.
|
||||
#### 🟧 Generation Speed Capable
|
||||
Generation speed refers to the rate at which the model can generate text-to-speech output. If a model is generation speed capable, it means the speed of the generated speech can be adjusted, making it faster or slower depending on the desired output.
|
||||
#### 🟧 Repetition Penalty Capable
|
||||
Repetition penalty is a technique used to discourage the model from repeating the same words or phrases multiple times in the same sounding way. If a model is repetition penalty capable, it can apply this penalty during generation to improve the diversity and naturalness of the output.
|
||||
#### 🟧 Multi-Languages Capable
|
||||
Multi-language capable models can generate speech in multiple languages. This means that the model has been trained on data from different languages and can switch between them during generation. Some models are language-specific.
|
||||
#### 🟧 Multi-Voice Capable
|
||||
Multi-voice capable models generate speech in multiple voices or speaking styles. This means that the model has been trained on data from different speakers and can mimic their voices during generation, or is a voice cloning model that can generate speech based on the input sample.
|
||||
""")
|
||||
gr.Markdown("""
|
||||
#### 🟧 Streaming Capable
|
||||
Streaming refers to the ability to generate speech output in real-time, without the need to generate the entire output before playback. If a model is streaming capable, it can generate speech on-the-fly, allowing for faster response times and more interactive applications.
|
||||
#### 🟧 Low VRAM Capable
|
||||
VRAM (Video Random Access Memory) is a type of memory used by GPUs to store and process data. If a model is low VRAM capable, it can efficiently utilize the available VRAM by moving data between CPU and GPU memory as needed, allowing for generation even on systems with limited VRAM where it may be competing with an LLM model for VRAM.
|
||||
#### 🟧 Temperature Capable
|
||||
Temperature is a hyperparameter that controls the randomness of the generated output. If a model is temperature capable, the temperature can be adjusted to make the output more or less random, affecting the creativity and variability of the generated speech.
|
||||
#### 🟧 Multi-Model Capable Engine
|
||||
If an engine is multi-model capable, it means that it can support and utilize multiple models for text-to-speech generation. This allows for greater flexibility and the ability to switch between different models depending on the desired output. Different models may be capable of different languages, specific languages, voices, etc.
|
||||
#### 🟧 Default Audio Output Format
|
||||
Specifies the file format in which the generated speech will be saved. Common audio formats include WAV, MP3, FLAC, Opus, AAC, and PCM. If you want different outputs, you can set the transcode function to change the output audio, though transcoding will add a little time to the generation and is not available for streaming generation.
|
||||
#### 🟧 Windows/Linux/Mac Support
|
||||
These indicators show whether the model and engine are compatible with Windows, Linux, or macOS. However, additional setup or requirements may be necessary to ensure full functionality on your operating system. Please note that full platform support has not been extensively tested.
|
||||
""")
|
||||
|
||||
#######################################################################################################################################################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# You will need to build a custom method to identify if your models are installed and download them. Store your models list in the available_models.json file, alongside your model_settings.json etc #
|
||||
#######################################################################################################################################################################################################
|
||||
with gr.Tab("Models/Voices Download"):
|
||||
with gr.Row():
|
||||
# Load the available models from the JSON file
|
||||
with open(os.path.join(this_dir, "available_models.json"), "r") as f:
|
||||
available_models = json.load(f)
|
||||
# Extract the model names for the dropdown
|
||||
model_names = [model["model_name"] for model in available_models["models"]]
|
||||
# Create the dropdown
|
||||
model_dropdown = gr.Dropdown(choices=sorted(model_names), label="Select Model", value=model_names[0])
|
||||
download_button = gr.Button("Download Model/Missing Files")
|
||||
|
||||
def download_model(model_name, force_download=False):
|
||||
# Find the selected model in the available models
|
||||
selected_model = next(model for model in available_models["models"] if model["model_name"] == model_name)
|
||||
# Get the folder path and files to download
|
||||
folder_path = os.path.join(main_dir, "models", "parler", model_name)
|
||||
files_to_download = selected_model["files_to_download"]
|
||||
# Check if all files are already downloaded
|
||||
all_files_exists = all(os.path.exists(os.path.join(folder_path, os.path.basename(url.split('?')[0]))) for url in files_to_download)
|
||||
if all_files_exists and not force_download:
|
||||
return "All files are already downloaded. No need to download again."
|
||||
else:
|
||||
# Create the folder if it doesn't exist
|
||||
os.makedirs(folder_path, exist_ok=True)
|
||||
# Download the missing files
|
||||
for url in files_to_download:
|
||||
file_name = os.path.basename(url.split('?')[0])
|
||||
file_path = os.path.join(folder_path, file_name)
|
||||
if not os.path.exists(file_path) or force_download:
|
||||
print(f"Downloading {file_name}...")
|
||||
response = requests.get(url, stream=True)
|
||||
total_size_in_bytes = int(response.headers.get("content-length", 0))
|
||||
block_size = 1024 # 1 Kibibyte
|
||||
progress_bar = tqdm(total=total_size_in_bytes, unit="iB", unit_scale=True)
|
||||
with open(file_path, "wb") as file:
|
||||
for data in response.iter_content(block_size):
|
||||
progress_bar.update(len(data))
|
||||
file.write(data)
|
||||
progress_bar.close()
|
||||
return "Model downloaded successfully!"
|
||||
|
||||
with gr.Row():
|
||||
download_status = gr.Textbox(label="Download Status")
|
||||
|
||||
download_button.click(download_model, inputs=model_dropdown, outputs=download_status)
|
||||
|
||||
def show_confirm_cancel(model_name):
|
||||
all_files_exists = all(os.path.exists(os.path.join(main_dir, "models", "parler", os.path.basename(file.split('?')[0]))) for model in available_models["models"] if model["model_name"] == model_name for file in model["files_to_download"])
|
||||
if all_files_exists:
|
||||
return [gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)]
|
||||
else:
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)]
|
||||
|
||||
def confirm_download(model_name):
|
||||
download_status_text = download_model(model_name, force_download=True)
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), download_status_text]
|
||||
|
||||
def cancel_download():
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), ""]
|
||||
|
||||
with gr.Row():
|
||||
confirm_button = gr.Button("Download Anyway", visible=False)
|
||||
cancel_button = gr.Button("Cancel", visible=False)
|
||||
|
||||
download_button.click(show_confirm_cancel, inputs=model_dropdown, outputs=[confirm_button, download_button, cancel_button])
|
||||
confirm_button.click(confirm_download, inputs=model_dropdown, outputs=[confirm_button, download_button, cancel_button, download_status])
|
||||
cancel_button.click(cancel_download, inputs=None, outputs=[confirm_button, download_button, cancel_button, download_status])
|
||||
|
||||
#####################################
|
||||
# Voices Editor - Unique to Parler #
|
||||
#####################################
|
||||
# Load voices from the JSON file
|
||||
def load_voices():
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
return json.load(f)
|
||||
else:
|
||||
return {"voices": []}
|
||||
|
||||
# Save voices to the JSON file
|
||||
def save_voices(voices):
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
with open(voices_file, "w") as f:
|
||||
json.dump(voices, f, indent=2)
|
||||
|
||||
# Convert voices to a list for display
|
||||
def voices_to_list(voices):
|
||||
return [f"{i+1}. {voice['voice_name']} - {voice['description']}" for i, voice in enumerate(voices["voices"])]
|
||||
|
||||
# Populate text boxes with selected voice data
|
||||
def populate_voice_data(selected):
|
||||
if selected:
|
||||
index = int(selected.split(".")[0]) - 1
|
||||
voices = load_voices()
|
||||
selected_voice = voices["voices"][index]
|
||||
return selected_voice["voice_name"], selected_voice["description"], index, gr.update(interactive=False), "Save Voice"
|
||||
return "", "", -1, gr.update(interactive=True), "Add/Save Voice"
|
||||
|
||||
# Add or update voice
|
||||
def add_update_voice(voice_name, description, selected_index, button_label):
|
||||
voices = load_voices()
|
||||
if selected_index >= 0:
|
||||
voices["voices"][selected_index]["description"] = description
|
||||
else:
|
||||
voices["voices"].append({"voice_name": voice_name, "description": description})
|
||||
save_voices(voices)
|
||||
return (gr.Dropdown(choices=voices_to_list(voices), value=voices_to_list(voices)[0] if voices["voices"] else None),
|
||||
"", "", -1, gr.update(interactive=True), "Add/Save Voice")
|
||||
|
||||
# Delete selected voice
|
||||
def delete_selected_voice(selected_index):
|
||||
voices = load_voices()
|
||||
if selected_index >= 0:
|
||||
del voices["voices"][selected_index]
|
||||
save_voices(voices)
|
||||
return (gr.Dropdown(choices=voices_to_list(voices), value=voices_to_list(voices)[0] if voices["voices"] else None),
|
||||
"", "", -1, gr.update(interactive=True), "Add/Save Voice")
|
||||
|
||||
# Clear text boxes
|
||||
def clear_text_boxes():
|
||||
return "", "", -1, gr.update(interactive=True), "Add/Save Voice"
|
||||
|
||||
# Load initial voices
|
||||
voices = load_voices()
|
||||
|
||||
with gr.Tab("Voice Editor"):
|
||||
with gr.Row():
|
||||
voice_name = gr.Textbox(label="Voice Name", scale=1)
|
||||
description = gr.Textbox(label="Voice Description", lines=4, scale=3)
|
||||
with gr.Row():
|
||||
add_update_button = gr.Button("Add/Save Voice")
|
||||
delete_button = gr.Button("Delete Selected Voice")
|
||||
clear_button = gr.Button("Clear Text Boxes")
|
||||
with gr.Row():
|
||||
voice_list = gr.Dropdown(
|
||||
choices=voices_to_list(voices),
|
||||
label="Voices",
|
||||
interactive=True,
|
||||
value=voices_to_list(voices)[0] if voices["voices"] else None
|
||||
)
|
||||
# Hidden state for selected index and button label
|
||||
selected_index = gr.Number(value=-1, visible=False)
|
||||
button_label = gr.State(value="Add/Save Voice")
|
||||
|
||||
# Button click events
|
||||
voice_list.change(populate_voice_data, inputs=voice_list, outputs=[voice_name, description, selected_index, voice_name, button_label])
|
||||
add_update_button.click(add_update_voice, inputs=[voice_name, description, selected_index, button_label], outputs=[voice_list, voice_name, description, selected_index, voice_name, button_label])
|
||||
delete_button.click(delete_selected_voice, inputs=selected_index, outputs=[voice_list, voice_name, description, selected_index, voice_name, button_label])
|
||||
clear_button.click(clear_text_boxes, outputs=[voice_name, description, selected_index, voice_name, button_label])
|
||||
|
||||
###################################################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Add any engine specific help, bugs, issues, operating system specifc requirements/setup in here #
|
||||
# Please use Markdown format, so gr.Markdown() with your markdown inside it. #
|
||||
###################################################################################################
|
||||
with gr.Tab("Engine Help"):
|
||||
with gr.Row():
|
||||
gr.Markdown("""
|
||||
### 🟧 Where are the Parler models stored?
|
||||
This extension will download the models to `/alltalk_tts/models/parler/` folder.<br>
|
||||
|
||||
### 🟧 Where are the voices stored for parler models?
|
||||
Parler is unlike any other TTS engine. It is actually mode like Stable Diffusion, where you write a text based description of what you want your image to look like, but in this case, you describe what you want the voice to sound like. This means you can have pretty much any voice you want, however, the downside to this is, there is no consistency of the voices, meaning that no 2x TTS generations will sound exactly like the same voice.<br>
|
||||
|
||||
### 🟧 So how do I create my voices for Parler?
|
||||
On the settings page for Parler, there is a Voice Editor tab. In there you can add/remove/ammend voices as you wish, or edit the `parler_voices.json` file stored in `system/tts_engines/parler/`.
|
||||
|
||||
### 🟧 Where are the outputs stored & Automatic output wav file deletion
|
||||
Voice outputs are stored in `/alltalk_tts/outputs/`. You can configure automatic maintenance deletion of old wav files by setting `Del WAV's older than` in the global settings.<br>
|
||||
|
||||
> When `Disabled`, your output wav files will be left untouched.<br>
|
||||
> When set to a setting `1 Day` or greater, your output wav files older than that time period will be automatically deleted on start-up of AllTalk.<br>
|
||||
""")
|
||||
gr.Markdown("""
|
||||
""")
|
||||
|
||||
return app
|
||||
|
||||
################################
|
||||
# REQUIRED CHANGE #
|
||||
# Sets up the Gradio interface #
|
||||
################################
|
||||
# This function sets up the Gradio interface for your TTS engine's settings page.
|
||||
# You need to change the name of the function calls to match the names you set in the functions above.
|
||||
#
|
||||
# Specifically, you need to update the following:
|
||||
#
|
||||
# 1. The name of the function `xxxx_at_gradio_settings_page` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_at_gradio_settings_page`.
|
||||
#
|
||||
# 2. The name of the function call `xxxx_model_alltalk_settings(model_config_data)`.
|
||||
# This should match the name you gave to the function that sets up the engine-specific settings page in Gradio.
|
||||
# If you named that function `mytts_model_alltalk_settings`, then the call should be:
|
||||
# `mytts_model_alltalk_settings(model_config_data)`
|
||||
#
|
||||
# After making these changes, this function will create and return the Gradio app for your TTS engine's settings page.
|
||||
def parler_at_gradio_settings_page(model_config_data):
|
||||
app = parler_model_alltalk_settings(model_config_data)
|
||||
return app
|
||||
def parler_at_gradio_settings_page(model_config_data):
|
||||
app = parler_model_alltalk_settings(model_config_data)
|
||||
import os
|
||||
import json
|
||||
import requests
|
||||
import gradio as gr
|
||||
from tqdm import tqdm
|
||||
from pathlib import Path
|
||||
this_dir = Path(__file__).parent.resolve() # Sets up self.this_dir as a variable for the folder THIS script is running in.
|
||||
main_dir = Path(__file__).parent.parent.parent.parent.resolve() # Sets up self.main_dir as a variable for the folder AllTalk is running in
|
||||
|
||||
##########################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Populate the voices list, using the method specific to your TTS engine #
|
||||
##########################################################################
|
||||
# This function is responsible for populating the list of available voices for your TTS engine.
|
||||
# You need to modify this function to use the appropriate method for your engine to retrieve the voice list.
|
||||
#
|
||||
# The current implementation lists all the WAV files in a "voices" directory, which may not be suitable for your engine.
|
||||
# You should replace the `xxxx_voices_file_list` function name to match your engine name. For example, if your engine
|
||||
# is named "mytts", the function should be named `mytts_voices_file_list`.
|
||||
#
|
||||
# You will also neef to update the code with your own implementation that retrieves the voice list according to your
|
||||
# engine's specific requirements. Typically this is the same code as will be in your model_engine.py file.
|
||||
#
|
||||
# For example, if your engine has a dedicated API or configuration file for managing voices, you should modify this
|
||||
# function to interact with that API or read from that configuration file.
|
||||
#
|
||||
# After making the necessary changes, this function should return a list of available voices that can be used
|
||||
# in your TTS engine's settings page.
|
||||
def parler_voices_file_list():
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
voices_data = json.load(f)
|
||||
voices_list = [voice["voice_name"] for voice in voices_data["voices"]]
|
||||
return sorted(voices_list)
|
||||
else:
|
||||
return []
|
||||
|
||||
|
||||
######################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Imports and saves the TTS engine-specific settings #
|
||||
######################################################
|
||||
# This function is responsible for importing and saving the settings specific to your TTS engine.
|
||||
# You need to make the following change:
|
||||
#
|
||||
# 1. Change the name of the function `xxxx_model_update_settings` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_model_update_settings`.
|
||||
#
|
||||
# After making this change, the function will load the model settings from a JSON file, update the settings and voice
|
||||
# dictionaries with the values provided as arguments, and save the updated settings back to the JSON file.
|
||||
#
|
||||
# You do not need to modify the function's logic or any other part of the code.
|
||||
def parler_model_update_settings(def_character_voice_gr, def_narrator_voice_gr, lowvram_enabled_gr, deepspeed_enabled_gr, temperature_set_gr, repetitionpenalty_set_gr, pitch_set_gr, generationspeed_set_gr, alloy_gr, echo_gr, fable_gr, nova_gr, onyx_gr, shimmer_gr):
|
||||
# Load the model_config_data from the JSON file
|
||||
with open(os.path.join(this_dir, "model_settings.json"), "r") as f:
|
||||
model_config_data = json.load(f)
|
||||
# Update the settings and openai_voices dictionaries with the new values
|
||||
model_config_data["settings"]["def_character_voice"] = def_character_voice_gr
|
||||
model_config_data["settings"]["def_narrator_voice"] = def_narrator_voice_gr
|
||||
model_config_data["openai_voices"]["alloy"] = alloy_gr
|
||||
model_config_data["openai_voices"]["echo"] = echo_gr
|
||||
model_config_data["openai_voices"]["fable"] = fable_gr
|
||||
model_config_data["openai_voices"]["nova"] = nova_gr
|
||||
model_config_data["openai_voices"]["onyx"] = onyx_gr
|
||||
model_config_data["openai_voices"]["shimmer"] = shimmer_gr
|
||||
model_config_data["settings"]["lowvram_enabled"] = lowvram_enabled_gr == "Enabled"
|
||||
model_config_data["settings"]["deepspeed_enabled"] = deepspeed_enabled_gr == "Enabled"
|
||||
model_config_data["settings"]["temperature_set"] = temperature_set_gr
|
||||
model_config_data["settings"]["repetitionpenalty_set"] = repetitionpenalty_set_gr
|
||||
model_config_data["settings"]["pitch_set"] = pitch_set_gr
|
||||
model_config_data["settings"]["generationspeed_set"] = generationspeed_set_gr
|
||||
# Save the updated model_config_data to the JSON file
|
||||
with open(os.path.join(this_dir, "model_settings.json"), "w") as f:
|
||||
json.dump(model_config_data, f, indent=4)
|
||||
return "Settings updated successfully!"
|
||||
|
||||
#######################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Sets up the engine-specific settings page in Gradio #
|
||||
#######################################################
|
||||
# This function sets up the Gradio interface for the settings page specific to your TTS engine.
|
||||
# You need to make the following changes:
|
||||
#
|
||||
# 1. Change the name of the function `xxxx_model_alltalk_settings` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_model_alltalk_settings`.
|
||||
#
|
||||
# 2. Change the name of the `submit_button.click` function call to match the name you gave to the function
|
||||
# that imports and saves your engine's settings (the function you modified above).
|
||||
#
|
||||
# 3. Change the name of the `voice_list` function call to match the name of the function that lists
|
||||
# the available voices for your TTS engine.
|
||||
#
|
||||
# 4. Change the 'title' of the `gr.Blocks` to match your engine's name e.g. title="mytts TTS"
|
||||
#
|
||||
# After making these changes, this function will create and return the Gradio interface for your TTS engine's
|
||||
# settings page, allowing users to configure various options and voice selections.
|
||||
def parler_model_alltalk_settings(model_config_data):
|
||||
features_list = model_config_data['model_capabilties']
|
||||
voice_list = parler_voices_file_list()
|
||||
with gr.Blocks(title="parler TTS") as app:
|
||||
with gr.Tab("Default Settings"):
|
||||
with gr.Row():
|
||||
lowvram_enabled_gr = gr.Radio(choices={"Enabled": "true", "Disabled": "false"}, label="Low VRAM" if model_config_data["model_capabilties"]["lowvram_capable"] else "Low VRAM N/A", value="Enabled" if model_config_data["settings"]["lowvram_enabled"] else "Disabled", interactive=model_config_data["model_capabilties"]["lowvram_capable"])
|
||||
deepspeed_enabled_gr = gr.Radio(choices={"Enabled": "true", "Disabled": "false"}, label="DeepSpeed Activate" if model_config_data["model_capabilties"]["deepspeed_capable"] else "DeepSpeed N/A", value="Enabled" if model_config_data["settings"]["deepspeed_enabled"] else "Disabled", interactive=model_config_data["model_capabilties"]["deepspeed_capable"])
|
||||
temperature_set_gr = gr.Slider(value=float(model_config_data["settings"]["temperature_set"]), minimum=0, maximum=1, step=0.05, label="Temperature" if model_config_data["model_capabilties"]["temperature_capable"] else "Temperature N/A", interactive=model_config_data["model_capabilties"]["temperature_capable"])
|
||||
repetitionpenalty_set_gr = gr.Slider(value=float(model_config_data["settings"]["repetitionpenalty_set"]), minimum=1, maximum=20, step=1, label="Repetition Penalty" if model_config_data["model_capabilties"]["repetitionpenalty_capable"] else "Repetition N/A", interactive=model_config_data["model_capabilties"]["repetitionpenalty_capable"])
|
||||
pitch_set_gr = gr.Slider(value=float(model_config_data["settings"]["pitch_set"]), minimum=-10, maximum=10, step=1, label="Pitch" if model_config_data["model_capabilties"]["pitch_capable"] else "Pitch N/A", interactive=model_config_data["model_capabilties"]["pitch_capable"])
|
||||
generationspeed_set_gr = gr.Slider(value=float(model_config_data["settings"]["generationspeed_set"]), minimum=0.25, maximum=2.00, step=0.25, label="Speed" if model_config_data["model_capabilties"]["generationspeed_capable"] else "Speed N/A", interactive=model_config_data["model_capabilties"]["generationspeed_capable"])
|
||||
with gr.Row():
|
||||
with gr.Column():
|
||||
gr.Markdown("### OpenAI Voice Mappings")
|
||||
with gr.Group():
|
||||
with gr.Row():
|
||||
alloy_gr = gr.Dropdown(value=model_config_data["openai_voices"]["alloy"], label="Alloy", choices=voice_list, allow_custom_value=True)
|
||||
echo_gr = gr.Dropdown(value=model_config_data["openai_voices"]["echo"], label="Echo", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Row():
|
||||
fable_gr = gr.Dropdown(value=model_config_data["openai_voices"]["fable"], label="Fable", choices=voice_list, allow_custom_value=True)
|
||||
nova_gr = gr.Dropdown(value=model_config_data["openai_voices"]["nova"], label="Nova", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Row():
|
||||
onyx_gr = gr.Dropdown(value=model_config_data["openai_voices"]["onyx"], label="Onyx", choices=voice_list, allow_custom_value=True)
|
||||
shimmer_gr = gr.Dropdown(value=model_config_data["openai_voices"]["shimmer"], label="Shimmer", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Column():
|
||||
gr.Markdown("### Default Voices")
|
||||
with gr.Row():
|
||||
def_character_voice_gr = gr.Dropdown(value=model_config_data["settings"]["def_character_voice"], label="Default/Character Voice", choices=voice_list, allow_custom_value=True)
|
||||
def_narrator_voice_gr = gr.Dropdown(value=model_config_data["settings"]["def_narrator_voice"], label="Narrator Voice", choices=voice_list, allow_custom_value=True)
|
||||
with gr.Group():
|
||||
with gr.Row():
|
||||
details_text = gr.Textbox(label="Details", show_label=False, lines=5, interactive=False, value="In this section, you can set the default settings for this TTS engine. Settings that are not supported by the current engine will be greyed out and cannot be selected. Default voices specified here will be used when no specific voice is provided in the TTS generation request. If a voice is specified in the request, it will override these default settings. When using the OpenAI API compatable API with this TTS engine, the voice mappings will be applied. As the OpenAI API has a limited set of 6 voices, these mappings ensure compatibility by mapping the OpenAI voices to the available voices in this TTS engine.")
|
||||
with gr.Row():
|
||||
submit_button = gr.Button("Update Settings")
|
||||
output_message = gr.Textbox(label="Output Message", interactive=False, show_label=False)
|
||||
submit_button.click(parler_model_update_settings, inputs=[def_character_voice_gr, def_narrator_voice_gr, lowvram_enabled_gr, deepspeed_enabled_gr, temperature_set_gr, repetitionpenalty_set_gr, pitch_set_gr, generationspeed_set_gr, alloy_gr, echo_gr, fable_gr, nova_gr, onyx_gr, shimmer_gr], outputs=output_message)
|
||||
|
||||
###########################################################################################
|
||||
# Do not change this section apart from "TTS Engine Name" value to match your engine name #
|
||||
###########################################################################################
|
||||
with gr.Tab("Engine Information"):
|
||||
with gr.Row():
|
||||
with gr.Group():
|
||||
gr.Textbox(label="Manufacturer Name", value=model_config_data['model_details']['manufacturer_name'], interactive=False)
|
||||
gr.Textbox(label="Manufacturer Website/TTS Engine Support", value=model_config_data['model_details']['manufacturer_website'], interactive=False)
|
||||
gr.Textbox(label="Engine/Model Description", value=model_config_data['model_details']['model_description'], interactive=False, lines=13)
|
||||
with gr.Column():
|
||||
with gr.Row():
|
||||
gr.Textbox(label="DeepSpeed Capable", value='Yes' if features_list['deepspeed_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Pitch Capable", value='Yes' if features_list['pitch_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Generation Speed Capable", value='Yes' if features_list['generationspeed_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Repetition Penalty Capable", value='Yes' if features_list['repetitionpenalty_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Multi Languages Capable", value='Yes' if features_list['languages_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Streaming Capable", value='Yes' if features_list['streaming_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Low VRAM Capable", value='Yes' if features_list['lowvram_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Temperature Capable", value='Yes' if features_list['temperature_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Multi Model Capable Engine", value='Yes' if features_list['multimodel_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Multi Voice Capable Models", value='Yes' if features_list['multivoice_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Default Audio output format", value=model_config_data['model_capabilties']['audio_format'], interactive=False)
|
||||
gr.Textbox(label="TTS Engine Name", value="parler", interactive=False)
|
||||
with gr.Row():
|
||||
gr.Textbox(label="Windows Support", value='Yes' if features_list['windows_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Linux Support", value='Yes' if features_list['linux_capable'] else 'No', interactive=False)
|
||||
gr.Textbox(label="Mac Support", value='Yes' if features_list['mac_capable'] else 'No', interactive=False)
|
||||
with gr.Row():
|
||||
gr.Markdown("""
|
||||
#### 🟧 DeepSpeed Capable
|
||||
DeepSpeed is a deep learning optimization library that can significantly speed up model training and inference. If a model is DeepSpeed capable, it means it can utilize DeepSpeed to accelerate the generation of text-to-speech output. This requires the model to be loaded into CUDA/VRAM on an Nvidia GPU and the model's inference method to support DeepSpeed.
|
||||
#### 🟧 Pitch Capable
|
||||
Pitch refers to the highness or lowness of a sound. If a model is pitch capable, it can adjust the pitch of the generated speech, allowing for more expressive and varied output.
|
||||
#### 🟧 Generation Speed Capable
|
||||
Generation speed refers to the rate at which the model can generate text-to-speech output. If a model is generation speed capable, it means the speed of the generated speech can be adjusted, making it faster or slower depending on the desired output.
|
||||
#### 🟧 Repetition Penalty Capable
|
||||
Repetition penalty is a technique used to discourage the model from repeating the same words or phrases multiple times in the same sounding way. If a model is repetition penalty capable, it can apply this penalty during generation to improve the diversity and naturalness of the output.
|
||||
#### 🟧 Multi-Languages Capable
|
||||
Multi-language capable models can generate speech in multiple languages. This means that the model has been trained on data from different languages and can switch between them during generation. Some models are language-specific.
|
||||
#### 🟧 Multi-Voice Capable
|
||||
Multi-voice capable models generate speech in multiple voices or speaking styles. This means that the model has been trained on data from different speakers and can mimic their voices during generation, or is a voice cloning model that can generate speech based on the input sample.
|
||||
""")
|
||||
gr.Markdown("""
|
||||
#### 🟧 Streaming Capable
|
||||
Streaming refers to the ability to generate speech output in real-time, without the need to generate the entire output before playback. If a model is streaming capable, it can generate speech on-the-fly, allowing for faster response times and more interactive applications.
|
||||
#### 🟧 Low VRAM Capable
|
||||
VRAM (Video Random Access Memory) is a type of memory used by GPUs to store and process data. If a model is low VRAM capable, it can efficiently utilize the available VRAM by moving data between CPU and GPU memory as needed, allowing for generation even on systems with limited VRAM where it may be competing with an LLM model for VRAM.
|
||||
#### 🟧 Temperature Capable
|
||||
Temperature is a hyperparameter that controls the randomness of the generated output. If a model is temperature capable, the temperature can be adjusted to make the output more or less random, affecting the creativity and variability of the generated speech.
|
||||
#### 🟧 Multi-Model Capable Engine
|
||||
If an engine is multi-model capable, it means that it can support and utilize multiple models for text-to-speech generation. This allows for greater flexibility and the ability to switch between different models depending on the desired output. Different models may be capable of different languages, specific languages, voices, etc.
|
||||
#### 🟧 Default Audio Output Format
|
||||
Specifies the file format in which the generated speech will be saved. Common audio formats include WAV, MP3, FLAC, Opus, AAC, and PCM. If you want different outputs, you can set the transcode function to change the output audio, though transcoding will add a little time to the generation and is not available for streaming generation.
|
||||
#### 🟧 Windows/Linux/Mac Support
|
||||
These indicators show whether the model and engine are compatible with Windows, Linux, or macOS. However, additional setup or requirements may be necessary to ensure full functionality on your operating system. Please note that full platform support has not been extensively tested.
|
||||
""")
|
||||
|
||||
#######################################################################################################################################################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# You will need to build a custom method to identify if your models are installed and download them. Store your models list in the available_models.json file, alongside your model_settings.json etc #
|
||||
#######################################################################################################################################################################################################
|
||||
with gr.Tab("Models/Voices Download"):
|
||||
with gr.Row():
|
||||
# Load the available models from the JSON file
|
||||
with open(os.path.join(this_dir, "available_models.json"), "r") as f:
|
||||
available_models = json.load(f)
|
||||
# Extract the model names for the dropdown
|
||||
model_names = [model["model_name"] for model in available_models["models"]]
|
||||
# Create the dropdown
|
||||
model_dropdown = gr.Dropdown(choices=sorted(model_names), label="Select Model", value=model_names[0])
|
||||
download_button = gr.Button("Download Model/Missing Files")
|
||||
|
||||
def download_model(model_name, force_download=False):
|
||||
# Find the selected model in the available models
|
||||
selected_model = next(model for model in available_models["models"] if model["model_name"] == model_name)
|
||||
# Get the folder path and files to download
|
||||
folder_path = os.path.join(main_dir, "models", "parler", model_name)
|
||||
files_to_download = selected_model["files_to_download"]
|
||||
# Check if all files are already downloaded
|
||||
all_files_exists = all(os.path.exists(os.path.join(folder_path, os.path.basename(url.split('?')[0]))) for url in files_to_download)
|
||||
if all_files_exists and not force_download:
|
||||
return "All files are already downloaded. No need to download again."
|
||||
else:
|
||||
# Create the folder if it doesn't exist
|
||||
os.makedirs(folder_path, exist_ok=True)
|
||||
# Download the missing files
|
||||
for url in files_to_download:
|
||||
file_name = os.path.basename(url.split('?')[0])
|
||||
file_path = os.path.join(folder_path, file_name)
|
||||
if not os.path.exists(file_path) or force_download:
|
||||
print(f"Downloading {file_name}...")
|
||||
response = requests.get(url, stream=True)
|
||||
total_size_in_bytes = int(response.headers.get("content-length", 0))
|
||||
block_size = 1024 # 1 Kibibyte
|
||||
progress_bar = tqdm(total=total_size_in_bytes, unit="iB", unit_scale=True)
|
||||
with open(file_path, "wb") as file:
|
||||
for data in response.iter_content(block_size):
|
||||
progress_bar.update(len(data))
|
||||
file.write(data)
|
||||
progress_bar.close()
|
||||
return "Model downloaded successfully!"
|
||||
|
||||
with gr.Row():
|
||||
download_status = gr.Textbox(label="Download Status")
|
||||
|
||||
download_button.click(download_model, inputs=model_dropdown, outputs=download_status)
|
||||
|
||||
def show_confirm_cancel(model_name):
|
||||
all_files_exists = all(os.path.exists(os.path.join(main_dir, "models", "parler", os.path.basename(file.split('?')[0]))) for model in available_models["models"] if model["model_name"] == model_name for file in model["files_to_download"])
|
||||
if all_files_exists:
|
||||
return [gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)]
|
||||
else:
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)]
|
||||
|
||||
def confirm_download(model_name):
|
||||
download_status_text = download_model(model_name, force_download=True)
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), download_status_text]
|
||||
|
||||
def cancel_download():
|
||||
return [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), ""]
|
||||
|
||||
with gr.Row():
|
||||
confirm_button = gr.Button("Download Anyway", visible=False)
|
||||
cancel_button = gr.Button("Cancel", visible=False)
|
||||
|
||||
download_button.click(show_confirm_cancel, inputs=model_dropdown, outputs=[confirm_button, download_button, cancel_button])
|
||||
confirm_button.click(confirm_download, inputs=model_dropdown, outputs=[confirm_button, download_button, cancel_button, download_status])
|
||||
cancel_button.click(cancel_download, inputs=None, outputs=[confirm_button, download_button, cancel_button, download_status])
|
||||
|
||||
#####################################
|
||||
# Voices Editor - Unique to Parler #
|
||||
#####################################
|
||||
# Load voices from the JSON file
|
||||
def load_voices():
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
if os.path.exists(voices_file):
|
||||
with open(voices_file, "r") as f:
|
||||
return json.load(f)
|
||||
else:
|
||||
return {"voices": []}
|
||||
|
||||
# Save voices to the JSON file
|
||||
def save_voices(voices):
|
||||
voices_file = os.path.join(main_dir, "system", "tts_engines", "parler", "parler_voices.json")
|
||||
with open(voices_file, "w") as f:
|
||||
json.dump(voices, f, indent=2)
|
||||
|
||||
# Convert voices to a list for display
|
||||
def voices_to_list(voices):
|
||||
return [f"{i+1}. {voice['voice_name']} - {voice['description']}" for i, voice in enumerate(voices["voices"])]
|
||||
|
||||
# Populate text boxes with selected voice data
|
||||
def populate_voice_data(selected):
|
||||
if selected:
|
||||
index = int(selected.split(".")[0]) - 1
|
||||
voices = load_voices()
|
||||
selected_voice = voices["voices"][index]
|
||||
return selected_voice["voice_name"], selected_voice["description"], index, gr.update(interactive=False), "Save Voice"
|
||||
return "", "", -1, gr.update(interactive=True), "Add/Save Voice"
|
||||
|
||||
# Add or update voice
|
||||
def add_update_voice(voice_name, description, selected_index, button_label):
|
||||
voices = load_voices()
|
||||
if selected_index >= 0:
|
||||
voices["voices"][selected_index]["description"] = description
|
||||
else:
|
||||
voices["voices"].append({"voice_name": voice_name, "description": description})
|
||||
save_voices(voices)
|
||||
return (gr.Dropdown(choices=voices_to_list(voices), value=voices_to_list(voices)[0] if voices["voices"] else None),
|
||||
"", "", -1, gr.update(interactive=True), "Add/Save Voice")
|
||||
|
||||
# Delete selected voice
|
||||
def delete_selected_voice(selected_index):
|
||||
voices = load_voices()
|
||||
if selected_index >= 0:
|
||||
del voices["voices"][selected_index]
|
||||
save_voices(voices)
|
||||
return (gr.Dropdown(choices=voices_to_list(voices), value=voices_to_list(voices)[0] if voices["voices"] else None),
|
||||
"", "", -1, gr.update(interactive=True), "Add/Save Voice")
|
||||
|
||||
# Clear text boxes
|
||||
def clear_text_boxes():
|
||||
return "", "", -1, gr.update(interactive=True), "Add/Save Voice"
|
||||
|
||||
# Load initial voices
|
||||
voices = load_voices()
|
||||
|
||||
with gr.Tab("Voice Editor"):
|
||||
with gr.Row():
|
||||
voice_name = gr.Textbox(label="Voice Name", scale=1)
|
||||
description = gr.Textbox(label="Voice Description", lines=4, scale=3)
|
||||
with gr.Row():
|
||||
add_update_button = gr.Button("Add/Save Voice")
|
||||
delete_button = gr.Button("Delete Selected Voice")
|
||||
clear_button = gr.Button("Clear Text Boxes")
|
||||
with gr.Row():
|
||||
voice_list = gr.Dropdown(
|
||||
choices=voices_to_list(voices),
|
||||
label="Voices",
|
||||
interactive=True,
|
||||
value=voices_to_list(voices)[0] if voices["voices"] else None
|
||||
)
|
||||
# Hidden state for selected index and button label
|
||||
selected_index = gr.Number(value=-1, visible=False)
|
||||
button_label = gr.State(value="Add/Save Voice")
|
||||
|
||||
# Button click events
|
||||
voice_list.change(populate_voice_data, inputs=voice_list, outputs=[voice_name, description, selected_index, voice_name, button_label])
|
||||
add_update_button.click(add_update_voice, inputs=[voice_name, description, selected_index, button_label], outputs=[voice_list, voice_name, description, selected_index, voice_name, button_label])
|
||||
delete_button.click(delete_selected_voice, inputs=selected_index, outputs=[voice_list, voice_name, description, selected_index, voice_name, button_label])
|
||||
clear_button.click(clear_text_boxes, outputs=[voice_name, description, selected_index, voice_name, button_label])
|
||||
|
||||
###################################################################################################
|
||||
# REQUIRED CHANGE #
|
||||
# Add any engine specific help, bugs, issues, operating system specifc requirements/setup in here #
|
||||
# Please use Markdown format, so gr.Markdown() with your markdown inside it. #
|
||||
###################################################################################################
|
||||
with gr.Tab("Engine Help"):
|
||||
with gr.Row():
|
||||
gr.Markdown("""
|
||||
### 🟧 Where are the Parler models stored?
|
||||
This extension will download the models to `/alltalk_tts/models/parler/` folder.<br>
|
||||
|
||||
### 🟧 Where are the voices stored for parler models?
|
||||
Parler is unlike any other TTS engine. It is actually mode like Stable Diffusion, where you write a text based description of what you want your image to look like, but in this case, you describe what you want the voice to sound like. For example `A female speaker with an enthusiastic and lively voice. Her tone is bright and energetic, with a fast pace and a lot of inflection. The audio is clear and vibrant.`
|
||||
|
||||
This means you can have pretty much any voice you want, however, the downside to this is, there is little consistency of the voices, meaning that no 2x TTS generations will sound exactly like the same voice. You can however use one of the `Native` built in voices to give a level of consistency. Please refer to the Parler website for more infomation https://github.com/huggingface/parler-tts.<br>
|
||||
|
||||
### 🟧 So how do I create my voices for Parler?
|
||||
On the settings page for Parler, there is a Voice Editor tab. In there you can add/remove/ammend voices as you wish, or edit the `parler_voices.json` file stored in `system/tts_engines/parler/`.
|
||||
|
||||
### 🟧 Where are the outputs stored & Automatic output wav file deletion
|
||||
Voice outputs are stored in `/alltalk_tts/outputs/`. You can configure automatic maintenance deletion of old wav files by setting `Del WAV's older than` in the global settings.<br>
|
||||
|
||||
> When `Disabled`, your output wav files will be left untouched.<br>
|
||||
> When set to a setting `1 Day` or greater, your output wav files older than that time period will be automatically deleted on start-up of AllTalk.<br>
|
||||
|
||||
### 🟧 Skipped text/speech
|
||||
At time of writing, this model is 1x day old and I could see reports of issues with some TTS not being generated, some longer text missing the middle portion of the text etc. You can confirm what AllTalk sent to the Parler TTS engine by looking at the command prompt/terminal window. If your text shows up there, it was sent to the Parler TTS engine. Please refer to the Parler website for more details https://github.com/huggingface/parler-tts.
|
||||
""")
|
||||
gr.Markdown("""
|
||||
### 🟧 How do I use Parler's voice generation system?
|
||||
To create a voice with Parler:
|
||||
|
||||
Describe the voice characteristics in your prompt. For example: `Jon's voice is monotone yet slightly fast in delivery, with a very close recording that almost has no background noise.`
|
||||
|
||||
Use one of the 34 inbuilt `Native` speaker names for a level of consistency: `Aaron, Alisa, Anna, Barbara, Bill, Brenda, Bruce, Carol, David, Eileen, Emily, Eric, Gary, James, Jason, Jenna, Jerry, Jon, Jordan, Joy, Karen, Laura, Lauren, Lea, Mike, Naomie, Patrick, Rebecca, Rick, Rose, Tina, Tom, Will, Yann`
|
||||
|
||||
Control audio quality in the voice characteristics prompt:
|
||||
|
||||
> Include `very clear audio` for highest quality<br>
|
||||
> Use `very noisy audio` for high levels of background noise
|
||||
|
||||
Adjust other features in your voice characteristics prompt:
|
||||
|
||||
> Gender<br>
|
||||
> Speaking rate<br>
|
||||
> Pitch<br>
|
||||
> Reverberation
|
||||
|
||||
Remember, while you can create diverse voices, each generation may sound slightly different even with the same description.
|
||||
|
||||
In the text sent for TTS generation, you can use punctuation to control speech rhythm (e.g. use `,` commas for small breaks)
|
||||
|
||||
### 🟧 UserWarning: 1Torch was not compiled with flash attention
|
||||
This is a limitation of Windows Pytorch not yet having full flash attention support on later builds of Pytorch (is my understanding at this time).
|
||||
""")
|
||||
|
||||
return app
|
||||
|
||||
################################
|
||||
# REQUIRED CHANGE #
|
||||
# Sets up the Gradio interface #
|
||||
################################
|
||||
# This function sets up the Gradio interface for your TTS engine's settings page.
|
||||
# You need to change the name of the function calls to match the names you set in the functions above.
|
||||
#
|
||||
# Specifically, you need to update the following:
|
||||
#
|
||||
# 1. The name of the function `xxxx_at_gradio_settings_page` to match your engine's name.
|
||||
# For example, if your engine is named "mytts", the function should be named `mytts_at_gradio_settings_page`.
|
||||
#
|
||||
# 2. The name of the function call `xxxx_model_alltalk_settings(model_config_data)`.
|
||||
# This should match the name you gave to the function that sets up the engine-specific settings page in Gradio.
|
||||
# If you named that function `mytts_model_alltalk_settings`, then the call should be:
|
||||
# `mytts_model_alltalk_settings(model_config_data)`
|
||||
#
|
||||
# After making these changes, this function will create and return the Gradio app for your TTS engine's settings page.
|
||||
def parler_at_gradio_settings_page(model_config_data):
|
||||
app = parler_model_alltalk_settings(model_config_data)
|
||||
return app
|
||||
def parler_at_gradio_settings_page(model_config_data):
|
||||
app = parler_model_alltalk_settings(model_config_data)
|
||||
return app
|
||||
@@ -1,96 +1,232 @@
|
||||
{
|
||||
"voices": [
|
||||
{
|
||||
"voice_name": "soothing_female",
|
||||
"description": "A female speaker with a soft, soothing voice. Her tone is gentle and calming, with a slight reverb as if she is in a small, cozy room. She speaks slowly and clearly."
|
||||
},
|
||||
{
|
||||
"voice_name": "authoritative_male",
|
||||
"description": "A male speaker with a strong, authoritative voice. His tone is firm and commanding, with a slight bass boost that adds to the depth. He speaks at a moderate pace."
|
||||
},
|
||||
{
|
||||
"voice_name": "friendly_young_male",
|
||||
"description": "A young male speaker with a friendly, approachable voice. He has a slightly high-pitched tone and speaks with enthusiasm. The audio is clear and bright."
|
||||
},
|
||||
{
|
||||
"voice_name": "whispering_female",
|
||||
"description": "A female speaker who whispers softly, creating an intimate and personal feel. The audio has a slight breathy quality, adding to the whispering effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "robotic_male",
|
||||
"description": "A male speaker with a robotic, synthetic voice. The tone is monotone and slightly mechanical, with clear and precise pronunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "dramatic_female",
|
||||
"description": "A female speaker with a dramatic and expressive voice. She speaks with varying pitch and intensity, in a high-quality audio environment. Her pace varies to add emphasis."
|
||||
},
|
||||
{
|
||||
"voice_name": "casual_male",
|
||||
"description": "A male speaker with a casual, conversational tone. He speaks in a relaxed manner, with natural pauses and inflections. The audio is clear and slightly ambient."
|
||||
},
|
||||
{
|
||||
"voice_name": "bright_young_female",
|
||||
"description": "A young female speaker with a bright, cheerful voice. She speaks with a lot of enthusiasm and a slightly fast pace, in a clear and vibrant audio environment."
|
||||
},
|
||||
{
|
||||
"voice_name": "calm_older_male",
|
||||
"description": "An older male speaker with a calm, soothing voice. His tone is deep and steady, with a slight reverb suggesting a large, quiet room. He speaks slowly and deliberately."
|
||||
},
|
||||
{
|
||||
"voice_name": "energetic_child",
|
||||
"description": "A child speaker with a high-pitched, energetic voice. The tone is playful and lively, with clear and crisp audio quality. The speech pace is fast."
|
||||
},
|
||||
{
|
||||
"voice_name": "mysterious_female",
|
||||
"description": "A female speaker with a low, mysterious voice. Her tone is hushed and slightly breathy, creating an air of secrecy. The audio has a slight echo effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "confident_male",
|
||||
"description": "A male speaker with a confident and assertive voice. His tone is strong and clear, with a moderate pace and slight emphasis on certain words to convey assurance."
|
||||
},
|
||||
{
|
||||
"voice_name": "gentle_young_female",
|
||||
"description": "A young female speaker with a gentle, kind voice. Her tone is soft and sweet, with clear audio quality. She speaks at a slow and soothing pace."
|
||||
},
|
||||
{
|
||||
"voice_name": "excited_male",
|
||||
"description": "A male speaker with an excited, enthusiastic voice. His tone is lively and energetic, with a fast pace and a lot of inflection. The audio is clear and vibrant."
|
||||
},
|
||||
{
|
||||
"voice_name": "professional_female",
|
||||
"description": "A female speaker with a professional and articulate voice. Her tone is clear and measured, with precise pronunciation and a moderate pace. The audio is high-quality."
|
||||
},
|
||||
{
|
||||
"voice_name": "relaxed_male",
|
||||
"description": "A male speaker with a relaxed and laid-back voice. His tone is warm and soothing, with a slight drawl. The audio has a comfortable and intimate feel."
|
||||
},
|
||||
{
|
||||
"voice_name": "melancholic_female",
|
||||
"description": "A female speaker with a melancholic and introspective voice. Her tone is soft and slightly sad, with a slow and deliberate pace. The audio has a gentle reverb effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "cheerful_child",
|
||||
"description": "A child speaker with a cheerful and happy voice. The tone is high-pitched and playful, with a fast pace and clear audio quality. The speech is lively and energetic."
|
||||
},
|
||||
{
|
||||
"voice_name": "serious_male",
|
||||
"description": "A male speaker with a serious and formal voice. His tone is deep and authoritative, with a slow and deliberate pace. The audio is clear and slightly reverberant."
|
||||
},
|
||||
{
|
||||
"voice_name": "enthusiastic_female",
|
||||
"description": "A female speaker with an enthusiastic and lively voice. Her tone is bright and energetic, with a fast pace and a lot of inflection. The audio is clear and vibrant."
|
||||
},
|
||||
{
|
||||
"voice_name": "soft_spoken_male",
|
||||
"description": "A male speaker with a soft-spoken and gentle voice. His tone is warm and soothing, with a slow and measured pace. The audio has a slight ambient quality."
|
||||
},
|
||||
{
|
||||
"voice_name": "deep_female",
|
||||
"description": "A female speaker with a deep and resonant voice. Her tone is strong and authoritative, with a slow and deliberate pace. The audio is clear and slightly reverberant."
|
||||
},
|
||||
{
|
||||
"voice_name": "lighthearted_male",
|
||||
"description": "A male speaker with a lighthearted and cheerful voice. His tone is warm and friendly, with a moderate pace and natural inflection. The audio is clear and bright."
|
||||
}
|
||||
]
|
||||
{
|
||||
"voices": [
|
||||
{
|
||||
"voice_name": "Native Aaron",
|
||||
"description": "Aaron's voice is a deep male voice with a resonant quality. He speaks slowly and clearly, with a commanding presence."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Alisa",
|
||||
"description": "Alisa's voice is a crisp female voice with a hint of assertiveness. She speaks at a brisk pace with clear enunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Anna",
|
||||
"description": "Anna's voice is a soft female voice with a slight Eastern European accent. She speaks at a careful pace with gentle tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Barbara",
|
||||
"description": "Barbara's voice is a sophisticated female voice with a slight huskiness. She speaks at a measured pace with refined intonation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Bill",
|
||||
"description": "Bill's voice is a mature male voice with a gravelly texture. He speaks slowly and deliberately, with a wise tone."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Brenda",
|
||||
"description": "Brenda's voice is a soft-spoken female voice with a gentle, soothing quality. She speaks slowly and clearly."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Bruce",
|
||||
"description": "Bruce's voice is a strong male voice with a commanding presence. He speaks at a steady pace with authoritative tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Carol",
|
||||
"description": "Carol's voice is a mature female voice with a warm, motherly quality. She speaks at a gentle pace with nurturing tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native David",
|
||||
"description": "David's voice is an authoritative male voice with a rich timbre. He speaks at a steady pace with precise diction."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Eileen",
|
||||
"description": "Eileen's voice is a mature female voice with a slight breathiness. She speaks at a relaxed pace with a warm tone."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Emily",
|
||||
"description": "Emily's voice is a sweet female voice with a youthful quality. She speaks at a lively pace with cheerful inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Eric",
|
||||
"description": "Eric's voice is a casual male voice with a hint of gruffness. He speaks at a relaxed pace with natural pauses."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Gary",
|
||||
"description": "Gary's voice is a deep, resonant male voice with a hint of gravel. He speaks slowly and deliberately, giving weight to each word."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native James",
|
||||
"description": "James's voice is a refined male voice with a slight British accent. He speaks at a moderate pace with crisp articulation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Jason",
|
||||
"description": "Jason's voice is a calm male voice with a neutral accent. He speaks at a moderate pace with even tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Jenna",
|
||||
"description": "Jenna's voice is a young female voice with a slight vocal fry. She speaks at a moderate pace with modern inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Jerry",
|
||||
"description": "Jerry's voice is a jovial male voice with a hint of a chuckle. He speaks at a relaxed pace with natural inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Jon",
|
||||
"description": "Jon's voice is monotone yet slightly fast in delivery, with a very close recording that almost has no background noise."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Jordan",
|
||||
"description": "Jordan's voice is a youthful, gender-neutral voice with a slight rasp. They speak at a variable pace with natural inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Joy",
|
||||
"description": "Joy's voice is a cheerful female voice with a musical quality. She speaks at a lively pace with expressive intonation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Karen",
|
||||
"description": "Karen's voice is a mature female voice with a slight huskiness. She speaks at a measured pace with a calming tone."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Laura",
|
||||
"description": "Laura's voice is a warm and friendly female tone with a slight lilt. She speaks at a moderate pace with clear enunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Lauren",
|
||||
"description": "Lauren's voice is a professional female voice with a confident tone. She speaks at a steady pace with clear diction."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Lea",
|
||||
"description": "Lea's voice is a bright and energetic female voice with a slight upward inflection at the end of sentences. She speaks at a brisk pace."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Mike",
|
||||
"description": "Mike's voice is a friendly male voice with a hint of enthusiasm. He speaks at a brisk pace with clear enunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Naomie",
|
||||
"description": "Naomie's voice is a vibrant female voice with a slight exotic accent. She speaks at a variable pace with expressive tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Patrick",
|
||||
"description": "Patrick's voice is a warm male voice with a slight Irish lilt. He speaks at a moderate pace with a friendly tone."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Rebecca",
|
||||
"description": "Rebecca's voice is a clear female voice with a hint of enthusiasm. She speaks at a moderate pace with engaging inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Rick",
|
||||
"description": "Rick's voice is a confident male voice with a slight regional accent. He speaks at a moderate pace with clear articulation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Rose",
|
||||
"description": "Rose's voice is a gentle female voice with a slight breathiness. She speaks slowly and softly, with a soothing quality."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Tina",
|
||||
"description": "Tina's voice is a bubbly female voice with an upbeat quality. She speaks at a lively pace with enthusiastic tones."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Tom",
|
||||
"description": "Tom's voice is a neutral male voice with a clear tone. He speaks at a steady pace with precise articulation."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Will",
|
||||
"description": "Will's voice is an energetic male voice with a youthful tone. He speaks at a fast pace with enthusiastic inflections."
|
||||
},
|
||||
{
|
||||
"voice_name": "Native Yann",
|
||||
"description": "Yann's voice is a sophisticated male voice with a slight foreign accent. He speaks at a measured pace with precise pronunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "soothing_female",
|
||||
"description": "A female speaker with a soft, soothing voice. Her tone is gentle and calming, with a slight reverb as if she is in a small, cozy room."
|
||||
},
|
||||
{
|
||||
"voice_name": "authoritative_male",
|
||||
"description": "Nicks voice is a male speaker with a strong, authoritative voice. His tone is firm and commanding, with a slight bass boost that adds to the depth. He speaks at a moderate pace."
|
||||
},
|
||||
{
|
||||
"voice_name": "friendly_young_male",
|
||||
"description": "A young male speaker with a friendly, approachable voice. He has a slightly high-pitched tone and speaks with enthusiasm. The audio is clear and bright."
|
||||
},
|
||||
{
|
||||
"voice_name": "whispering_female",
|
||||
"description": "A female speaker who whispers softly, creating an intimate and personal feel. The audio has a slight breathy quality, adding to the whispering effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "robotic_male",
|
||||
"description": "A male speaker with a robotic, synthetic voice. The tone is monotone and slightly mechanical, with clear and precise pronunciation."
|
||||
},
|
||||
{
|
||||
"voice_name": "dramatic_female",
|
||||
"description": "A female speaker with a dramatic and expressive voice. She speaks with varying pitch and intensity, in a high-quality audio environment. Her pace varies to add emphasis."
|
||||
},
|
||||
{
|
||||
"voice_name": "casual_male",
|
||||
"description": "A male speaker with a casual, conversational tone. He speaks in a relaxed manner, with natural pauses and inflections. The audio is clear and slightly ambient."
|
||||
},
|
||||
{
|
||||
"voice_name": "bright_young_female",
|
||||
"description": "A young female speaker with a bright, cheerful voice. She speaks with a lot of enthusiasm and a slightly fast pace, in a clear and vibrant audio environment."
|
||||
},
|
||||
{
|
||||
"voice_name": "calm_older_male",
|
||||
"description": "An older male speaker with a calm, soothing voice. His tone is deep and steady, with a slight reverb suggesting a large, quiet room. He speaks slowly and deliberately."
|
||||
},
|
||||
{
|
||||
"voice_name": "energetic_child",
|
||||
"description": "A child speaker with a high-pitched, energetic voice. The tone is playful and lively, with clear and crisp audio quality. The speech pace is fast."
|
||||
},
|
||||
{
|
||||
"voice_name": "mysterious_female",
|
||||
"description": "A female speaker with a low, mysterious voice. Her tone is hushed and slightly breathy, creating an air of secrecy. The audio has a slight echo effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "confident_male",
|
||||
"description": "A male speaker with a confident and assertive voice. His tone is strong and clear, with a moderate pace and slight emphasis on certain words to convey assurance."
|
||||
},
|
||||
{
|
||||
"voice_name": "gentle_young_female",
|
||||
"description": "A young female speaker with a gentle, kind voice. Her tone is soft and sweet, with clear audio quality. She speaks at a slow and soothing pace."
|
||||
},
|
||||
{
|
||||
"voice_name": "excited_male",
|
||||
"description": "A male speaker with an excited, enthusiastic voice. His tone is lively and energetic, with a fast pace and a lot of inflection. The audio is clear and vibrant."
|
||||
},
|
||||
{
|
||||
"voice_name": "professional_female",
|
||||
"description": "A female speaker with a professional and articulate voice. Her tone is clear and measured, with precise pronunciation and a moderate pace. The audio is high-quality."
|
||||
},
|
||||
{
|
||||
"voice_name": "relaxed_male",
|
||||
"description": "A male speaker with a relaxed and laid-back voice. His tone is warm and soothing, with a slight drawl. The audio has a comfortable and intimate feel."
|
||||
},
|
||||
{
|
||||
"voice_name": "melancholic_female",
|
||||
"description": "A female speaker with a melancholic and introspective voice. Her tone is soft and slightly sad, with a slow and deliberate pace. The audio has a gentle reverb effect."
|
||||
},
|
||||
{
|
||||
"voice_name": "cheerful_child",
|
||||
"description": "A child speaker with a cheerful and happy voice. The tone is high-pitched and playful, with a fast pace and clear audio quality. The speech is lively and energetic."
|
||||
},
|
||||
{
|
||||
"voice_name": "serious_male",
|
||||
"description": "A male speaker with a serious and formal voice. His tone is deep and authoritative, with a slow and deliberate pace. The audio is clear and slightly reverberant."
|
||||
},
|
||||
{
|
||||
"voice_name": "enthusiastic_female",
|
||||
"description": "A female speaker with an enthusiastic and lively voice. Her tone is bright and energetic, with a fast pace and a lot of inflection. The audio is clear and vibrant."
|
||||
},
|
||||
{
|
||||
"voice_name": "soft_spoken_male",
|
||||
"description": "A male speaker with a soft-spoken and gentle voice. His tone is warm and soothing, with a slow and measured pace. The audio has a slight ambient quality."
|
||||
},
|
||||
{
|
||||
"voice_name": "deep_female",
|
||||
"description": "A female speaker with a deep and resonant voice. Her tone is strong and authoritative, with a slow and deliberate pace. The audio is clear and slightly reverberant."
|
||||
},
|
||||
{
|
||||
"voice_name": "lighthearted_male",
|
||||
"description": "A male speaker with a lighthearted and cheerful voice. His tone is warm and friendly, with a moderate pace and natural inflection. The audio is clear and bright."
|
||||
}
|
||||
]
|
||||
}
|
||||
+3
-2
@@ -1137,7 +1137,7 @@ class JSONInput(BaseModel):
|
||||
rvcnarrator_voice_gen: str = Field(..., description="rvcnarrator_voice_gen needs to be the name of a valid pth file in the 'folder\\file.pth' format or the word 'Disabled'.")
|
||||
rvcnarrator_pitch: float = Field(..., description="RVC Narrator pitch needs to be a number between -24 and 24")
|
||||
text_not_inside: str = Field(..., pattern="^(character|narrator|silent)$", description="text_not_inside needs to be 'character', 'narrator' or 'silent'.")
|
||||
language: str = Field(..., pattern="^(ar|zh-cn|cs|nl|en|fr|de|hu|hi|it|ja|ko|pl|pt|ru|es|tr)$", description="language needs to be one of the following: ar, zh-cn, cs, nl, en, fr, de, hu, hi, it, ja, ko, pl, pt, ru, es, tr.")
|
||||
language: str = Field(..., pattern="^(ar|zh-cn|zh|cs|nl|en|fr|de|hu|hi|it|ja|ko|pl|pt|ru|es|tr)$", description="language needs to be one of the following: ar, zh-cn, zh, cs, nl, en, fr, de, hu, hi, it, ja, ko, pl, pt, ru, es, tr.")
|
||||
output_file_name: str = Field(..., pattern="^[a-zA-Z0-9_]+$", description="output_file_name needs to be the name without any special characters or file extension, e.g., 'filename'.")
|
||||
output_file_timestamp: bool = Field(..., description="output_file_timestamp needs to be true or false.")
|
||||
autoplay: bool = Field(..., description="autoplay needs to be a true or false value.")
|
||||
@@ -1216,7 +1216,8 @@ def get_params_speed():
|
||||
def get_params_temperature():
|
||||
return model_engine.temperature_set
|
||||
def get_params_repetition():
|
||||
return model_engine.repetitionpenalty_set
|
||||
value = model_engine.repetitionpenalty_set
|
||||
return float(str(value).replace(',', '.'))
|
||||
def get_params_pitch():
|
||||
return model_engine.pitch_set
|
||||
def get_character_voice_gen():
|
||||
|
||||
Reference in new issue
Block a user