Click to expand
@@ -254,19 +273,27 @@ I would suggest that if its in an interview format, you cut out the interviewer
#### β« Important requirements CUDA 11.8
As mentioned you must have a small portion of the Nvidia CUDA Toolkit **11.8** installed. Not higher or lower versions. Specifically **11.8**. You do not have to uninstall any other versions, change any graphics drivers, reinstall torch or anything like that. To keep the download+install as small as possible, you will need to:
- Download the **xxx (network)** install of the Nvidia Cuda Toolkit 11.8 from [here](https://developer.nvidia.com/cuda-11-8-0-download-archive)
-- When you run the installation, select **Custom Advanced** as your installation type. Uncheck `CUDA`, `Other Components` and `Driver Components` at the top then expand `CUDA` > `Development` > `Compiler` > and select `nvcc` then expand `Libraries` and select `CUBLAS`
-- Back at the top of `CUDA` expand `Runtime` > `Libraries` and select `CUBLAS`. You can now next through the install.
+- Run the installer. At minimum, you need to [minimally] install the `nvcc` compiler and the `CUBLAS` development and runtime libraries:
+ - Select **Custom Advanced** as your installation type.
+ - Uncheck all the checkboxes in the list.
+ - Now check the following elements:
+ - `CUDA` > `Development` > `Compiler` > `nvcc`
+ - `CUDA` > `Development` > `Compiler` > `Libraries` > `CUBLAS`
+ - `CUDA` > `Runtime` > `Libraries` > `CUBLAS`
+ - You can now proceed through the install.
- When that has installed, open a terminal/command prompt and type `nvcc --version`. If it reports back `Cuda compilation tools, release 11.8.` you are good to go. **Specifically, 11.8**. If not continue to the next step.
-- For both Windows an Linux, you will need to ensure that `nvcc` and the 11.8 cuda library files are in your environments search path. You can undo the changes below after finetuning if you prefer.
- **Windows** - Edit the Windows PATH environment variable and add `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin`
- **Linux** - The path may be different depending on what flavour of Linux you are running, so you may need to seek out specific instructions on the internet. Generic paths **may** be:
- `export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}` and
- `export LD_LIBRARY_PATH=/usr/local/cuda-11.8/bin`
- Add these to your '~/.bashrc' if you want this to be permanent and not something you have to set each time you open a new terminal.
-- When you have made the changes, open a new terminal/command prompt and `nvcc --version`. It should report back `Cuda compilation tools, release 11.8.` at which point, you are good to go.
-- If it doesnt report that, check you have correctly set the search environment paths, dont have overlapping other versions of cuda paths etc.
+- For both Windows and Linux, you need to ensure that `nvcc` and the 11.8 cuda library files are in your environments search path. You can undo the changes below after finetuning if you prefer:
+ - **Windows**: Edit the Windows `Path` environment variable and add `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin`
+ - **Linux**: The path may be different depending on what flavour of Linux you are running, so you may need to seek out specific instructions on the internet. Generic paths **may** be:
+ - `export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}` and
+ - `export LD_LIBRARY_PATH=/usr/local/cuda-11.8/bin`
+ - Add these to your '~/.bashrc' if you want this to be permanent and not something you have to set each time you open a new terminal.
+- When you have made the changes, open a **new** terminal/command prompt (in order to load the new search paths) and `nvcc --version`. It should report back `Cuda compilation tools, release 11.8.` at which point, you are good to go.
+- If it doesn't report that, check you have correctly set the search environment paths, dont have overlapping other versions of cuda paths etc.
#### β« Starting Finetuning
+**NOTE:** Please make sure you have started AllTalk at least once after updating, so that it downloads the additional files needed for finetuning.
+
1) Close all other applications that are using your GPU/VRAM and copy your audio samples into:
`/alltalk_tts/finetune/put-voice-samples-in-here/`
3) In a command prompt/terminal window you need to move into your Text generation webUI folder:
@@ -291,6 +318,8 @@ As mentioned you must have a small portion of the Nvidia CUDA Toolkit **11.8** i
12) When you have finished finetuning, the final tab will tell you what to do with your files and how to move your newly trained model to the correct location on disk.
## π΅π’π‘ DeepSpeed Installation Options
+**NOTE**: You **DO NOT** need to set Text-generation-webUI's **--deepspeed** setting for AllTalk to be able to use DeepSpeed. These are two completely separate things and incorrectly setting that on Text-generation-webUI may cause other complications.
+
#### π΅ Linux Installation
Click to expand: Linux DeepSpeed installation
@@ -441,6 +470,31 @@ def get_shm_size():
**Removal** - If it became necessary to uninstall DeepSpeed, you can do so with `cmd_windows.bat` and then `pip uninstall deepspeed`
+## π¦ Running AllTalk as a standalone app
+AllTalk will run as a standalone app, as long as you install its requirements files into whatever Python environment you are using. You can follow the steps to install the AllTalk's requirements into whatever Python environment you wish. Because I dont know what Python environment you are wanting to use, I can only give you a loose set of installation instructions.
+
+Please note, at time of writing, the TTS engine requires Python **3.9.x** to **3.11.x** [TTS Engine details here](https://pypi.org/project/TTS/)
+
+#### π¦ (Option 1) I already have AllTalk installed as an extension of Text-generation-webui
+If you already have AllTalk as a extension of Text-generation-webui, and wish to run it as standalone, load Text-generation-webui's Python environment `cmd_windows.bat`, `./cmd_linux.sh`, `cmd_macos.sh` or `cmd_wsl.bat`, move into the AllTalk folder `cd extensions` > `cd alltalk_tts` and start AllTalk with `python script.py`. There is nothing beyond this you would need to do.
+
+#### π¦ (Option 2) I wish to do a custom install of AllTalk
+1) In your chosen disk location git clone this repository:
+`git clone https://github.com/erew123/alltalk_tts`
+
+2) Move into the **alltalk_tts** folder:
+`cd alltalk_tts`
+
+3) If you have a custom Python environment, now is the time to load it!
+
+4) Install one of the two requirements files. Whichever one of the two is correct for your machine type:
+**Nvidia graphics card machines** - `pip install -r requirements_nvidia.txt`
+**Other machines (mac, amd etc)** - `pip install -r requirements_other.txt`
+
+5) Start AllTalk with `python script.py`
+
+Deepspeed and other such things can be installed. Please read the relevant instructions for those items, however, make the relevant changes to load your correct Python environment when installing any requirements files and starting AllTalk.
+
## π API Suite and JSON-CURL
### π Overview
The Text-to-Speech (TTS) Generation API allows you to generate speech from text input using various configuration options. This API supports both character and narrator voices, providing flexibility for creating dynamic and engaging audio content.
diff --git a/diagnostics.py b/diagnostics.py
index 6ede8fa..ca5466a 100644
--- a/diagnostics.py
+++ b/diagnostics.py
@@ -6,6 +6,7 @@ try:
import torch
import os # Import the os module
import re
+ import sys
import glob
import textwrap
import packaging.version
@@ -17,7 +18,7 @@ except ImportError as e:
print(f"\033[91mError importing module: {e}\033[0m\n")
print("\033[94mPlease ensure you started the Text-generation-webUI Python environment with either\033[0m")
print("\033[92mcmd_linux.sh\033[0m, \033[92mcmd_windows.bat\033[0m, \033[92mcmd_macos.sh\033[0m, or \033[92mcmd_wsl.bat\033[0m")
- print("\033[94mand then try running the diagnostics again.\033[0m")
+ print("\033[94mfrom the text-generation-webui directory, and then try running the diagnostics again.\033[0m")
exit(1)
try:
@@ -110,6 +111,18 @@ def log_system_info():
# Package versions using importlib_metadata
package_versions = {d.metadata['Name']: d.version for d in distributions()}
+ # Python environment information
+ python_executable = sys.executable
+ python_version_info = sys.version_info
+ python_virtual_env = os.environ.get('VIRTUAL_ENV', 'N/A')
+
+ # Conda environment information
+ conda_env = os.environ.get('CONDA_DEFAULT_ENV', 'N/A')
+
+ # Get Path environment information
+ search_path = sys.path
+ path_env = os.environ.get('PATH', 'N/A')
+
# Compare with requirements file
requirements_file = get_requirements_file()
if requirements_file:
@@ -134,14 +147,24 @@ def log_system_info():
# Log and print information
logging.info(f"OS Version: {os_version}")
- logging.info(f"Note: Windows 11 build is 10.x.22xxx")
- logging.info(f"Python Version: {python_version}")
+ logging.info(f"Note: Windows 11 will list as build is 10.x.22xxx")
logging.info(f"Torch Version: {torch_version}")
logging.info(f"System RAM: {system_ram}")
logging.info(f"CUDA_HOME: {cuda_home}")
logging.info(f"Port Status: {port_status}")
+ logging.info(f"Python Version: {platform.python_version()}")
+ logging.info(f"Python Version Info: {python_version_info}")
+ logging.info(f"Python Executable: {python_executable}")
+ logging.info(f"Python Virtual Environment: {python_virtual_env} (Should be N/A when in Text-generation-webui Conda Python environment)")
+ logging.info(f"Conda Environment: {conda_env}")
+ logging.info("\nPython Search Path:")
+ for path in search_path:
+ logging.info(f" {path}")
+ logging.info("\nOS PATH Environment Variable:")
+ for path in path_env.split(';'):
+ logging.info(f" {path}")
if required_packages: # Check if the dictionary is not empty
- logging.info("Package Versions:")
+ logging.info("\nPackage Versions:")
max_package_length = max(len(package) for package in required_packages.keys())
for package_name, (operator, required_version) in required_packages.items():
installed_version = installed_packages.get(package_name, 'Not installed')
@@ -153,12 +176,22 @@ def log_system_info():
# Print to screen
print(f"\n\033[94mOS Version:\033[0m \033[92m{os_version}\033[0m")
- print(f"\033[94mOS Ver note:\033[0m \033[92mWindows 11 build is 10.x.22xxx\033[0m")
+ print(f"\033[94mOS Ver note:\033[0m \033[92m(Windows 11 will say build is 10.x.22xxx)\033[0m")
print(f"\033[94mCUDA_HOME:\033[0m \033[92m{cuda_home}\033[0m")
print(f"\033[94mSystem RAM:\033[0m \033[92m{system_ram}\033[0m")
print(f"\033[94mPort Status:\033[0m \033[92m{port_status}\033[0m")
print(f"\033[94mTorch Version:\033[0m \033[92m{torch_version}\033[0m")
- print(f"\033[94mPython Version:\033[0m \033[92m{python_version}\033[0m")
+ print(f"\033[94mPython Version:\033[0m \033[92m{platform.python_version()}\033[0m")
+ print(f"\033[94mPython Version Info:\033[0m \033[92m{python_version_info}\033[0m")
+ print(f"\033[94mPython Executable:\033[0m \033[92m{python_executable}\033[0m")
+ print(f"\033[94mPython Virtual Environment:\033[0m \033[92m{python_virtual_env}\033[0m (Should be N/A when in Text-generation-webui Conda Python environment)")
+ print(f"\033[94mConda Environment:\033[0m \033[92m{conda_env}\033[0m")
+ print(f"\n\033[94mPython Search Path:\033[0m")
+ for path in search_path:
+ print(f" {path}")
+ print(f"\n\033[94mOS Search PATH Environment Variable:\033[0m")
+ for path in path_env.split(';'):
+ print(f" {path}")
if required_packages: # Check if the dictionary is not empty
print("\033[94m\nRequirements file package comparison:\033[0m")
max_package_length = max(len(package) for package in required_packages.keys())
diff --git a/finetune.py b/finetune.py
index 3ab2be7..f156a6f 100644
--- a/finetune.py
+++ b/finetune.py
@@ -14,6 +14,7 @@ import gc
import time
import shutil
import pandas
+import glob
import json
from pathlib import Path
from tqdm import tqdm
@@ -28,6 +29,7 @@ out_path = this_dir / "finetune" / "tmp-trn"
progress = 0
theme = gr.themes.Default()
refresh_symbol = 'π'
+os.environ['TRAINER_TELEMETRY'] = '0'
# Define the path to the modeldownload config file file
modeldownload_config_file_path = this_dir / "modeldownload.json"
@@ -39,6 +41,7 @@ if modeldownload_config_file_path.exists():
# Extract settings from the loaded JSON
base_path = Path(settings.get("base_path", ""))
model_path = Path(settings.get("model_path", ""))
+ base_model_path = Path(settings.get("model_path", ""))
files_to_download = settings.get("files_to_download", {})
else:
# Default settings if the JSON file doesn't exist or is empty
@@ -59,7 +62,7 @@ def create_temporary_file(folder, suffix=".wav"):
return os.path.join(folder, unique_filename)
def format_audio_list(target_language, whisper_model, out_path, gradio_progress=progress):
- audio_files = [os.path.join(audio_folder, file) for file in os.listdir(audio_folder) if file.endswith('.wav')]
+ audio_files = [os.path.join(audio_folder, file) for file in os.listdir(audio_folder) if file.endswith(('.mp3', '.flac', '.wav'))]
buffer=0.2
eval_percentage=0.15
speaker_name="coqui"
@@ -470,10 +473,10 @@ def run_tts(lang, tts_text, speaker_audio_file):
return "Speech generated !", out_path, speaker_audio_file
-def get_available_voices(minimum_size_kb=1300):
+def get_available_voices(minimum_size_kb=1200):
voice_files = [
voice for voice in Path(f"{this_dir}/finetune/tmp-trn/wavs").glob("*.wav")
- if voice.stat().st_size > minimum_size_kb * 1300 # Convert KB to bytes
+ if voice.stat().st_size > minimum_size_kb * 1200 # Convert KB to bytes
]
return sorted([str(file) for file in voice_files]) # Return full path as string
@@ -494,7 +497,114 @@ xtts_config_files = find_jsons(main_directory, "config.json")
# XTTS vocab files (vocab.json)
xtts_vocab_files = find_jsons(main_directory, "vocab.json")
+##########################
+#### STEP 4 AND OTHER ####
+##########################
+def find_latest_best_model(folder_path):
+ search_path = folder_path / "XTTS_FT-*" / "best_model.pth"
+ files = glob.glob(str(search_path), recursive=True)
+ latest_file = max(files, key=os.path.getctime, default=None)
+ return latest_file
+
+def compact_model():
+ this_dir = Path(__file__).parent.resolve()
+
+ best_model_path_str = find_latest_best_model(this_dir / "finetune" / "tmp-trn" / "training")
+
+ # Check if the best model file exists
+ if best_model_path_str is None:
+ print("No trained model was found.")
+ return "No trained model was found."
+
+ # Convert model_path_str to Path
+ best_model_path = Path(best_model_path_str)
+
+ # Attempt to load the model
+ try:
+ checkpoint = torch.load(best_model_path, map_location=torch.device("cpu"))
+ except Exception as e:
+ print("Error loading checkpoint:", e)
+ raise
+
+ del checkpoint["optimizer"]
+
+ for key in list(checkpoint["model"].keys()):
+ if "dvae" in key:
+ del checkpoint["model"][key]
+
+ # Define the target directory
+ target_dir = this_dir / "models" / "trainedmodel"
+
+ # Create the target directory if it doesn't exist
+ target_dir.mkdir(parents=True, exist_ok=True)
+
+ # Save the modified checkpoint in the target directory
+ torch.save(checkpoint, target_dir / "model.pth")
+
+ # Specify the files you want to copy
+ files_to_copy = ["vocab.json", "config.json", "dvae.pth", "mel_stats.pth"]
+
+ for file_name in files_to_copy:
+ src_path = this_dir / base_path / base_model_path / file_name
+ dest_path = target_dir / file_name
+ shutil.copy(str(src_path), str(dest_path))
+
+ source_wavs_dir = this_dir / "finetune" / "tmp-trn" / "wavs"
+ target_wavs_dir = target_dir / "wavs"
+ target_wavs_dir.mkdir(parents=True, exist_ok=True)
+
+ # Iterate through files in the source directory
+ for file_path in source_wavs_dir.iterdir():
+ # Check if it's a file and larger than 1000 KB
+ if file_path.is_file() and file_path.stat().st_size > 1000 * 1024:
+ # Copy the file to the target directory
+ shutil.copy(str(file_path), str(target_wavs_dir / file_path.name))
+
+ print("[FINETUNE] Model copied to '/models/trainedmodel/'")
+ return "Model copied to '/models/trainedmodel/'"
+
+
+def delete_training_data():
+ # Define the folder to be deleted
+ folder_to_delete = Path(this_dir / "finetune" / "tmp-trn")
+
+ # Check if the folder exists before deleting
+ if folder_to_delete.exists():
+ # Delete the folder and its contents
+ shutil.rmtree(folder_to_delete)
+ print(f"[FINETUNE] Folder {folder_to_delete} and its contents deleted successfully.")
+ return "Folder '/finetune/tmp-trn/' and its contents deleted successfully."
+ else:
+ print(f"[FINETUNE] Folder {folder_to_delete} does not exist.")
+ return "Folder '/finetune/tmp-trn/' does not exist."
+
+def delete_voice_sample_contents():
+ # Define the folder to be cleared
+ folder_to_clear = Path(this_dir / "finetune" / "put-voice-samples-in-here")
+
+ # Check if the folder exists before clearing its contents
+ if folder_to_clear.exists() and folder_to_clear.is_dir():
+ # List all files and subdirectories in the folder
+ for item in os.listdir(folder_to_clear):
+ item_path = folder_to_clear / item
+ if item_path.is_file():
+ # If it's a file, remove it
+ os.remove(item_path)
+ elif item_path.is_dir():
+ # If it's a subdirectory, remove it recursively
+ shutil.rmtree(item_path)
+
+ print(f"[FINETUNE] Contents of {folder_to_clear} deleted successfully.")
+ return f"Contents of 'put-voice-samples-in-here deleted' successfully."
+ else:
+ print(f"[FINETUNE] Folder {folder_to_clear} does not exist.")
+ return f"Folder 'put-voice-samples-in-here' does not exist."
+
+
+#######################
+#### OTHER Generic ####
+#######################
# define a logger to redirect
class Logger:
def __init__(self, filename="finetune.log"):
@@ -620,7 +730,7 @@ if __name__ == "__main__":
## Finetuning Information
### π₯ Important Note
#### π₯ finetune.py needs to be run from the /alltalk_tts/ folder. Don't move the location of this script.
- #### π₯ Have you run AllTalk at least once? It needs to have downloaded the voice model, before we can finetune it.
+ #### π₯ Have you run AllTalk at least once? It needs to have downloaded+updated the voice model, before we can finetune it.
### π¦ What you need to run finetuning
#### π¦ An Nvidia GPU. Tested on Windows with extended shared VRAM and training used about 16GB's total (which worked on a 12GB card).
#### π¦ I have not been able to test this on a GPU with less than 12GB of VRAM, so cannot say if it will work or how that would affect performance.
@@ -743,6 +853,7 @@ if __name__ == "__main__":
print("[FINETUNE] Dataset Generated. Move to Step 2")
return "Dataset Generated. Move to Step 2", train_meta, eval_meta
+
#######################
#### GRADIO STEP 2 ####
#######################
@@ -839,6 +950,7 @@ if __name__ == "__main__":
## STEP 3 - Inference/Testing
### π₯ Important Note
#### π₯ This step will error if you are using TTS version 0.22.0. Please re-run pip install -r requirements_finetune.txt if it errors loading the model.
+ #### π₯ If you dont see multiple speaker reference files and you used more than 3 minutes of speech, try refreshing the page as it may not have loaded them correctly.
### π¦ What you need to do
#### π¦ The model is now trained and you are at the testing stage. Hopefully all the dropdowns should be pre-populated now.
#### π¦ You need to Load Fine-tuned XTTS model and then select your Speaker Reference Audio. You can choose various Speaker Reference Audios to see which works best.
@@ -939,22 +1051,25 @@ if __name__ == "__main__":
f"""
## What to do next
### π¦ What you need to do
- #### π¦ The wav(s) you liked from Speaker Reference Audios are in /finetune/tmp-trn/wavs/ so copy the one(s) you selected to your /alltalk_tts/voices/ samples folder.
- #### π¦ You can delete your original samples from inside of the /finetune/put-voice-samples-in-here/ folder.
- #### π¦ In the folder /finetune/tmp-trn/training/XTTS_FT(date here)/ you will find your best_model.pth, config.json and voice.json (also the last epoch training best_model_xxx.pth)
- #### π¦ You can delete the last epoch training model file best_model_xxx.pth (the one with the numbers).
- #### π¦ You will need to rename best_model.pth to model.pth.
- #### π¦ You will close the Finetuning application and copy your 3 files listed above over the top of the existing model files inside your /alltalk_tts/models/xttsv2_2.0.2/ folder.
- #### π¦ Start up Alltalk and you should be off and running. Preferably use the XTTS Local method for best results.
+ #### π¦ There are 3x buttons below:
+ #### πΉ Compact & move model. This will compress the trained model and move it, along with any large enough wav files created to /models/trainedmodel/. You can use the wav files as sample voices..
+ #### πΉ Delete Generated Training data. This will delete the finetuned model and any other training data generated during this process. You will more than likely want to do this after you have Compacted & moved the model.
+ #### πΉ Delete original voice samples. This will delete the voice samples you placed in put-voice-samples-in-here (if you no longer have a use for them).
+ #### π¦ I will at some point add an option within AllTalk to load a finetuned model as a seperate model. However, for now, you will have to copy it over the model in the /models/xttsv2_2.0.2/ model, if you wish to use it.
+ #### π¦ If you wish to further train this model, you will also need to copy it over the model in the /models/xttsv2_2.0.2/ model folder..
### π¨ Clearing up disk space
- #### π¨ When you are happy with everything, you can delete the /alltalk_tts/finetune/tmp-trn/ folder. This will delete all training data that was generated.
#### π¨ If you are not going to train anything again, you can delete the whisper model from inside of your huggingface cache (3GB approx)
#### Linux ~/.cache/huggingface/hub/(folder-here)
#### Windows C:\users(your-username)\.cache\huggingface\hub\(folder-here).
#### π¨ You can also uninstall the Nvidia CUDA 11.8 Toolkit if you wish and remove your environment variable entries.
"""
)
-
+ final_progress_data = gr.Label(
+ label="Progress:"
+ )
+ compact_btn = gr.Button(value="Compact and move model to '/models/trainedmodel/' (THIS WILL OVER-WRITE ANY MODEL IN THAT LOCATION)")
+ delete_training_btn = gr.Button(value="Delete generated training data")
+ delete_voicesamples_btn = gr.Button(value="Delete original voice samples")
prompt_compute_btn.click(
fn=preprocess_dataset,
@@ -1005,6 +1120,19 @@ if __name__ == "__main__":
],
outputs=[progress_gen, tts_output_audio, reference_audio],
)
+
+ compact_btn.click(
+ fn=compact_model,
+ outputs=[final_progress_data],
+ )
+ delete_training_btn.click(
+ fn=delete_training_data,
+ outputs=[final_progress_data],
+ )
+ delete_voicesamples_btn.click(
+ fn=delete_voice_sample_contents,
+ outputs=[final_progress_data],
+ )
demo.queue().launch(
show_api=False,
diff --git a/finetune/compact.py b/finetune/compact.py
new file mode 100644
index 0000000..71c7ac3
--- /dev/null
+++ b/finetune/compact.py
@@ -0,0 +1,26 @@
+import torch
+import os
+MODEL_PATH = "best_model.pth"
+# Check if best_model.pth exists
+if not os.path.exists(MODEL_PATH):
+ print(f"\nCannot find \033[94m{MODEL_PATH}\033[0m in the current folder.")
+
+ # Check if model.pth exists as an alternative
+ ALTERNATIVE_PATH = "model.pth"
+ if os.path.exists(ALTERNATIVE_PATH):
+ print(f"\nIf your model is currently called \033[93m{ALTERNATIVE_PATH}\033[0m please rename it to \033[94mbest_model.pth\033[0m.")
+ else:
+ print(f"Please put \033[91mbest_model.pth\033[0m in the current folder or rename your model file to \033[94mbest_model.pth\033[0m.")
+ exit(1)
+# Attempt to load the model
+try:
+ checkpoint = torch.load(MODEL_PATH, map_location=torch.device("cpu"))
+except Exception as e:
+ print("Error loading checkpoint:", e)
+ raise
+del checkpoint["optimizer"]
+for key in list(checkpoint["model"].keys()):
+ if "dvae" in key:
+ del checkpoint["model"][key]
+torch.save(checkpoint, "model.pth")
+print(f"\nYour compacted model is created \033[93mmodel.pth\033[0m.")
diff --git a/requirements_nvidia.txt b/requirements_nvidia.txt
index 64bdcfe..82579fd 100644
--- a/requirements_nvidia.txt
+++ b/requirements_nvidia.txt
@@ -10,3 +10,4 @@ requests>=2.31.0
tqdm>=4.66.1
importlib-metadata>=4.8.1
packaging>=23.2
+pydantic>=1.10.13
diff --git a/requirements_other.txt b/requirements_other.txt
index 667d885..aad8a62 100644
--- a/requirements_other.txt
+++ b/requirements_other.txt
@@ -10,3 +10,4 @@ requests>=2.31.0
tqdm>=4.66.1
importlib-metadata>=4.8.1
packaging>=23.2
+pydantic>=1.10.13
diff --git a/script.py b/script.py
index 3090386..30eb714 100644
--- a/script.py
+++ b/script.py
@@ -249,7 +249,7 @@ else:
# Cleanly kill off this script, but allow text-generation-webui to keep running, albeit without this alltalk_tts
sys.exit(1)
-timeout = 60 # Adjust the timeout as needed
+timeout = 120 # Adjust the timeout as needed
# Introduce a delay before starting the check loop
time.sleep(25) # Wait 25 secs before checking if the tts_server.py has started up.
@@ -262,7 +262,7 @@ while time.time() - start_time < timeout:
except requests.RequestException as e:
# Print the exception for debugging purposes
print(
- f"[{params['branding']}Startup] \033[91mWarning\033[0m TTS Subprocess has NOT started up yet, Will keep trying for 60 seconds maximum. Please wait."
+ f"[{params['branding']}Startup] \033[91mWarning\033[0m TTS Subprocess has NOT started up yet, Will keep trying for 120 seconds maximum. Please wait."
)
time.sleep(1)
else:
@@ -483,6 +483,9 @@ def voice_preview(string):
return f"[{params['branding']}Server] Audio generation failed. Status: {generate_response.get('status')}"
+def replace_asterisk(match):
+ return match.group(0) + '"<*'
+
#################################
#### TTS STANDARD GENERATION ####
#################################
@@ -491,6 +494,7 @@ def output_modifier(string, state):
if not params["activate"]:
return string
original_string = string
+ #print("ORGINAL STRING: ", original_string)
cleaned_string = before_audio_generation(string, params)
if cleaned_string is None:
return
@@ -501,32 +505,35 @@ def output_modifier(string, state):
if process_lock.acquire(blocking=False):
try:
if params["narrator_enabled"]:
- # print(original_string)
+ #print(original_string)
# Do Some basic stripping and remove any double CR's
processed_string = (
original_string
.replace("***", "*")
.replace("**", "*")
.replace("\n\n", "\n")
+ .replace("\n", "")
)
- # Clean up two asterisks (narrators) being made between a newline, making it one sentence.
processed_string = re.sub(r'\.\*\n\*', '. ', processed_string)
- # Clean up a few other bits.
- processed_string = (
- processed_string
- .replace("'", "'")
- .replace("\n", " ")
- # Add special characters to the quote so that we can use it to identify things later after its been split
- .replace('"', '"<')
- # Capture new conversations which wont have things like "e in them
- .replace('"', '"<')
- )
- #capturing another outlier in inital character paragraph
- # print("processed string 1 is:", processed_string)
+ processed_string = (processed_string .replace('"\n\n*', '"<*'))
+ processed_string = (processed_string .replace('"\n', '< '))
+ processed_string = (processed_string .replace("'", "'"))
+ processed_string = (processed_string .replace("\n", " "))
+ processed_string = (processed_string .replace('"', '"<'))
+ processed_string = (processed_string .replace('"', '"<'))
+ #pattern = re.compile(r'(?<=[a-zA-Z!])"<(?=[a-zA-Z])')
+ pattern = re.compile(r'(?<=[a-zA-Z!?\.])"<(?=[a-zA-Z])')
+ processed_string = pattern.sub('" ', processed_string)
+ pattern = re.compile(r'(?<=[a-zA-Z])\*(?=[a-zA-Z])')
+ processed_string = pattern.sub('*"', processed_string)
processed_string = processed_string.replace('"<. *', '"< *"')
processed_string = processed_string.replace('< *"', '< *')
processed_string = processed_string.replace('. *', '< *')
- # print("processed string 2 is:", processed_string)
+ processed_string = processed_string.replace('! *', '"< *')
+ processed_string = processed_string.replace('? *', '< *')
+ processed_string = processed_string.replace('* ', '* "<')
+ processed_string = processed_string.replace('*. ', '* "<')
+ processed_string = re.sub(r'\*(?=[a-zA-Z])', replace_asterisk, processed_string)
# Set up a tracking of the individual wav files.
audio_files_all_paragraphs = []
# Split the line using " and ".* " (so end of sentences, leaving special characters added to the start of all OTHER sentences, bar possibly the first one if its starting with a *
@@ -534,18 +541,18 @@ def output_modifier(string, state):
audio_files_paragraph = []
for i, part in enumerate(parts):
# Skip parts that are too short
- if len(part.strip()) <= 1:
+ if len(part.strip()) <= 3:
continue
# Figure out which type of line it is, then replace characters as necessary to avoid TTS trying to pronunce them, htmlunescape after.
# Character will always be a < with a letter immediately after it
if '<' in part and '< *' not in part and '<*' not in part and '< *' not in part and '< ' not in part and '< ' not in part:
cleaned_part = html.unescape(part.replace('<', ''))
voice_to_use = params["voice"]
- #Narrator will always be an * or < with an * a position or two after it.
+ #Narrator will always be an * or < with an * a position or two after it.
elif '<*' in part or '< *' in part or '< *' in part or '*' in part:
cleaned_part = html.unescape(part.replace('<*', '').replace('< *', '').replace('< *', '').replace('*', ''))
voice_to_use = params["narrator_voice"]
- #If the other two dont capture it, aka, the AI gave no * or " on the line, use non_quoted_text_is aka user interface, user can choose Char or Narrator
+ #If the other two dont capture it, aka, the AI gave no * or " on the line, use non_quoted_text_is aka user interface, user can choose Char or Narrator
elif non_quoted_text_is:
cleaned_part = html.unescape(part.replace('< ', '').replace('< ', '').replace('< ', ''))
voice_to_use = params["voice"]
@@ -576,9 +583,14 @@ def output_modifier(string, state):
)
cleaned_part = html.unescape(processed_string)
# Process the part and give it a non-character name if being used vai API or standalone.
- output_filename = get_output_filename(state)
+ if "character_menu" in state:
+ output_file = Path(f'{params["output_folder_wav"]}/{state["character_menu"]}_{int(time.time())}.wav')
+ else:
+ output_file = Path(f'{params["output_folder_wav"]}/TTSOUT_{int(time.time())}.wav')
+ output_file_str = output_file.as_posix()
+ output_file = get_output_filename(state)
generate_response = send_generate_request(
- cleaned_part, params["voice"], language_code, output_filename
+ cleaned_part, params["voice"], language_code, output_file_str
)
audio_path = generate_response.get("data", {}).get("audio_path")
final_output_file = audio_path
@@ -618,11 +630,10 @@ def output_modifier(string, state):
def get_output_filename(state):
- # Check if character name exists and if not, just call it TTSOUT_
if "character_menu" in state:
- return Path(f'{params["output_folder_wav"]}/{state["character_menu"]}_{int(time.time())}.wav').as_posix()
+ return Path(f'{params["output_folder_wav"]}/{state["character_menu"]}_{str(uuid.uuid4())[:8]}.wav').as_posix()
else:
- return Path(f'{params["output_folder_wav"]}/TTSOUT_{int(time.time())}.wav').as_posix()
+ return Path(f'{params["output_folder_wav"]}/TTSOUT_{str(uuid.uuid4())[:8]}.wav').as_posix()
###############################################
@@ -782,7 +793,7 @@ def ui():
)
tts_radio_buttons_play = gr.HTML(visible=False)
explanation_text = gr.HTML(
- f"NOTE: Switching Model Type, Low VRAM & DeepSpeed, each takes 15 seconds. Each TTS generation method has a slightly different sound. DeepSpeed checkbox is only visible if DeepSpeed is present. Readme & Settings: http://{params['ip_address']}:{params['port_number']}"
+ f"
βοΈ Click here for the Settings page, help & instructionsβοΈ
NOTE: Switching Model Type, Low VRAM & DeepSpeed, each takes 15 seconds. Each TTS generation method has a slightly different sound. DeepSpeed checkbox is only visible if DeepSpeed is installed on your system."
)
with gr.Row():
@@ -881,4 +892,4 @@ if running_in_standalone:
try:
time.sleep(1) # Add a small delay to avoid high CPU usage
except KeyboardInterrupt:
- break # Allow graceful exit on Ctrl+C
\ No newline at end of file
+ break # Allow graceful exit on Ctrl+C
diff --git a/tts_server.py b/tts_server.py
index 938bbc6..61fb07b 100644
--- a/tts_server.py
+++ b/tts_server.py
@@ -22,7 +22,6 @@ from fastapi.templating import Jinja2Templates
from fastapi.staticfiles import StaticFiles
from jinja2 import Template
from contextlib import asynccontextmanager
-from pydantic import field_validator
###########################
#### STARTUP VARIABLES ####
@@ -625,6 +624,7 @@ async def get_audio(filename: str):
########################
import html
import re
+import uuid
import numpy as np
import soundfile as sf
from typing import Union, Dict
@@ -647,11 +647,10 @@ class JSONInput(BaseModel):
autoplay: bool = Field(..., description="autoplay needs to be a true or false value.")
autoplay_volume: float = Field(..., ge=0.1, le=1.0, description="autoplay_volume needs to be from 0.1 to 1.0")
- @field_validator("autoplay_volume")
@classmethod
def validate_autoplay_volume(cls, value):
if not (0.1 <= value <= 1.0):
- raise ValidationError("Autoplay volume must be between 0.1 and 1.0")
+ raise ValueError("Autoplay volume must be between 0.1 and 1.0")
return value
@@ -804,7 +803,7 @@ async def tts_generate(
print("ELSE - NARRATOR\n")
cleaned_part = html.unescape(part.replace('< ', '').replace('< ', '').replace('< ', ''))
voice_to_use = narrator_voice_gen
- output_file = this_dir / "outputs" / f"{output_file_name}_{int(time.time())}_{i}.wav"
+ output_file = this_dir / "outputs" / f"{output_file_name}_{uuid.uuid4()}_{int(time.time())}_{i}.wav"
output_file_str = output_file.as_posix()
await generate_audio(cleaned_part, voice_to_use, language, output_file_str)
audio_path = output_file_str
@@ -1054,6 +1053,12 @@ simple_webpage = """
AllTalk is a web interface, based around the Coqui TTS voice cloning/speech generation system. To generate TTS, you can use the provided gradio interface or interact with the server using JSON/CURL commands.
Note: When loading up a new character in Text generation webUI it may look like nothing is happening for 20-30 seconds. Its actually processing the introduction section (greeting message) of the text and once that is completed, it will appear. You can see the activity occuring in the console window. Refreshing the page multiple times will try force the TTS engine to keep re-generating the text, so please just wait and check the console if needed.
Note: Ensure that your RP character card has asterisks around anything for the narration and double quotes around anything spoken. There is a complication ONLY with the greeting card so, ensuring it has the correct use of quotes and asterisks will help make sure that the greeting card sounds correct. I will aim to address this issue in a future update. In Text-generation-webUI parameters menu > character tab > greeting make sure that anything in there that is the narrator is in asterisks and anything spoken is in double quotes, then hit the save (πΎ) button.
+AllTalk Minor Bug Fixes Changelog & Known issues
+If I squash any minor bugs or find any issues, I will try to apply an update asap. If you think something isnt working correctly or you have a problem, check these two links below first.
+
AllTalk
Github link here
Update instructions link here
Help and issues link here
Text generation webUI
Web interface link here
Documentation link here
Back to top of page
@@ -1153,7 +1158,7 @@ simple_webpage = """
Save your generated wav file in the /alltalk_tts/voices/ folder.
Its worth mentioning that using AI generated audio clips may introduce unwanted sounds as its already a copy/simulation of a voice.
Why doesnt it sound like XXX Person?
-The reasons can be that you:
+Maybe you might be interested in trying Finetuning of the model. Otherwise, the reasons can be that you:
- Didn't downsample it as above.
- Have a bad quality voice sample.