Merge pull request #567 from Paladinium/main_improve_github_actions

Improve paths-filter for Docker github action

.github/ISSUE_TEMPLATE/bug-report-support-request.md

@@ -1,6 +1,6 @@
 ---
-name: Support request OR Bug report
+name: Support Request/Issue
-about: Create a report to help us improve
+about: Use this to create a support request or report a bug/issue
 title: ''
 labels: ''
 assignees: ''

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1 @@
+blank_issues_enabled: false

.github/ISSUE_TEMPLATE/feature_request.md

@@ -1,12 +1,15 @@
 ---
 name: Feature request
-about: Suggest an idea for this project
+about: Use this to create a Feature request
 title: ''
 labels: ''
 assignees: ''
 
 ---
 
+**Please confirm you have read the current Features request list**
+https://github.com/erew123/alltalk_tts/discussions/74
+
 **Is your feature request related to a problem? Please describe.**
 A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
 

.github/actions/docker-setup/action.yml

@@ -0,0 +1,73 @@
+name: 'Setting up docker'
+
+inputs:
+  DOCKERHUB_USERNAME:
+    description: "Docker hub username"
+    required: true
+  DOCKERHUB_REPO_NAME:
+    description: "Docker hub repo name"
+    required: true
+  DOCKERHUB_TOKEN:
+    description: "Docker hub token"
+    required: true
+
+runs:
+  using: "composite"
+  steps:
+    - name: Free Disk Space (Ubuntu)
+      uses: jlumbroso/free-disk-space@v1.3.1
+      with:
+        tool-cache: true
+        swap-storage: false
+        large-packages: false
+    - name: Read variables
+      shell: bash
+      run: |
+        source ./docker/variables.sh
+        for key in "${!ALLTALK_VARS[@]}"
+        do
+          echo "${key}=${ALLTALK_VARS[${key}]}"
+          echo "${key}=${ALLTALK_VARS[${key}]}" >> $GITHUB_ENV
+        done
+
+    - name: Extract version from Git Tag
+      shell: bash {0}
+      run: |
+        # Use the git tag if existing, otherwise 'latest':
+        DOCKER_TAG=$( echo "${{ github.ref }}" | grep -q tags | cut -d'/' -f 3 )
+        if [ -z "${DOCKER_TAG}" ]; then
+          DOCKER_TAG="latest"
+        fi
+        echo "Using DOCKER_TAG=${DOCKER_TAG}"
+        echo "DOCKER_TAG=${DOCKER_TAG}" >> $GITHUB_ENV
+
+    - name: Extract Docker Hub User and Repo Name
+      shell: bash
+      env:
+        DOCKERHUB_USERNAME: ${{ inputs.DOCKERHUB_USERNAME }}
+        DOCKERHUB_REPO_NAME: ${{ inputs.DOCKERHUB_REPO_NAME }}
+      run: |
+        if [ -n "${DOCKERHUB_USERNAME}" ]; then
+          echo "Using Github repository variable 'DOCKERHUB_USERNAME' as Docker hub username with value: '${DOCKERHUB_USERNAME}'"
+          echo "DOCKERHUB_USERNAME=${DOCKERHUB_USERNAME}" >> $GITHUB_ENV
+        else
+          echo "Using Github repository name as Docker hub username"
+          echo "DOCKERHUB_USERNAME=$(echo ${{ github.repository }} | cut -d'/' -f 1)" >> $GITHUB_ENV
+        fi
+
+        if [ -n "${DOCKERHUB_REPO_NAME}" ]; then
+          echo "Using Github repository variable 'DOCKERHUB_REPO_NAME' as Docker hub repo name with value: '${DOCKERHUB_REPO_NAME}'"
+          echo "DOCKERHUB_REPO_NAME=${DOCKERHUB_REPO_NAME}" >> $GITHUB_ENV
+        else
+          echo "Using Github repository name as Docker hub repo name"
+          echo "DOCKERHUB_REPO_NAME=$(echo ${{ github.repository }} | cut -d'/' -f 2)" >> $GITHUB_ENV
+        fi
+
+    - name: Set up Docker Buildx
+      uses: docker/setup-buildx-action@v3
+
+    - name: Login to DockerHub
+      uses: docker/login-action@v3
+      with:
+        username: ${{ inputs.DOCKERHUB_USERNAME }}
+        password: ${{ inputs.DOCKERHUB_TOKEN }}

.github/workflows/publish-docker-v2.yml

@@ -0,0 +1,124 @@
+name: Publish Docker v2
+
+on:
+  workflow_dispatch:
+  release:
+    types: [ published ]
+  push:
+    branches:
+      - 'alltalkbeta'
+    tags:
+      - 'v*'
+
+# Limit the concurrency of entire workflow runs for a specific branch:
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build_base_environment_docker_image:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        uses: ./.github/actions/docker-setup
+        with:
+          DOCKERHUB_USERNAME: ${{ vars.DOCKERHUB_USERNAME }}
+          DOCKERHUB_REPO_NAME: ${{ vars.DOCKERHUB_REPO_NAME }}
+          DOCKERHUB_TOKEN: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            ${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}_environment
+          tags: |
+            # Use sematic versioning on push tag event:
+            type=semver,pattern={{version}}
+            # set latest tag for main and alltalkbeta branch:
+            type=raw,value=latest,enable=${{ github.ref == format('refs/heads/{0}', 'main') || github.ref == format('refs/heads/{0}', 'alltalkbeta') }}
+
+      - uses: dorny/paths-filter@v3
+        id: changes
+        with:
+          base: 'alltalkbeta'
+          filters: |
+            docker-base:
+              - 'docker/variables.sh'
+              - 'docker/base/Dockerfile'
+            docker-workflow:
+              - '.github/workflows/publish-docker*.yml'
+
+      - name: Build and push base Docker image
+        if: |
+          steps.changes.outputs.docker-base == 'true' ||
+          steps.changes.outputs.docker-workflow == 'true' ||
+          github.event_name == 'release' ||
+          (github.event_name == 'push' && contains(github.ref, 'refs/tags/'))
+        uses: docker/build-push-action@v6
+        with:
+          context: ./docker/base
+          file: ./docker/base/Dockerfile
+          platforms: "linux/amd64"
+          push: true
+          build-args: |
+            CUDA_VERSION=${{ env.CUDA_VERSION }}
+            PYTHON_VERSION=${{ env.PYTHON_VERSION }}
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=registry,ref=${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}_environment:buildcache
+          cache-to: type=registry,ref=${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}_environment:buildcache,mode=max
+
+  build_main_docker_image:
+    needs: build_base_environment_docker_image
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        tts: [ xtts, vits, piper ]
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Setup docker
+        uses: ./.github/actions/docker-setup
+        with:
+          DOCKERHUB_USERNAME: ${{ vars.DOCKERHUB_USERNAME }}
+          DOCKERHUB_REPO_NAME: ${{ vars.DOCKERHUB_REPO_NAME }}
+          DOCKERHUB_TOKEN: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            ${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}
+          flavor: |
+            suffix=-${{ matrix.tts }},onlatest=true
+          tags: |
+            # Use sematic versioning on push tag event:
+            type=semver,pattern={{version}}
+            # set latest tag for main and alltalkbeta branch:
+            type=raw,value=latest,enable=${{ github.ref == format('refs/heads/{0}', 'main') || github.ref == format('refs/heads/{0}', 'alltalkbeta') }}
+
+      - name: Build and push base Docker image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: "linux/amd64"
+          push: true
+          build-args: |
+            TTS_MODEL=${{ matrix.tts }}
+            ALLTALK_DIR=${{ env.ALLTALK_DIR }}
+            DEEPSPEED_VERSION=${{ env.DEEPSPEED_VERSION }}
+            DOCKER_TAG=${{ env.DOCKER_TAG }}
+            DOCKER_REPOSITORY=${{ env.DOCKERHUB_USERNAME }}/
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=registry,ref=${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}:buildcache-${{ matrix.tts }}
+          cache-to: type=registry,ref=${{ env.DOCKERHUB_USERNAME }}/${{ env.DOCKERHUB_REPO_NAME }}:buildcache-${{ matrix.tts }},mode=max

.github/workflows/publish-docker.yml

@@ -1,8 +1,8 @@
-name: Publish Docker image
+name: Publish Docker image (main branch)
 
 on:
-  push:
+#  push:
-    branches: [main]
+#    branches: [main]
   release:
     types: [published]
   workflow_dispatch:

.github/workflows/python-ci.yml

@@ -0,0 +1,53 @@
+# This workflow will install Python dependencies, run tests and lint with a single version of Python
+# For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python
+
+name: Python CI
+
+on: [push, pull_request]
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python 3.11.9
+        uses: actions/setup-python@v3
+        if: ${{ hashFiles('./test') != '' }}
+        with:
+          python-version: "3.11.9"
+      - name: Install dependencies
+        if: ${{ hashFiles('./test') != '' }}
+        run: |
+          python -m pip install --upgrade pip
+          if [ -f system/requirements/requirements_unit_test.txt ]; then
+            pip install -r system/requirements/requirements_unit_test.txt;
+          fi
+      - name: Test with pytest
+        if: ${{ hashFiles('./test') != '' }}
+        run: |
+          pytest test --doctest-modules --junitxml=junit/test-results.xml --cov=com --cov-report=xml --cov-report=html
+      - name: Surface failing tests
+        if: always()
+        uses: pmeier/pytest-results-action@v0.6.0
+        with:
+          # A list of JUnit XML files, directories containing the former, and wildcard
+          # patterns to process.
+          # See @actions/glob for supported patterns.
+          path: junit/test-results.xml
+
+          # (Optional) Add a summary of the results at the top of the report
+          summary: true
+
+          # (Optional) Select which results should be included in the report.
+          # Follows the same syntax as `pytest -r`
+          display-options: fEX
+
+          # (Optional) Fail the workflow if no JUnit XML was found.
+          fail-on-empty: false
+
+          # (Optional) Title of the test results section in the workflow summary
+          title: Test results

Dockerfile

@@ -1,12 +1,20 @@
-FROM ubuntu:22.04
+FROM nvidia/cuda:12.1.1-devel-ubuntu22.04
 ENV HOST 0.0.0.0
 RUN apt-get update && apt-get upgrade -y \
     && apt-get install -y git build-essential portaudio19-dev \
     python3 python3-pip gcc wget \
+    ocl-icd-opencl-dev opencl-headers clinfo \
+    libclblast-dev libopenblas-dev libaio-dev \
+    && mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd \
     && ln -s /usr/bin/python3 /usr/bin/python
+
 WORKDIR /app
+ENV CUDA_DOCKER_ARCH=all
+COPY system/requirements/requirements_docker.txt system/requirements/requirements_docker.txt
+RUN pip install --no-cache-dir --no-deps -r system/requirements/requirements_docker.txt && \
+    pip install --no-cache-dir deepspeed
 COPY . .
-RUN pip install --no-cache-dir -r requirements_other.txt
+
-EXPOSE 7851
+EXPOSE 7851 7852
 RUN chmod +x launch.sh
 ENTRYPOINT ["sh", "-c", "./launch.sh"]

atsetup.bat

@@ -1,7 +1,7 @@
 @echo off
 cd /D "%~dp0"
-
+setlocal enabledelayedexpansion
-set PATH=%PATH%;%SystemRoot%\system32
+@set "PATH=%PATH%;%SystemRoot%\system32"
 
 :: Generate the ESC character
 for /F %%a in ('echo prompt $E ^| cmd') do set "ESC=%%a"
@@ -25,6 +25,15 @@ set "L_CYAN=%ESC%[96m"
 set "L_WHITE=%ESC%[97m"
 set "RESET=%ESC%[0m"
 
+echo "%CD%"| findstr /C:" " >nul &&  echo. && echo    %L_BLUE%ALLTALK WINDOWS SETUP UTILITY%RESET% && echo. &&  echo. &&  echo    You are trying to install AllTalk in a folder that has a space in the folder path e.g. && echo. && echo       C:\%L_RED%program files%RESET%\alltalk_tts && echo.	&& echo    This causes errors with Conda and Python scripts. Please follow this link for reference: &&  echo. && echo      %L_CYAN%https://docs.anaconda.com/free/working-with-conda/reference/faq/#installing-anaconda%RESET% && echo. && echo    Please use a folder path that has no spaces in it e.g. && echo. && echo       C:\myfiles\alltalk_tts\ && echo. && echo. && pause && goto :end
+
+@rem Check for special characters in installation path
+set "SPCHARMESSAGE="WARNING: Special characters were detected in the installation path!" "         This can cause the installation to fail!""
+echo "%CD%"| findstr /R /C:"[!#\$%&()\*+,;<=>?@\[\]\^`{|}~]" >nul && (
+	call :PrintBigMessage %SPCHARMESSAGE%
+)
+set SPCHARMESSAGE=
+
 @rem Check if curl is available
 curl --version >nul 2>&1
 if "%ERRORLEVEL%" NEQ "0" (
@@ -120,7 +129,7 @@ if "%StandaloneOption%"=="3" goto STReapplyrequirements
 if "%StandaloneOption%"=="4" goto STDeleteCustomStandalone
 if "%StandaloneOption%"=="5" goto STPurgepipcache
 if "%StandaloneOption%"=="8" goto GenerateDiagsStandalone
-if "%StandaloneOption%"=="9" goto End
+if "%StandaloneOption%"=="9" goto EndStandalone
 goto StandaloneMenu
 
 :InstallNvidiaTextGen
@@ -128,6 +137,8 @@ pip install -r system\requirements\requirements_textgen.txt
 if %ERRORLEVEL% neq 0 (
     echo.
     echo    There was an error installing the requirements.
+    echo    Have you started your Text-gen-webui Python environment
+    echo    with cmd_{yourOS} before running atsetup.bat?
     echo    Press any key to return to the menu.
     echo.
     pause
@@ -383,7 +394,7 @@ echo     Downloading and installing PyTorch. This step can take a long time
 echo     depending on your internet connection and hard drive speed. Please
 echo     be patient.
 echo.
-pip install torch>=2.2.1+cu121 torchaudio>=2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
+pip install torch==2.2.2+cu121 torchaudio>=2.2.2+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
 echo Installing other requirements.
 echo.
 pip install -r system\requirements\requirements_standalone.txt
@@ -524,11 +535,11 @@ echo     Downloading and installing PyTorch. This step can take a long time
 echo     depending on your internet connection and hard drive speed. Please
 echo     be patient.
 echo.
-pip install torch>=2.2.1+cu121 torchaudio>=2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
+pip install torch==2.2.2+cu121 torchaudio>=2.2.2+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
 echo Installing other requirements.
 echo.
 pip install -r system\requirements\requirements_standalone.txt
-curl -LO https://github.com/erew123/alltalk_tts/releases/download/DeepSpeed-14.0/deepspeed-0.14.0+ce78a63-cp311-cp311-win_amd64.whl
+C:/Windows/system32/curl.exe -LO https://github.com/erew123/alltalk_tts/releases/download/DeepSpeed-14.0/deepspeed-0.14.0+ce78a63-cp311-cp311-win_amd64.whl
 echo Installing DeepSpeed...
 pip install deepspeed-0.14.0+ce78a63-cp311-cp311-win_amd64.whl
 del deepspeed-0.14.0+ce78a63-cp311-cp311-win_amd64.whl
@@ -633,6 +644,16 @@ Echo.
 pause
 goto StandaloneMenu
 
+:EndStandalone
+echo Exiting AllTalk Setup Utility...
+echo.
+Echo    Remember, after installation you can....
+Echo    Run %L_YELLOW%start_alltalk.bat%RESET% to start AllTalk.
+Echo    Run %L_YELLOW%start_finetune.bat%RESET% to start Finetuning.
+Echo    Run %L_YELLOW%start_environment.bat%RESET% to start the AllTalk Python environment.
+Echo.
+exit /b
+
 :End
 echo Exiting AllTalk Setup Utility...
 exit /b

atsetup.sh

@@ -28,6 +28,40 @@ check_curl() {
     fi
 }
 
+# Check if the current directory path contains a space
+containsSpace=false
+currentPath=$(pwd)
+if echo "$currentPath" | grep -q ' '; then
+    containsSpace=true
+fi
+
+if [ "$containsSpace" = true ]; then
+    echo
+    echo -e "    ${L_BLUE}ALLTALK LINUX SETUP UTILITY${NC}"
+    echo
+    echo
+    echo -e "    You are trying to install AllTalk in a folder that has a space in the"
+    echo -e "    folder path e.g."
+    echo 
+    echo -e "       /home/${L_RED}program files${NC}/alltalk_tts"
+    echo 
+    echo -e "    This causes errors with Conda and Python scripts. Please follow this"
+    echo -e "    link for reference:"
+    echo 
+    echo -e "      ${L_CYAN}https://docs.anaconda.com/free/working-with-conda/reference/faq/#installing-anaconda${NC}"
+    echo 
+    echo -e "    Please use a folder path that has no spaces in it e.g." 
+    echo 
+    echo -e "       /home/myfiles/alltalk_tts/"
+    echo 
+    echo
+    read -p "Press Enter to continue..." 
+    exit 1
+else
+    # Continue with the main menu
+    echo "Continue with the main menu."
+fi
+
 # Main Menu
 main_menu() {
     while true; do
@@ -291,7 +325,7 @@ install_custom_standalone() {
     echo "    Downloading and installing PyTorch. This step can take a long time"
     echo "    depending on your internet connection and hard drive speed. Please"
     echo "    be patient."
-    pip install torch>=2.2.1+cu121 torchaudio>=2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu1
+    pip install torch==2.2.1+cu121 torchaudio==2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
     echo
     echo "    Installing additional requirements."
     echo
@@ -469,7 +503,7 @@ reapply_standalone() {
     echo "    Downloading and installing PyTorch. This step can take a long time"
     echo "    depending on your internet connection and hard drive speed. Please"
     echo "    be patient."
-    pip install torch>=2.2.1+cu121 torchaudio>=2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu1
+    pip install torch>=2.2.1+cu121 torchaudio>=2.2.1+cu121 --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121
     echo
     echo "    Installing additional requirements."
     echo

docker-compose.yml

@@ -1,13 +1,27 @@
 version: '3.8'
 services:
   alltalk-tts:
-    image: erew123/alltalk_tts:cpu
+    image: ${IMAGE:-alltalk:v1-9c}
     restart: unless-stopped
     ports:
       - "7851:7851"
+      - "7852:7852"
     volumes:
       - ./outputs:/app/outputs/
       - ./models:/app/models/
       - ./voices:/app/voices/
       - ./finetune/put-voice-samples-in-here:/app/finetune/put-voice-samples-in-here
       - ./dockerconfig.json:/app/confignew.json
+    healthcheck:
+      test: ["CMD", "wget", "-qO-", "http://localhost:7851/api/ready"]
+      interval: 1m30s
+      timeout: 30s
+      retries: 5
+      start_period: 30s
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [ gpu ]

finetune.py

@@ -291,12 +291,12 @@ def create_temporary_file(folder, suffix=".wav"):
     unique_filename = f"custom_tempfile_{int(time.time())}_{random.randint(1, 1000)}{suffix}"
     return os.path.join(folder, unique_filename)
 
-def format_audio_list(target_language, whisper_model, out_path, eval_split_number, gradio_progress=progress):
+def format_audio_list(target_language, whisper_model, out_path, eval_split_number, speaker_name_input, gradio_progress=progress):
     pfc_check_fail()
     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 = eval_split_number / 100.0
-    speaker_name="coqui"
+    speaker_name=speaker_name_input
     audio_total_size = 0
     os.makedirs(out_path, exist_ok=True)
     temp_folder = os.path.join(out_path, "temp")  # Update with your folder name
@@ -533,7 +533,7 @@ def basemodel_or_finetunedmodel_choice(value):
     elif value == "Existing finetuned model":
         basemodel_or_finetunedmodel = False
 
-def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path, max_audio_length=255995):
+def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, learning_rate, output_path, max_audio_length=255995):
     pfc_check_fail()
     #  Logging parameters
     RUN_NAME = "XTTS_FT"
@@ -657,7 +657,7 @@ def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv,
         optimizer="AdamW",
         optimizer_wd_only_on_weights=OPTIMIZER_WD_ONLY_ON_WEIGHTS,
         optimizer_params={"betas": [0.9, 0.96], "eps": 1e-8, "weight_decay": 1e-2},
-        lr=5e-06,  # learning rate
+        lr=learning_rate,  # learning rate
         lr_scheduler="MultiStepLR",
         # it was adjusted accordly for the new step scheme
         lr_scheduler_params={"milestones": [50000 * 18, 150000 * 18, 300000 * 18], "gamma": 0.5, "last_epoch": -1},
@@ -735,6 +735,9 @@ def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
     config.load_json(xtts_config)
     XTTS_MODEL = Xtts.init_from_config(config)
     print("[FINETUNE] \033[94mStarting Step 3\033[0m Loading XTTS model!")
+    print(xtts_checkpoint)
+    print(xtts_vocab)
+    print(xtts_speakers_pth)
     XTTS_MODEL.load_checkpoint(config, checkpoint_path=xtts_checkpoint, vocab_path=xtts_vocab, use_deepspeed=False, speaker_file_path=xtts_speakers_pth)
     if torch.cuda.is_available():
         XTTS_MODEL.cuda()
@@ -773,18 +776,22 @@ def get_available_voices(minimum_size_kb=1200):
     ]
     return sorted([str(file) for file in voice_files])  # Return full path as string
 
+def find_best_models(directory):
+    """Find files named 'best_model.pth' in the given directory."""
+    return [str(file) for file in Path(directory).rglob("best_model.pth")]
+
 def find_models(directory, extension):
     """Find files with a specific extension in the given directory."""
-    return [file for file in Path(directory).rglob(f"*.{extension}")]
+    return [str(file) for file in Path(directory).rglob(f"*.{extension}")]
 
 def find_jsons(directory, filename):
     """Find files with a specific filename in the given directory."""
-    return list(Path(directory).rglob(filename))
+    return [str(file) for file in Path(directory).rglob(filename)]
 
 # Your main directory
 main_directory = Path(this_dir) / "finetune" / "tmp-trn"
-# XTTS checkpoint files (*.pth)
+# XTTS checkpoint files (best_model.pth)
-xtts_checkpoint_files = find_models(main_directory, "pth")
+xtts_checkpoint_files = find_best_models(main_directory)
 # XTTS config files (config.json)
 xtts_config_files = find_jsons(main_directory, "config.json")
 # XTTS vocab files (vocab.json)
@@ -800,68 +807,70 @@ def find_latest_best_model(folder_path):
     latest_file = max(files, key=os.path.getctime, default=None)
     return latest_file
 
-def compact_model():
+def compact_model(xtts_checkpoint_copy):
     this_dir = Path(__file__).parent.resolve()
-
+    print("THIS DIR:", this_dir)
-    best_model_path_str = find_latest_best_model(this_dir / "finetune" / "tmp-trn" / "training")
+    best_model_path_str = str(xtts_checkpoint_copy)  # Convert to string
-
+    print("best_model_path_str", best_model_path_str)
+    
     # Check if the best model file exists
-    if best_model_path_str is None:
+    if not best_model_path_str:
         print("[FINETUNE] No trained model was found.")
         return "No trained model was found."
-
+    
-    # Convert model_path_str to Path
+    print(f"[FINETUNE] Best model path: {best_model_path_str}")
-    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"))
+        checkpoint = torch.load(best_model_path_str, map_location=torch.device("cpu"))
+        print(f"[FINETUNE] Checkpoint loaded: {best_model_path_str}")
     except Exception as e:
         print("[FINETUNE] Error loading checkpoint:", e)
         raise
+    
+    # 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)
 
     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")
+    torch.save(checkpoint, str(target_dir / "model.pth"))  # Convert to string
 
     # Specify the files you want to copy
-    files_to_copy = ["vocab.json", "config.json", "dvae.pth", "mel_stats.pth","speakers_xtts.pth", ]
+    files_to_copy = ["vocab.json", "config.json", "speakers_xtts.pth", "mel_stats.pth", "dvae.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))
+        shutil.copy(str(src_path), str(dest_path))  # Convert to string
-
+    
     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))
+            shutil.copy(str(file_path), str(target_wavs_dir / file_path.name))  # Convert to string
     
     print("[FINETUNE] Model copied to '/models/trainedmodel/'")
     return "Model copied to '/models/trainedmodel/'"
 
 
-def compact_lastfinetuned_model():
+def compact_lastfinetuned_model(xtts_checkpoint_copy):
     this_dir = Path(__file__).parent.resolve()
 
-    best_model_path_str = find_latest_best_model(this_dir / "finetune" / "tmp-trn" / "training")
+    best_model_path_str = xtts_checkpoint_copy
+    print(f"[FINETUNE] Best model path: {best_model_path_str}")
 
     # Check if the best model file exists
     if best_model_path_str is None:
@@ -880,21 +889,21 @@ def compact_lastfinetuned_model():
 
     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" / "lastfinetuned"
 
     # Create the target directory if it doesn't exist
     target_dir.mkdir(parents=True, exist_ok=True)
 
+    for key in list(checkpoint["model"].keys()):
+        if "dvae" in key:
+            del checkpoint["model"][key]
+
     # 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", "speakers_xtts.pth",]
+    files_to_copy = ["vocab.json", "config.json", "speakers_xtts.pth", "mel_stats.pth", "dvae.pth",]
 
     for file_name in files_to_copy:
         src_path = this_dir / base_path / base_model_path / file_name
@@ -916,15 +925,16 @@ def compact_lastfinetuned_model():
     return "Model copied to '/models/lastfinetuned/'"
 
 
-def compact_legacy_model():
+def compact_custom_model(xtts_checkpoint_copy, folder_path):
     this_dir = Path(__file__).parent.resolve()
 
-    best_model_path_str = os.path.join(this_dir, "finetune", "best_model.pth")
+    best_model_path_str = xtts_checkpoint_copy
+    print(f"[FINETUNE] Best model path: {best_model_path_str}")
 
     # Check if the best model file exists
-    if not os.path.exists(best_model_path_str):
+    if best_model_path_str is None:
-        print("[FINETUNE] No model called best_model.pth was found in /finetune/")
+        print("[FINETUNE] No trained model was found.")
-        return "No model called best_model.pth was found in /finetune/"
+        return "No trained model was found."
 
     # Convert model_path_str to Path
     best_model_path = Path(best_model_path_str)
@@ -938,21 +948,41 @@ def compact_legacy_model():
 
     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 / "finetune"
+    target_dir = this_dir / "models" / folder_path
 
     # Create the target directory if it doesn't exist
     target_dir.mkdir(parents=True, exist_ok=True)
 
+    for key in list(checkpoint["model"].keys()):
+        if "dvae" in key:
+            del checkpoint["model"][key]
+
     # Save the modified checkpoint in the target directory
     torch.save(checkpoint, target_dir / "model.pth")
 
-    print("[FINETUNE] model.pth created in '/finetune/'")
+    # Specify the files you want to copy
-    return "model.pth created in '/finetune/'"
+    files_to_copy = ["vocab.json", "config.json", "speakers_xtts.pth", "mel_stats.pth", "dvae.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/",folder_path,"/")
+    return f"Model copied to '/models/{folder_path}/'"
+
 
 def delete_training_data():
     # Define the folder to be deleted
@@ -1264,6 +1294,9 @@ if __name__ == "__main__":
                 ### 🟥 <u>Important Note - Windows - "UserWarning: huggingface_hub cache-system uses symlinks."</u>
                 ◽ This error is caused by Huggingfaces download software downloading the Whisper model. If you get this error, please restart your Windows command prompt with "Run as Administrator" and restart finetuning.<br>
                 ◽ This should only occur the 1st time it downloads the Whisper model. [Huggingface Reference here](https://huggingface.co/docs/huggingface_hub/en/guides/manage-cache#limitations)
+                ### 🟥 <u>Important Note - Language support.</u>
+                ◽ Although I have done my best to help automate this step, the Whisper model is not great with all languages and may fail to build your training data correctly. The Large-v3 may be better at certain languages.<br>
+                ◽ You can find information about the Whisper model [here](https://github.com/openai/whisper?tab=readme-ov-file#available-models-and-languages) and you can find data about manually building training data [here](https://docs.coqui.ai/en/latest/formatting_your_dataset.html), as well as details below about the file structure this step performs.<br>
                 ### 🟦 <u>What you need to do</u>
                 ◽ Please read Coqui's guide on what makes a good dataset [here](https://docs.coqui.ai/en/latest/what_makes_a_good_dataset.html#what-makes-a-good-dataset)<br>
                 ◽ Place your audio files in <span style="color: #3366ff;">{str(audio_folder)}</span>          
@@ -1276,6 +1309,7 @@ if __name__ == "__main__":
                 ◽ Whisper is making a best effort to find spoken audio and break it down into smaller audio files. The content of these audio files is then transcribed into CSV fles (which you can edit in Excel or similar).<br>
                 ◽ These files (audio and CSV) are used at the next step to train the model "this is what the audio sounds like and these are the words being spoken".<br>
                 ◽ If you wish to manually look at the CSV files before running the next step, you are welcome to do so and edit them as necessary. The greater the accuracy of the text the better the training will be.<br>
+                &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; ◽ <span style="color: #3366ff;">/alltalk_tts/finetune/tmp-trn/lang.txt</span> Contains a two digit langauage code e.g. `en`, `fr` etc.<br>
                 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; ◽ <span style="color: #3366ff;">/alltalk_tts/finetune/tmp-trn/metadata_train.csv</span> Contains the list of wavs to text that will be used. This is what the model is trained with.<br>
                 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; ◽ <span style="color: #3366ff;">/alltalk_tts/finetune/tmp-trn/metadata_eval.csv</span> Lists the evaluation wav to text, used to generate TTS while training and evaluate the quality of generated TTS to the original sample.<br>
                 &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; ◽ <span style="color: #3366ff;">/alltalk_tts/finetune/tmp-trn/wavs/</span> These are the wavs split out from your original training samples and match the contents of the CSV files.<br>
@@ -1334,6 +1368,11 @@ if __name__ == "__main__":
                     maximum=95,  # Maximum value
                     step=1,  # Increment step
                 )
+                speaker_name_input = gr.Textbox(
+                label="The name of the speaker/person you are training",
+                value="personsname",
+                visible=True,
+            )
             progress_data = gr.Label(
                 label="Progress:"
             )
@@ -1345,14 +1384,15 @@ if __name__ == "__main__":
 
             prompt_compute_btn = gr.Button(value="Step 1 - Create dataset")
         
-            def preprocess_dataset(language,  whisper_model, out_path, eval_split_number, progress=gr.Progress(track_tqdm=True)):
+            def preprocess_dataset(language, whisper_model, out_path, eval_split_number, speaker_name_input, progress=gr.Progress(track_tqdm=True)):
                 clear_gpu_cache()
                 test_for_audio_files = [file for file in os.listdir(audio_folder) if any(file.lower().endswith(ext) for ext in ['.wav', '.mp3', '.flac'])]
                 if not test_for_audio_files:
                     return "I cannot find any mp3, wav or flac files in the folder called 'put-voice-samples-in-here'", "", ""
                 else:
                     try:
-                        train_meta, eval_meta, audio_total_size = format_audio_list(target_language=language, whisper_model=whisper_model, out_path=out_path, eval_split_number=eval_split_number, gradio_progress=progress)
+
+                        train_meta, eval_meta, audio_total_size = format_audio_list(target_language=language, whisper_model=whisper_model, out_path=out_path, eval_split_number=eval_split_number, speaker_name_input=speaker_name_input, gradio_progress=progress)
                     except:
                         traceback.print_exc()
                         error = traceback.format_exc()
@@ -1373,111 +1413,206 @@ if __name__ == "__main__":
 #### GRADIO STEP 2 ####
 #######################
         with gr.Tab("💻 Step 2 - Training"):
-
+            with gr.Tab("Training"):
-            gr.Markdown(
-                f"""
-                ### 💻 <u>Training</u><br>                
-                ### 🟦 <u>What you need to do</u>
-                ◽ The <span style="color: #3366ff;">Train CSV</span> and <span style="color: #3366ff;">Eval CSV</span> should already be populated. If not, just go back to Step 1 and click "Create Dataset" again.<br>
-                ◽ The default settings below are the suggested settings for most purposes, however you may choose to alter them depending on your specific use case.<br>
-                ◽ There are no absolute correct settings for training. It will vary based on:<br>
-                &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;- What you are training (A human voice, A cartoon voice, A new language entirely etc)<br>
-                &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;- How much audio you have (you may want less or more eval or epochs)<br>
-                ◽ The key indicator of sufficient model training is whether it sounds right to you. Coqui suggests beginning with the base settings for training. If the resulting audio doesn't meet your expectations, additional training sessions may be necessary.<br>
-                ◽ **Number of epochs:** Defines how many times the entire dataset is passed through the neural network; more epochs can improve learning but increase the risk of memorizing the training data.<br>
-                ◽ **Batch size:** Sets how many training examples are used in one iteration. Larger sizes speed up training but require more computational memory.<br>
-                ◽ **Grad(ient) accumulation steps:** Allows larger batches by spreading the computation across multiple steps, enabling training with limited hardware resources.<br>
-                ◽ **Max permitted audio size in seconds:** Limits the length of audio that the model processes. Any audio exceeding this duration will be truncated to fit the model's input constraints.<br>
-                ### 🟨 <u>What this step is doing</u><br>
-                ◽ Very simply put, it's taking all our wav files generated in Step 1, along with our recorded speech in out excel documents and its training the model on that voice e.g. listen to this audio file and this is what is being said in it, so learn to reproduce it.<br>
-                ### 🟩 <u>How long will this take?</u><br>
-                ◽ On a RTX 4070 with factory settings (as below) and 20 minutes of audio, this took 20 minutes. Again, it will vary by system and how much audio you are throwing at it. You have time to go make a coffee.<br>
-                ◽ Look at your command prompt/terminal window if you want to see what it is doing.<br>
-                """
-            )
-            with gr.Row():
-                train_csv = gr.Textbox(
-                    label="Train CSV:",
-                )
-                eval_csv = gr.Textbox(
-                    label="Eval CSV:",
-                )
-            with gr.Row():
-                num_epochs =  gr.Slider(
-                    label="Number of epochs:",
-                    minimum=1,
-                    maximum=100,
-                    step=1,
-                    value=args.num_epochs,
-                )
-                batch_size = gr.Slider(
-                    label="Batch size:",
-                    minimum=2,
-                    maximum=512,
-                    step=1,
-                    value=args.batch_size,
-                )
-                grad_acumm = gr.Slider(
-                    label="Grad accumulation steps:",
-                    minimum=2,
-                    maximum=128,
-                    step=1,
-                    value=args.grad_acumm,
-                )
-                max_audio_length = gr.Slider(
-                    label="Max permitted audio size in seconds:",
-                    minimum=2,
-                    maximum=20,
-                    step=1,
-                    value=args.max_audio_length,
-                )
-            with gr.Row():
-                model_to_train_choices = ["Base Model"]
-                if finetuned_model:
-                    model_to_train_choices.append("Existing finetuned model")
-                model_to_train = gr.Radio(
-                    choices=model_to_train_choices,
-                    label="Which model do you want to train?",
-                    value="Base Model"
-                )
                 gr.Markdown(
-                f"""
+                    f"""
-                If you have an existing finetuned model in <span style="color: #3366ff;">/models/trainedmodel/</span> you will have an option to select <span style="color: #3366ff;">Existing finetuned model</span>, allowing you to further train it. Otherwise, you will only have Base Model as an option.
+                    ### 💻 <u>Training</u><br>
-                """
+                    ### 🟥 <u>Important Note - Language support.</u>
+                    ◽ If this step is failing/erroring you may wish to check your training data was created correctly (Detailed in Step 1), confirming that wav files have been generated and your `metadata_train.csv` and `metadata_eval.csv` files have been populated.<br>             
+                    ### 🟦 <u>What you need to do</u>
+                    ◽ The <span style="color: #3366ff;">Train CSV</span> and <span style="color: #3366ff;">Eval CSV</span> should already be populated. If not, just go back to Step 1 and click "Create Dataset" again.<br>
+                    ◽ The default settings below are the suggested settings for most purposes, however you may choose to alter them depending on your specific use case.<br>
+                    ◽ There are no absolute correct settings for training. It will vary based on:<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;- What you are training (A human voice, A cartoon voice, A new language entirely etc)<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;- How much audio you have (you may want less or more eval or epochs)<br>
+                    ◽ The key indicator of sufficient model training is whether it sounds right to you. Coqui suggests beginning with the base settings for training. If the resulting audio doesn't meet your expectations, additional training sessions may be necessary.<br>
+                    ### 🟨 <u>What this step is doing</u><br>
+                    ◽ Very simply put, it's taking all our wav files generated in Step 1, along with our recorded speech in out excel documents and its training the model on that voice e.g. listen to this audio file and this is what is being said in it, so learn to reproduce it.<br>
+                    ### 🟩 <u>How long will this take?</u><br>
+                    ◽ On a RTX 4070 with factory settings (as below) and 20 minutes of audio, this took 20 minutes. Again, it will vary by system and how much audio you are throwing at it. You have time to go make a coffee.<br>
+                    ◽ Look at your command prompt/terminal window if you want to see what it is doing.<br>
+                    """
+                )
+                with gr.Row():
+                    train_csv = gr.Textbox(
+                        label="Train CSV:",
+                        scale=2,
+                    )
+                    eval_csv = gr.Textbox(
+                        label="Eval CSV:",
+                        scale=2,
+                    )
+                    learning_rates = gr.Dropdown(
+                        value=5e-6,
+                        label="Learning Rate",
+                        choices=[
+                            ("1e-6", 1e-6),
+                            ("5e-6", 5e-6),
+                            ("1e-5", 1e-5),
+                            ("5e-5", 5e-5),
+                            ("1e-4", 1e-4),
+                            ("5e-4", 5e-4),
+                            ("1e-3", 1e-3),
+                        ],
+                        type="value",
+                        allow_custom_value=True,
+                        scale=0,
+                    )
+                with gr.Row():
+                    num_epochs =  gr.Slider(
+                        label="Number of epochs:",
+                        minimum=1,
+                        maximum=100,
+                        step=1,
+                        value=args.num_epochs,
+                    )
+                    batch_size = gr.Slider(
+                        label="Batch size:",
+                        minimum=2,
+                        maximum=512,
+                        step=1,
+                        value=args.batch_size,
+                    )
+                    grad_acumm = gr.Slider(
+                        label="Grad accumulation steps:",
+                        minimum=1,
+                        maximum=32,
+                        step=1,
+                        value=args.grad_acumm,
+                    )
+                    max_audio_length = gr.Slider(
+                        label="Max permitted audio size in seconds:",
+                        minimum=2,
+                        maximum=20,
+                        step=1,
+                        value=args.max_audio_length,
+                    )
+                with gr.Row():
+                    model_to_train_choices = ["Base Model"]
+                    if finetuned_model:
+                        model_to_train_choices.append("Existing finetuned model")
+                    model_to_train = gr.Radio(
+                        choices=model_to_train_choices,
+                        label="Which model do you want to train?",
+                        value="Base Model"
+                    )
+                    gr.Markdown(
+                    f"""
+                    If you have an existing finetuned model in <span style="color: #3366ff;">/models/trainedmodel/</span> you will have an option to select <span style="color: #3366ff;">Existing finetuned model</span>, allowing you to further train it. Otherwise, you will only have Base Model as an option.
+                    """
+                    )
+
+                progress_train = gr.Label(
+                    label="Progress:"
+                )
+                logs_tts_train = gr.Textbox(
+                    label="Logs:",
+                    interactive=False,
+                )
+                demo.load(read_logs, None, logs_tts_train, every=1)
+                train_btn = gr.Button(value="Step 2 - Run the training")
+
+                def train_model(language, train_csv, eval_csv, learning_rates, num_epochs, batch_size, grad_acumm, output_path, max_audio_length):
+                    clear_gpu_cache()
+                    if not train_csv or not eval_csv:
+                        return "You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !", "", "", "", ""
+                    try:
+                        # convert seconds to waveform frames
+                        max_audio_length = int(max_audio_length * 22050)
+                        learning_rate = float(learning_rates)  # Convert the learning rate value to a float
+                        config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, learning_rate, output_path=str(output_path), max_audio_length=max_audio_length)
+                    except:
+                        traceback.print_exc()
+                        error = traceback.format_exc()
+                        return f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}", "", "", "", ""
+
+                    # copy original files to avoid parameters changes issues
+                    shutil.copy(config_path, exp_path)
+                    shutil.copy(vocab_file, exp_path)
+
+                    ft_xtts_checkpoint = os.path.join(exp_path, "best_model.pth")
+                    print("[FINETUNE] Model training done. Move to Step 3")
+                    clear_gpu_cache()
+                    return "Model training done. Move to Step 3", config_path, vocab_file, ft_xtts_checkpoint, speaker_wav
+            with gr.Tab("Info - Grad Accumulation"):
+                                gr.Markdown(
+                    f"""
+                    ### 🟩 <u>Grad Accumulation</u><br>
+                    ◽ Gradient accumulation is a technique that allows you to simulate a larger batch size without actually increasing the memory consumption. Here's how it works:<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ Instead of updating the model's parameters after every batch, gradient accumulation enables you to process multiple batches and accumulate their gradients.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ By setting steps greater than 1, you can process multiple batches before a single optimization step e.g if steps is set to 4, the gradients will be accumulated over 4 batches before updating the model's parameters.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ This means that you can effectively use a larger batch size while keeping the actual batch size per iteration smaller, thereby reducing the memory footprint.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ Increasing the steps allows you to find a balance between memory consumption and computational efficiency. You can process more examples per optimization step without exceeding the available memory.<br><br>
+                    ◽ However, it's important to note that increasing the gradient accumulation steps does have an impact on the training process:<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ Since the model's parameters are updated less frequently (every grad_accum_steps batches), the training dynamics may be slightly different compared to updating the model after every batch.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ You may need to adjust the learning rate and other hyperparameters accordingly to compensate for the less frequent updates. Typically, you can increase the learning rate slightly when using gradient accumulation.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ The training progress will be slower in terms of the number of optimization steps per epoch, as the model updates occur less frequently.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ However, the overall training time may still be reduced compared to using a smaller batch size without gradient accumulation, as it allows for better utilization of GPU resources.<br>
+                    &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;◽ To start, you can try setting steps to a value like 4 and see if it resolves the OOM error. If the error persists, you can experiment with higher values until you find a balance that works for your specific setup.
+                    """
+                )
+            with gr.Tab("Info - Batch Size"):
+                                gr.Markdown(
+                    f"""
+                    ### 🟩 <u>Batch Size</u><br>
+                    ◽ Batch size refers to the number of training examples processed in a single iteration (forward and backward pass) before updating the model's parameters.<br>
+                    ◽ A larger batch size allows for more efficient computation and utilization of hardware resources, especially when using GPUs.<br>
+                    ◽ However, increasing the batch size also increases the memory consumption during training, as more examples need to be stored in memory.<br>
+                    ◽ If the batch size is too large, it can lead to out-of-memory (OOM) errors, especially when training on GPUs with limited memory.<br>
+                    ◽ On the other hand, using a smaller batch size can result in slower training and may require more iterations to converge.<br>
+                    ◽ The optimal batch size depends on various factors, such as the model architecture, available hardware resources, and the specific problem being solved.<br>
+                    ◽ It's common to experiment with different batch sizes to find a balance between training speed and memory efficiency.<br>
+                    ◽ If you encounter OOM errors, try reducing the batch size until the training can proceed without memory issues.<br>
+                    ◽ Keep in mind that the batch size also affects the model's generalization and convergence behavior, so it's important to monitor the model's performance while adjusting the batch size.
+                    """
+                )
+            with gr.Tab("Info - Learning Rate"):
+                                gr.Markdown(
+                    f"""
+                    ### 🟩 <u>Learning Rate</u><br>
+                    ◽ 1e-6: Very small learning rate, slow but stable learning.<br>
+                    ◽ 5e-6: Small learning rate, slow but stable learning.<br>
+                    ◽ 1e-5: Moderate learning rate, balanced between stability and convergence speed.<br>
+                    ◽ 5e-5: Higher learning rate, faster convergence but potential instability.<br>
+                    ◽ 1e-4: High learning rate, faster convergence but increased risk of instability.<br>
+                    ◽ 5e-4: Very high learning rate, fast convergence but higher risk of instability.<br>
+                    ◽ 1e-3: Extremely high learning rate, very fast convergence but high risk of instability.<br><br>
+                    The optimal learning rate depends on the model architecture, dataset, and other hyperparameters.
+                    """
+                )
+            with gr.Tab("Info - Epochs"):
+                                gr.Markdown(
+                    f"""
+                    ### 🟩 <u>Epochs</u><br>
+                    ◽ An epoch represents a single pass through the entire training dataset during the training process.<br>
+                    ◽ In each epoch, the model sees and learns from all the training examples once.<br>
+                    ◽ The number of epochs determines how many times the model will iterate over the complete training dataset.<br>
+                    ◽ Increasing the number of epochs allows the model to learn more from the data and potentially improve its performance.<br>
+                    ◽ However, training for too many epochs can lead to overfitting, where the model becomes too specialized to the training data and fails to generalize well to unseen data.<br>
+                    ◽ The optimal number of epochs depends on factors such as the complexity of the problem, the size of the dataset, and the model's capacity.<br>
+                    ◽ It's common to use techniques like early stopping or validation monitoring to determine the appropriate number of epochs.<br>
+                    ◽ Early stopping involves monitoring the model's performance on a validation set and stopping the training when the performance starts to degrade, indicating potential overfitting.<br>
+                    ◽ Validation monitoring involves evaluating the model's performance on a separate validation set after each epoch and selecting the model checkpoint that performs best on the validation set.<br>
+                    ◽ It's recommended to start with a reasonable number of epochs and monitor the model's performance to determine if more epochs are needed or if early stopping should be applied
+                    """
+                )
+            with gr.Tab("Info - Max Audio"):
+                                gr.Markdown(
+                    f"""
+                    ### 🟩 <u>Max Permitted Audio Size (in seconds)</u><br>
+                    ◽ The max permitted audio size determines the maximum duration of audio files that can be used as input to the model during training and inference.<br>
+                    ◽ It is specified in seconds and represents the longest audio clip that the model can process.<br>
+                    ◽ Limiting the audio size helps to control the memory usage and computational requirements of the model.<br>
+                    ◽ If the audio files in your dataset vary in length, it's important to consider the max permitted audio size to ensure that all files can be processed effectively.<br>
+                    ◽ Audio files longer than the max permitted size will typically be truncated or split into smaller segments before being fed into the model.<br>
+                    ◽ The choice of max permitted audio size depends on the specific requirements of your application and the available hardware resources.<br>
+                    ◽ Increasing the max permitted audio size allows the model to handle longer audio clips but also increases the memory consumption and computational burden.<br>
+                    ◽ If you encounter memory issues or slow processing times, you may need to reduce the max permitted audio size to find a suitable balance.<br>
+                    ◽ It's important to consider the nature of your audio data and choose a max permitted audio size that captures the relevant information while being computationally feasible.
+                    """
                 )
 
-            progress_train = gr.Label(
+                
-                label="Progress:"
-            )
-            logs_tts_train = gr.Textbox(
-                label="Logs:",
-                interactive=False,
-            )
-            demo.load(read_logs, None, logs_tts_train, every=1)
-            train_btn = gr.Button(value="Step 2 - Run the training")
-
-            def train_model(language, train_csv, eval_csv, num_epochs, batch_size, grad_acumm, output_path, max_audio_length):
-                clear_gpu_cache()
-                if not train_csv or not eval_csv:
-                    return "You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !", "", "", "", ""
-                try:
-                    # convert seconds to waveform frames
-                    max_audio_length = int(max_audio_length * 22050)
-                    config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path=str(output_path), max_audio_length=max_audio_length)
-                except:
-                    traceback.print_exc()
-                    error = traceback.format_exc()
-                    return f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}", "", "", "", ""
-
-                # copy original files to avoid parameters changes issues
-                shutil.copy(config_path, exp_path)
-                shutil.copy(vocab_file, exp_path)
-
-                ft_xtts_checkpoint = os.path.join(exp_path, "best_model.pth")
-                print("[FINETUNE] Model training done. Move to Step 3")
-                clear_gpu_cache()
-                return "Model training done. Move to Step 3", config_path, vocab_file, ft_xtts_checkpoint, speaker_wav
-            
 #######################
 #### GRADIO STEP 3 ####
 #######################
@@ -1572,7 +1707,7 @@ if __name__ == "__main__":
                         refresh_button = create_refresh_button(
                             [xtts_checkpoint, xtts_config, xtts_vocab, speaker_reference_audio],
                             [
-                                lambda: {"choices": find_models(main_directory, "pth"), "value": ""},
+                                lambda: {"choices": find_best_models(main_directory), "value": ""},
                                 lambda: {"choices": find_jsons(main_directory, "config.json"), "value": ""},
                                 lambda: {"choices": find_jsons(main_directory, "vocab.json"), "value": ""},
                                 lambda: {"choices": get_available_voices(), "value": ""},
@@ -1605,7 +1740,8 @@ if __name__ == "__main__":
                 ◽ **Note:** Sample wav files will be copied into the model folder. You will need to MANUALLY copy/move the WAV file(s) you want to use to the AllTalk voices folder, to make them available within the interface.<br>
                 ◽ **Note:** Text-generation-webui users, <span style="color: #3366ff;">OPTION A - Compact and move model to '/trainedmodel/'</span>. will make the model available for use within Text-gen-webui.<br>
                 ### 🟩 <u>Using your Finetuned model as you main model</u>
-                ◽ If you wish to use the model you have just fintuned as your main model, you will need to MANUALLY copy it over the base model in <span style="color: #3366ff;">/models/trainedmodel/</span> after you have used Option A or B.<br>
+                ◽ **Currently** only Text-gen-webui and SillyTavern have loaders built for loading models in the <span style="color: #3366ff;">'/finetuned/'</span> folder. So if you wish to use your finetuned model as elsewhere in AllTalk please copy it over your base model e.g. copy your model over the top of <span style="color: #3366ff;">/models/xttsv2_2.0.2/</span><br>
+                ◽ If you wish to use the model you have just fintuned as your main model, you will need to MANUALLY copy it over the base model e.g copy it over <span style="color: #3366ff;">/models/xttsv2_2.0.2/</span> after you have used Option A or B.<br>
                 ◽ Should you ever wish to revert your model back to the base model, you can delete the files in <span style="color: #3366ff;">/models/xttsv2_2.0.2/</span> and new copies will be re-downloaded when AllTalk starts up.<br>
                 ### 🟨 <u>Clearing up disk space</u>
                 ◽ If you are not going to train anything again, you can delete the whisper model from inside of your huggingface cache (3GB approx) <br>
@@ -1617,6 +1753,21 @@ if __name__ == "__main__":
             final_progress_data = gr.Label(
                 label="Progress:"
             )
+            with gr.Row():
+                xtts_checkpoint_copy = gr.Dropdown(
+                    [str(file) for file in xtts_checkpoint_files],
+                    label="XTTS checkpoint path (Select the model you want to copy over):",
+                    value="",
+                    allow_custom_value=True,
+                    scale=3,
+                )
+                # Create refresh button
+                refresh_button = create_refresh_button(
+                    [xtts_checkpoint_copy,],
+                    [
+                        lambda: {"choices": find_best_models(main_directory), "value": ""},
+                    ],
+                    elem_class="refresh-button-class")
             with gr.Row():
                 compact_btn = gr.Button(value="OPTION A - Compact and move model to '/trainedmodel/'")
                 gr.Markdown(
@@ -1631,6 +1782,9 @@ if __name__ == "__main__":
                 This will compact the model and move it, along with the reference wav's to <span style="color: #3366ff;">/models/lastfinetuned/</span> and will <span style="color: red;">OVERWRITE</span> any model in that location. This will just leave you a finetuned model that you can do with as you please. It will <span style="color: red;">NOT</span> be available in Text-gen-webui unless you copy it over one of the other model loader locations.
                 """
                 )
+            with gr.Row():
+                compact_custom_btn = gr.Button(value="OPTION C - Compact and move model to a folder name of your choosing")
+                folder_path = gr.Textbox(label="Enter a new folder name (will be sub the models folder)", lines=1, value="mycustomfolder")
             with gr.Row():
                 delete_training_btn = gr.Button(value="Delete generated training data")
                 gr.Markdown(
@@ -1653,6 +1807,7 @@ if __name__ == "__main__":
                     whisper_model,
                     out_path,
                     eval_split_number,
+                    speaker_name_input,
                 ],
                 outputs=[
                     progress_data,
@@ -1668,6 +1823,7 @@ if __name__ == "__main__":
                     lang,
                     train_csv,
                     eval_csv,
+                    learning_rates,
                     num_epochs,
                     batch_size,
                     grad_acumm,
@@ -1695,14 +1851,20 @@ if __name__ == "__main__":
                     speaker_reference_audio,
                 ],
                 outputs=[progress_gen, tts_output_audio, reference_audio],
-            )
+            )    
-            
             compact_btn.click(
                 fn=compact_model,
+                inputs=[xtts_checkpoint_copy],
                 outputs=[final_progress_data],
             )
             compact_lastfinetuned_btn.click(
                 fn=compact_lastfinetuned_model,
+                inputs=[xtts_checkpoint_copy],
+                outputs=[final_progress_data],
+            )
+            compact_custom_btn.click(
+                fn=compact_custom_model,
+                inputs=[xtts_checkpoint_copy, folder_path],
                 outputs=[final_progress_data],
             )
             delete_training_btn.click(

nvidia.Dockerfile

@@ -1,20 +1,22 @@
-FROM nvidia/cuda:12.1.1-devel-ubuntu22.04
+FROM nvidia/cuda:12.1.1-devel-ubuntu22.04
-ENV HOST 0.0.0.0
+ENV HOST 0.0.0.0
-RUN apt-get update && apt-get upgrade -y \
+RUN apt-get update && apt-get upgrade -y \
-    && apt-get install -y git build-essential portaudio19-dev \
+    && apt-get install -y git build-essential portaudio19-dev \
-    python3 python3-pip gcc wget \
+    python3 python3-pip gcc wget \
-    ocl-icd-opencl-dev opencl-headers clinfo \
+    ocl-icd-opencl-dev opencl-headers clinfo \
-    libclblast-dev libopenblas-dev \
+    libclblast-dev libopenblas-dev \
-    && mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd \
+    && mkdir -p /etc/OpenCL/vendors && echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd \
-    && ln -s /usr/bin/python3 /usr/bin/python
+    && ln -s /usr/bin/python3 /usr/bin/python
-WORKDIR /app
+
-COPY . .
+RUN mkdir /app/pip_cache
-ENV CUDA_DOCKER_ARCH=all
+ENV PIP_CACHE_DIR=/app/pip_cache
-
+
-RUN pip install --no-cache-dir -r requirements_nvidia.txt && \
+WORKDIR /app
-    pip install --no-cache-dir -r requirements_finetune.txt && \
+COPY . .
-    pip install --no-cache-dir -r requirements_other.txt && \
+ENV CUDA_DOCKER_ARCH=all
-    pip install --no-cache-dir deepspeed
+RUN pip install --no-cache-dir --no-deps -r system/requirements/requirements_standalone.txt --cache-dir=/app/pip_cache && \
-EXPOSE 7851
+    pip install --no-cache-dir deepspeed --cache-dir=/app/pip_cache
-RUN chmod +x launch.sh
+
+EXPOSE 7851
+RUN chmod +x launch.sh
 ENTRYPOINT ["sh", "-c", "./launch.sh"]

script.py

@@ -174,7 +174,7 @@ def get_available_voices():
 ############################################
 def delete_old_files(folder_path, days_to_keep):
     current_time = datetime.now()
-    print(f"[{params['branding']}Startup] \033[92mWAV file deletion      :\033[93m {formatted_date}\033[0m",delete_output_wavs_setting,)
+    print(f"[{params['branding']}Startup] \033[92mWAV file deletion      :\033[93m", delete_output_wavs_setting,"\033[0m")
     for file_name in os.listdir(folder_path):
         file_path = os.path.join(folder_path, file_name)
         if os.path.isfile(file_path):
@@ -230,7 +230,7 @@ if finetuned_model:
         (trained_model_directory / file).exists() for file in required_files
     )
 if finetuned_model:
-    print(f"[{params['branding']}Startup] Finetuned model \033[93mDetected\033[0m")
+    print(f"[{params['branding']}Startup] \033[92mFinetuned model        :\033[93m Detected\033[0m")
 
 ####################################################
 #### SET GRADIO BUTTONS BASED ON confignew.json ####
@@ -698,9 +698,11 @@ def output_modifier(string, state):
                         print(f"[{params['branding']}TTSGen] {voice_description}")
 
                     # Replace multiple exclamation marks, question marks, or other punctuation with a single instance
-                    cleaned_part = re.sub(r"([!?.])\1+", r"\1", part)
+                    cleaned_part = re.sub(r"([!?.\u3002\uFF1F\uFF01\uFF0C])\1+", r"\1", part)
+                    # Replace "Chinese ellipsis" with a single dot
+                    cleaned_part = re.sub(r"\u2026{1,2}", ". ", cleaned_part)
                     # Further clean to remove any other unwanted characters
-                    cleaned_part = re.sub(r'[^a-zA-Z0-9\s\.,;:!?\-\'"\u0100-\u017F\u0400-\u04FFÀ-ÿ\u0150\u0151\u0170\u0171\u4E00-\u9FFF\u3400-\u4DBF\u20000-\u2A6DF\u3000-\u303F\$]', '', cleaned_part)
+                    cleaned_part = re.sub(r'[^a-zA-Z0-9\s.,;:!?\-\'"$\u0400-\u04FF\u00C0-\u00FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]', '', cleaned_part)
                     # Remove all newline characters (single or multiple)
                     cleaned_part = re.sub(r"\n+", " ", cleaned_part)
 
@@ -725,9 +727,11 @@ def output_modifier(string, state):
                 # Decode HTML entities first
                 cleaned_part = html.unescape(cleaned_string)
                 # Replace multiple instances of certain punctuation marks with a single instance
-                cleaned_part = re.sub(r"([!?.])\1+", r"\1", cleaned_part)
+                cleaned_part = re.sub(r"([!?.\u3002\uFF1F\uFF01\uFF0C])\1+", r"\1", cleaned_part)
+                # Replace "Chinese ellipsis" with a single dot
+                cleaned_part = re.sub(r"\u2026{1,2}", ". ", cleaned_part)
                 # Further clean to remove any other unwanted characters
-                cleaned_part = re.sub(r'[^a-zA-Z0-9\s\.,;:!?\-\'"\u0100-\u017F\u0400-\u04FFÀ-ÿ\u0150\u0151\u0170\u0171\u4E00-\u9FFF\u3400-\u4DBF\u20000-\u2A6DF\u3000-\u303F\$]', '', cleaned_part)
+                cleaned_part = re.sub(r'[^a-zA-Z0-9\s.,;:!?\-\'"$\u0400-\u04FF\u00C0-\u00FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]', '', cleaned_part)
                 # Remove all newline characters (single or multiple)
                 cleaned_part = re.sub(r"\n+", " ", cleaned_part)
                 # Process the part and give it a non-character name if being used vai API or standalone.

system/admin.html

@@ -304,6 +304,7 @@
                 <option value="nl">Dutch</option>
                 <option value="fr">French</option>
                 <option value="de">German</option>
+                <option value="hi">Hindi</option>
                 <option value="hu">Hungarian</option>
                 <option value="it">Italian</option>
                 <option value="ja">Japanese</option>
@@ -2500,6 +2501,7 @@
         <code>en</code> English<br>
         <code>fr</code> French<br>
         <code>de</code> German<br>
+        <code>hi</code> Hindi<br>
         <code>hu</code> Hungarian<br>
         <code>it</code> Italian<br>
         <code>ja</code> Japanese<br>

system/at_admin/at_settings.html

@@ -124,6 +124,7 @@
             <option value="English" {% if data.language=='English' %}selected{% endif %}>English</option>
             <option value="French" {% if data.language=='French' %}selected{% endif %}>French</option>
             <option value="German" {% if data.language=='German' %}selected{% endif %}>German</option>
+            <option value="Hindi" {% if data.language=='Hindi' %}selected{% endif %}>Hindi</option>
             <option value="Hungarian" {% if data.language=='Hungarian' %}selected{% endif %}>Hungarian</option>
             <option value="Italian" {% if data.language=='Italian' %}selected{% endif %}>Italian</option>
             <option value="Japanese" {% if data.language=='Japanese' %}selected{% endif %}>Japanese</option>

system/config/languages.json

@@ -6,6 +6,7 @@
   "English": "en",
   "French": "fr",
   "German": "de",
+  "Hindi": "hi",
   "Hungarian": "hu",
   "Italian": "it",
   "Japanese": "ja",

system/requirements/requirements_docker.txt

@@ -0,0 +1,173 @@
+absl-py==2.1.0
+aiofiles==23.2.1
+aiohttp==3.9.3
+aiosignal==1.3.1
+altair==5.2.0
+annotated-types==0.6.0
+anyascii==0.3.2
+anyio==4.3.0
+attrs==23.2.0
+audioread==3.0.1
+av==11.0.0
+Babel==2.14.0
+bangla==0.0.2
+blinker==1.7.0
+blis==0.7.11
+bnnumerizer==0.0.2
+bnunicodenormalizer==0.1.6
+catalogue==2.0.10
+certifi==2024.2.2
+cffi==1.16.0
+charset-normalizer==3.3.2
+click==8.1.7
+cloudpathlib==0.16.0
+colorama==0.4.6
+coloredlogs==15.0.1
+confection==0.1.4
+contourpy==1.2.0
+coqpit==0.0.17
+ctranslate2==4.1.0
+cutlet==0.4.0
+cycler==0.12.1
+cymem==2.0.8
+Cython==3.0.9
+dateparser==1.1.8
+decorator==5.1.1
+docopt==0.6.2
+einops==0.7.0
+encodec==0.1.1
+exceptiongroup
+fastapi==0.110.0
+faster-whisper==1.0.1
+ffmpy==0.3.2
+filelock==3.13.3
+Flask==3.0.2
+flatbuffers==24.3.25
+fonttools==4.50.0
+frozenlist==1.4.1
+fsspec==2024.3.1
+fugashi==1.3.1
+fuzzywuzzy>=0.18.0
+g2pkk==0.1.2
+gradio==3.50.2
+gradio_client==0.6.1
+grpcio==1.62.1
+gruut==2.2.3
+gruut-ipa==0.13.0
+gruut-lang-de==2.0.0
+gruut-lang-en==2.0.0
+gruut-lang-es==2.0.0
+gruut-lang-fr==2.0.2
+h11==0.14.0
+hangul-romanize==0.1.0
+httpcore==1.0.4
+httpx==0.27.0
+huggingface-hub==0.22.1
+humanfriendly==10.0
+idna==3.6
+importlib_metadata==7.1.0
+importlib_resources==6.4.0
+inflect==7.0.0
+itsdangerous==2.1.2
+jaconv==0.3.4
+jamo==0.4.1
+jieba==0.42.1
+Jinja2==3.1.3
+joblib==1.3.2
+jsonlines==1.2.0
+jsonschema==4.21.1
+jsonschema-specifications==2023.12.1
+kiwisolver==1.4.5
+langcodes==3.3.0
+lazy_loader==0.3
+librosa==0.10.1
+llvmlite==0.42.0
+Markdown==3.6
+MarkupSafe==2.1.5
+matplotlib==3.8.3
+mojimoji==0.0.13
+mpmath==1.3.0
+msgpack==1.0.8
+multidict==6.0.5
+murmurhash==1.0.10
+networkx==2.8.8
+nltk==3.8.1
+num2words==0.5.13
+numba==0.59.1
+numpy==1.26.4
+nvidia-cublas-cu11>=11.11.3.6
+nvidia-cudnn-cu11>=9.0.0.312
+onnxruntime==1.17.1
+orjson==3.9.15
+packaging==24.0
+pandas==1.5.3
+pillow==10.2.0
+platformdirs==4.2.0
+pooch==1.8.1
+preshed==3.0.9
+protobuf==5.26.0
+psutil==5.9.8
+pycparser==2.21
+pydantic==2.6.4
+pydantic_core==2.16.3
+pydub==0.25.1
+pynndescent==0.5.11
+pyparsing==3.1.2
+pypinyin==0.51.0
+pyreadline3==3.4.1
+pysbd==0.3.4
+python-crfsuite==0.9.10
+python-dateutil==2.9.0.post0
+python-Levenshtein>=0.25.0
+python-multipart==0.0.9
+pytz==2024.1
+PyYAML==6.0.1
+referencing==0.34.0
+regex==2023.12.25
+requests==2.31.0
+rpds-py==0.18.0
+safetensors==0.4.2
+scikit-learn==1.4.1.post1
+scipy==1.12.0
+semantic-version==2.10.0
+six==1.16.0
+smart-open==6.4.0
+sniffio==1.3.1
+sounddevice==0.4.6
+soundfile==0.12.1
+soxr==0.3.7
+spacy==3.7.4
+spacy-legacy==3.0.12
+spacy-loggers==1.0.5
+srsly==2.4.8
+starlette==0.36.3
+SudachiDict-core==20240109
+SudachiPy==0.6.8
+sympy==1.12
+tensorboard==2.16.2
+tensorboard-data-server==0.7.2
+thinc==8.2.3
+threadpoolctl==3.4.0
+tokenizers==0.15.2
+toolz==0.12.1
+torch>=2.2.0
+torchaudio>=2.2.0
+tqdm==4.66.2
+trainer==0.0.36
+transformers==4.39.1
+TTS==0.22.0
+typer==0.9.4
+typing_extensions==4.10.0
+tzdata==2024.1
+tzlocal==5.2
+umap-learn==0.5.5
+Unidecode==1.3.8
+unidic-lite==1.0.8
+urllib3==2.2.1
+uvicorn==0.29.0
+wasabi==1.1.2
+weasel==0.3.4
+websockets==11.0.3
+Werkzeug==3.0.1
+yarl==1.9.4
+zipp==3.18.1

system/requirements/requirements_textgen.txt

@@ -3,8 +3,8 @@ soundfile>=0.12.1
 uvicorn>=0.24.0.post1
 transformers>=4.37.1
 TTS>=0.22.0
-torch>=2.1.0+cu118
+torch>=2.1.0
-torchaudio>=2.1.0+cu118
+torchaudio>=2.1.0
 fastapi>=0.104.1
 Jinja2>=3.1.2
 requests>=2.31.0
@@ -22,3 +22,4 @@ nvidia-cublas-cu11>=11.11.3.6
 nvidia-cudnn-cu11>=9.0.0.312
 fuzzywuzzy>=0.18.0
 python-Levenshtein>=0.25.0
+spacy>=3.7.2

system/st_files/alltalk.js

@@ -281,7 +281,7 @@ class AllTalkTtsProvider {
             if (narratorVoiceSelect) narratorVoiceSelect.disabled = !this.settings.narrator_enabled;
         }
         const modelSelect = document.getElementById('switch_model');
-        if (this.settings.finetuned_model === 'true') {
+        if (this.settings.finetuned_model === true) {
             const ftOption = document.createElement('option');
             ftOption.value = 'XTTSv2 FT';
             ftOption.textContent = 'XTTSv2 FT';
@@ -830,4 +830,4 @@ function updateStatus(message) {
                 break;
         }
     }
-}
+}

system/tts_generator/crunker.d.ts

@@ -176,4 +176,4 @@ export default class Crunker {
      * @internal
      */
     private _renderURL;
-}
+}

system/tts_generator/crunker.esm.js

@@ -1,2 +1,2 @@
-var e={d:(t,n)=>{for(var r in n)e.o(n,r)&&!e.o(t,r)&&Object.defineProperty(t,r,{enumerable:!0,get:n[r]})},o:(e,t)=>Object.prototype.hasOwnProperty.call(e,t)},t={};function n(e){return n="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},n(e)}function r(e){return function(e){if(Array.isArray(e))return a(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return a(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return a(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function a(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function o(e,t){if(!(e instanceof t))throw new TypeError("Cannot call a class as a function")}function i(e,t){for(var r=0;r<t.length;r++){var a=t[r];a.enumerable=a.enumerable||!1,a.configurable=!0,"value"in a&&(a.writable=!0),Object.defineProperty(e,(o=a.key,i=void 0,i=function(e,t){if("object"!==n(e)||null===e)return e;var r=e[Symbol.toPrimitive];if(void 0!==r){var a=r.call(e,t||"default");if("object"!==n(a))return a;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(o,"string"),"symbol"===n(i)?i:String(i)),a)}var o,i}e.d(t,{Z:()=>u});var u=function(){function e(){var t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},n=t.sampleRate;o(this,e),this._context=this._createContext(n),n||(n=this._context.sampleRate),this._sampleRate=n}var t,n,a;return t=e,n=[{key:"_createContext",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:44100;return window.AudioContext=window.AudioContext||window.webkitAudioContext||window.mozAudioContext,new AudioContext({sampleRate:e})}},{key:"context",get:function(){return this._context}},{key:"fetchAudio",value:async function(){for(var e=this,t=arguments.length,n=new Array(t),r=0;r<t;r++)n[r]=arguments[r];return await Promise.all(n.map((async function(t){var n;return n=t instanceof File||t instanceof Blob?await t.arrayBuffer():await fetch(t).then((function(e){return e.headers.has("Content-Type")&&!e.headers.get("Content-Type").includes("audio/")&&console.warn("Crunker: Attempted to fetch an audio file, but its MIME type is `".concat(e.headers.get("Content-Type").split(";")[0],"`. We'll try and continue anyway. (file: \"").concat(t,'")')),e.arrayBuffer()})),await e._context.decodeAudioData(n)})))}},{key:"mergeAudio",value:function(e){var t=this._context.createBuffer(this._maxNumberOfChannels(e),this._sampleRate*this._maxDuration(e),this._sampleRate);return e.forEach((function(e){for(var n=0;n<e.numberOfChannels;n++){for(var r=t.getChannelData(n),a=e.getChannelData(n),o=e.getChannelData(n).length-1;o>=0;o--)r[o]+=a[o];t.getChannelData(n).set(r)}})),t}},{key:"concatAudio",value:function(e){var t=this._context.createBuffer(this._maxNumberOfChannels(e),this._totalLength(e),this._sampleRate),n=0;return e.forEach((function(e){for(var r=0;r<e.numberOfChannels;r++)t.getChannelData(r).set(e.getChannelData(r),n);n+=e.length})),t}},{key:"padAudio",value:function(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:0,n=arguments.length>2&&void 0!==arguments[2]?arguments[2]:0;if(0===n)return e;if(t<0)throw new Error('Crunker: Parameter "padStart" in padAudio must be positive');if(n<0)throw new Error('Crunker: Parameter "seconds" in padAudio must be positive');for(var r=this._context.createBuffer(e.numberOfChannels,Math.ceil(e.length+n*e.sampleRate),e.sampleRate),a=0;a<e.numberOfChannels;a++){var o=e.getChannelData(a);r.getChannelData(a).set(o.subarray(0,Math.ceil(t*e.sampleRate)+1),0),r.getChannelData(a).set(o.subarray(Math.ceil(t*e.sampleRate)+2,r.length+1),Math.ceil((t+n)*e.sampleRate))}return r}},{key:"sliceAudio",value:function(e,t,n){var r=arguments.length>3&&void 0!==arguments[3]?arguments[3]:0,a=arguments.length>4&&void 0!==arguments[4]?arguments[4]:0;if(t>=n)throw new Error('Crunker: "start" time should be less than "end" time in sliceAudio method');for(var o=Math.round((n-t)*this._sampleRate),i=Math.round(t*this._sampleRate),u=this._context.createBuffer(e.numberOfChannels,o,this._sampleRate),l=0;l<e.numberOfChannels;l++)for(var s=e.getChannelData(l),c=u.getChannelData(l),f=0;f<o;f++)c[f]=s[i+f],f<r*this._sampleRate&&(c[f]*=f/(r*this._sampleRate)),f>o-a*this._sampleRate&&(c[f]*=(o-f)/(a*this._sampleRate));return u}},{key:"play",value:function(e){var t=this._context.createBufferSource();return t.buffer=e,t.connect(this._context.destination),t.start(),Line truncated
+var e={d:(t,n)=>{for(var r in n)e.o(n,r)&&!e.o(t,r)&&Object.defineProperty(t,r,{enumerable:!0,get:n[r]})},o:(e,t)=>Object.prototype.hasOwnProperty.call(e,t)},t={};function n(e){return n="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},n(e)}function r(e){return function(e){if(Array.isArray(e))return a(e)}(e)||function(e){if("undefined"!=typeof Symbol&&null!=e[Symbol.iterator]||null!=e["@@iterator"])return Array.from(e)}(e)||function(e,t){if(!e)return;if("string"==typeof e)return a(e,t);var n=Object.prototype.toString.call(e).slice(8,-1);"Object"===n&&e.constructor&&(n=e.constructor.name);if("Map"===n||"Set"===n)return Array.from(e);if("Arguments"===n||/^(?:Ui|I)nt(?:8|16|32)(?:Clamped)?Array$/.test(n))return a(e,t)}(e)||function(){throw new TypeError("Invalid attempt to spread non-iterable instance.\nIn order to be iterable, non-array objects must have a [Symbol.iterator]() method.")}()}function a(e,t){(null==t||t>e.length)&&(t=e.length);for(var n=0,r=new Array(t);n<t;n++)r[n]=e[n];return r}function o(e,t){if(!(e instanceof t))throw new TypeError("Cannot call a class as a function")}function i(e,t){for(var r=0;r<t.length;r++){var a=t[r];a.enumerable=a.enumerable||!1,a.configurable=!0,"value"in a&&(a.writable=!0),Object.defineProperty(e,(o=a.key,i=void 0,i=function(e,t){if("object"!==n(e)||null===e)return e;var r=e[Symbol.toPrimitive];if(void 0!==r){var a=r.call(e,t||"default");if("object"!==n(a))return a;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)}(o,"string"),"symbol"===n(i)?i:String(i)),a)}var o,i}e.d(t,{Z:()=>u});var u=function(){function e(){var t=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},n=t.sampleRate;o(this,e),this._context=this._createContext(n),n||(n=this._context.sampleRate),this._sampleRate=n}var t,n,a;return t=e,n=[{key:"_createContext",value:function(){var e=arguments.length>0&&void 0!==arguments[0]?arguments[0]:44100;return window.AudioContext=window.AudioContext||window.webkitAudioContext||window.mozAudioContext,new AudioContext({sampleRate:e})}},{key:"context",get:function(){return this._context}},{key:"fetchAudio",value:async function(){for(var e=this,t=arguments.length,n=new Array(t),r=0;r<t;r++)n[r]=arguments[r];return await Promise.all(n.map((async function(t){var n;return n=t instanceof File||t instanceof Blob?await t.arrayBuffer():await fetch(t).then((function(e){return e.headers.has("Content-Type")&&!e.headers.get("Content-Type").includes("audio/")&&console.warn("Crunker: Attempted to fetch an audio file, but its MIME type is `".concat(e.headers.get("Content-Type").split(";")[0],"`. We'll try and continue anyway. (file: \"").concat(t,'")')),e.arrayBuffer()})),await e._context.decodeAudioData(n)})))}},{key:"mergeAudio",value:function(e){var t=this._context.createBuffer(this._maxNumberOfChannels(e),this._sampleRate*this._maxDuration(e),this._sampleRate);return e.forEach((function(e){for(var n=0;n<e.numberOfChannels;n++){for(var r=t.getChannelData(n),a=e.getChannelData(n),o=e.getChannelData(n).length-1;o>=0;o--)r[o]+=a[o];t.getChannelData(n).set(r)}})),t}},{key:"concatAudio",value:function(e){var t=this._context.createBuffer(this._maxNumberOfChannels(e),this._totalLength(e),this._sampleRate),n=0;return e.forEach((function(e){for(var r=0;r<e.numberOfChannels;r++)t.getChannelData(r).set(e.getChannelData(r),n);n+=e.length})),t}},{key:"padAudio",value:function(e){var t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:0,n=arguments.length>2&&void 0!==arguments[2]?arguments[2]:0;if(0===n)return e;if(t<0)throw new Error('Crunker: Parameter "padStart" in padAudio must be positive');if(n<0)throw new Error('Crunker: Parameter "seconds" in padAudio must be positive');for(var r=this._context.createBuffer(e.numberOfChannels,Math.ceil(e.length+n*e.sampleRate),e.sampleRate),a=0;a<e.numberOfChannels;a++){var o=e.getChannelData(a);r.getChannelData(a).set(o.subarray(0,Math.ceil(t*e.sampleRate)+1),0),r.getChannelData(a).set(o.subarray(Math.ceil(t*e.sampleRate)+2,r.length+1),Math.ceil((t+n)*e.sampleRate))}return r}},{key:"sliceAudio",value:function(e,t,n){var r=arguments.length>3&&void 0!==arguments[3]?arguments[3]:0,a=arguments.length>4&&void 0!==arguments[4]?arguments[4]:0;if(t>=n)throw new Error('Crunker: "start" time should be less than "end" time in sliceAudio method');for(var o=Math.round((n-t)*this._sampleRate),i=Math.round(t*this._sampleRate),u=this._context.createBuffer(e.numberOfChannels,o,this._sampleRate),l=0;l<e.numberOfChannels;l++)for(var s=e.getChannelData(l),c=u.getChannelData(l),f=0;f<o;f++)c[f]=s[i+f],f<r*this._sampleRate&&(c[f]*=f/(r*this._sampleRate)),f>o-a*this._sampleRate&&(c[f]*=(o-f)/(a*this._sampleRate));return u}},{key:"play",value:function(e){var t=this._context.createBufferSource();return t.buffer=e,t.connect(this._context.destination),t.start(),Line truncated
 //# sourceMappingURL=crunker.esm.js.map

system/tts_generator/crunker.esm.js.map

@@ -1 +1 @@
-{"version":3,"file":"crunker.esm.js","mappings":"AACA,IAAIA,EAAsB,CCA1BA,EAAwB,CAACC,EAASC,KACjC,IAAI,IAAIC,KAAOD,EACXF,EAAoBI,EAAEF,EAAYC,KAASH,EAAoBI,EAAEH,EAASE,IAC5EE,OAAOC,eAAeL,EAASE,EAAK,CAAEI,YAAY,EAAMC,IAAKN,EAAWC,IAE1E,ECNDH,EAAwB,CAACS,EAAKC,IAAUL,OAAOM,UAAUC,eAAeC,KAAKJ,EAAKC,I,8jDCoBlF,IAGqBI,EAAO,WAU1B,aAAqE,6DAAJ,CAAC,EAApDC,EAAU,EAAVA,WAAU,UACtBC,KAAKC,SAAWD,KAAKE,eAAeH,GAEpCA,IAAAA,EAAeC,KAAKC,SAASF,YAE7BC,KAAKG,YAAcJ,CACrB,C,UA6YC,O,EA3YD,E,EAAA,6BAKA,WAAkE,IAA3CA,EAAqB,UAAH,6CAAG,MAE1C,OADAK,OAAOC,aAAeD,OAAOC,cAAiBD,OAAeE,oBAAuBF,OAAeG,gBAC5F,IAAIF,aAAa,CAAEN,WAAAA,GAC5B,GAEA,mBAIA,WACE,OAAOC,KAAKC,QACd,GAEA,wBAGAO,iBAA4E,kCAAxDC,EAAS,yBAATA,EAAS,gBAC3B,aAAaC,QAAQC,IACnBF,EAAUG,KAAIJ,eAAOK,GACnB,IAAIC,EAkBJ,OAfEA,EADED,aAAoBE,MAAQF,aAAoBG,WACnCH,EAASI,oBAETC,MAAML,GAAUM,MAAK,SAACC,GASnC,OARIA,EAASC,QAAQC,IAAI,kBAAoBF,EAASC,QAAQ7B,IAAI,gBAAiB+B,SAAS,WAC1FC,QAAQC,KAAK,oEAAD,OAERL,EAASC,QAAQ7B,IAAI,gBAAiBkC,MAAM,KAAK,GAAE,sDACPb,EAAQ,OAInDO,EAASH,aAClB,UAGW,EAAKhB,SAAS0B,gBAAgBb,EAC7C,IAEJ,GAEA,wBAOA,SAAWc,GACT,IAAMC,EAAS7B,KAAKC,SAAS6B,aAC3B9B,KAAK+B,qBAAqBH,GAC1B5B,KAAKG,YAAcH,KAAKgC,aAAaJ,GACrC5B,KAAKG,aAgBP,OAbAyB,EAAQK,SAAQ,SAACnB,GACf,IAAK,IAAIoB,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IAAiB,CAIpF,IAHA,IAAME,EAAaP,EAAOQ,eAAeH,GACnCI,EAAaxB,EAAOuB,eAAeH,GAEhCK,EAAIzB,EAAOuB,eAAeH,GAAeM,OAAS,EAAGD,GAAK,EAAGA,IACpEH,EAAWG,IAAMD,EAAWC,GAG9BV,EAAOQ,eAAeH,GAAeO,IAAIL,EAC3C,CACF,IAEOP,CACT,GAEA,yBAOA,SAAYD,GACV,IAAMC,EAAS7B,KAAKC,SAAS6B,aAC3B9B,KAAK+B,qBAAqBH,GAC1B5B,KAAK0C,aAAad,GAClB5B,KAAKG,aAEHwC,EAAS,EAUb,OARAf,EAAQK,SAAQ,SAACnB,GACf,IAAK,IAAIoB,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IACnEL,EAAOQ,eAAeH,GAAeO,IAAI3B,EAAOuB,eAAeH,GAAgBS,GAGjFA,GAAU7B,EAAO0B,MACnB,IAEOX,CACT,GAEA,sBAUA,SAASf,GAA6E,IAAxD8B,EAAmB,UAAH,6CAAG,EAAGC,EAAkB,UAAH,6CAAG,EACpE,GAAgB,IAAZA,EAAe,OAAO/B,EAE1B,GAAI8B,EAAW,EAAG,MAAM,IAAIE,MAAM,8DAClC,GAAID,EAAU,EAAG,MAAM,IAAIC,MAAM,6DAQjC,IANA,IAAMC,EAAgB/C,KAAKC,SAAS6B,aAClChB,EAAOqB,iBACPa,KAAKC,KAAKnC,EAAO0B,OAASK,EAAU/B,EAAOf,YAC3Ce,EAAOf,YAGAmC,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IAAiB,CACpF,IAAMgB,EAAcpC,EAAOuB,eAAeH,GAC1Ca,EACGV,eAAeH,GACfO,IAAIS,EAAYC,SAAS,EAAGH,KAAKC,KAAKL,EAAW9B,EAAOf,YAAc,GAAI,GAE7EgD,EACGV,eAAeH,GACfO,IACCS,EAAYC,SAASH,KAAKC,KAAKL,EAAW9B,EAAOf,YAAc,EAAGgD,EAAcP,OAAS,GACzFQ,KAAKC,MAAML,EAAWC,GAAW/B,EAAOf,YAE9C,CAEA,OAAOgD,CACT,GAEA,wBASA,SAAWjC,EAAqBsC,EAAeC,GAAmE,IAAtDC,EAAiB,UAAH,6CAAG,EAAGC,EAAkB,UAAH,6CAAG,EAChG,GAAIH,GAASC,EAAK,MAAM,IAAIP,MAAM,6EAMlC,IAJA,IAAMN,EAASQ,KAAKQ,OAAOH,EAAMD,GAASpD,KAAKG,aACzCwC,EAASK,KAAKQ,MAAMJ,EAAQpD,KAAKG,aACjCsD,EAAYzD,KAAKC,SAAS6B,aAAahB,EAAOqB,iBAAkBK,EAAQxC,KAAKG,aAE1EuD,EAAU,EAAGA,EAAU5C,EAAOqB,iBAAkBuB,IAIvD,IAHA,IAAMC,EAAY7C,EAAOuB,eAAeqB,GAClCtB,EAAaqB,EAAUpB,eAAeqB,GAEnCnB,EAAI,EAAGA,EAAIC,EAAQD,IAC1BH,EAAWG,GAAKoB,EAAUhB,EAASJ,GAG/BA,EAAIe,EAAStD,KAAKG,cACpBiC,EAAWG,IAAMA,GAAKe,EAAStD,KAAKG,cAIlCoC,EAAIC,EAASe,EAAUvD,KAAKG,cAC9BiC,EAAWG,KAAOC,EAASD,IAAMgB,EAAUvD,KAAKG,cAKtD,OAAOsD,CACT,GAEA,kBAGA,SAAK3C,GACH,IAAM8C,EAAS5D,KAAKC,SAAS4D,qBAM7B,OAJAD,EAAO9C,OAASA,EAChB8C,EAAOE,QAAQ9D,KAAKC,SAAS8D,aAC7BH,EAAOR,QAEAQ,CACT,GAEA,oBAUA,SAAO9C,GAAuE,IAAlDkD,EAAe,UAAH,6CAAG,YACnCC,EAAWjE,KAAKkE,YAAYpD,GAC5BqD,EAAWnE,KAAKoE,cAAcH,EAAUnD,EAAOqB,iBAAkBrB,EAAOf,YACxEsE,EAAY,IAAIrD,KAAK,CAACmD,GAAW,CAAEH,KAAAA,IAEzC,MAAO,CACLM,KAAMD,EACNE,IAAKvE,KAAKwE,WAAWH,GACrBI,QAASzE,KAAK0E,oBAAoBL,GAEtC,GAEA,sBAMA,SAASC,GAA6D,IAAjDK,EAAmB,UAAH,6CAAG,UAChCC,EAAIC,SAASC,cAAc,KAOjC,OALAF,EAAEG,MAAMC,QAAU,OAClBJ,EAAEK,KAAOjF,KAAKwE,WAAWF,GACzBM,EAAEM,SAAW,GAAH,OAAMP,EAAQ,YAAIL,EAAKN,KAAKtC,MAAM,KAAK,IACjDkD,EAAEO,QAEKP,CACT,GAEA,0BAOA,SAAgBQ,GACd,OAAOpF,KAAKqF,oBAAiBC,EAAYF,GAC3C,GAEA,mBAGA,WAEE,OADApF,KAAKC,SAASsF,QACPvF,IACT,GAEA,0BAKA,SAAqB4B,GACnB,OAAOoB,KAAKwC,IAAG,MAARxC,KAAI,EAAQpB,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAO2E,QAAQ,KAC5D,GAEA,kCAKA,SAA6B7D,GAC3B,OAAOoB,KAAKwC,IAAG,MAARxC,KAAI,EAAQpB,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAOqB,gBAAgB,KACpE,GAEA,0BAKA,SAAqBP,GACnB,OAAOA,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAO0B,MAAM,IAAEkD,QAAO,SAACd,EAAGe,GAAC,OAAKf,EAAIe,CAAC,GAAE,EACxE,GAEA,0BAKA,WACE,MAAO,iBAAkBvF,QAAU,uBAAwBA,QAAU,oBAAqBA,MAC5F,GAEA,2BAOA,SAAsBU,EAAsB8E,EAAuB7F,GACjE,IAEM8F,EADiBC,EACJF,EAIbG,EALiBD,EAKDhF,EAAO0B,OACvBwD,EAFkB,GAEiBD,EAEnC9E,EAAc,IAAIgF,YALD,EAK8BD,GAC/CE,EAAO,IAAIC,SAASlF,GAgB1B,OAdAjB,KAAKoG,aAAaF,EAAM,EAAG,QAC3BA,EAAKG,UAAU,EAAGL,GAAgB,GAClChG,KAAKoG,aAAaF,EAAM,EAAG,QAC3BlG,KAAKoG,aAAaF,EAAM,GAAI,QAC5BA,EAAKG,UAAU,GAAI,IAAI,GACvBH,EAAKI,UAAU,GAAI,GAAG,GACtBJ,EAAKI,UAAU,GAAIV,GAAe,GAClCM,EAAKG,UAAU,GAAItG,GAAY,GAC/BmG,EAAKG,UAAU,GAAItG,EAAa8F,GAAY,GAC5CK,EAAKI,UAAU,GAAIT,GAAY,GAC/BK,EAAKI,UAAU,GAtBE,IAsBY,GAC7BtG,KAAKoG,aAAaF,EAAM,GAAI,QAC5BA,EAAKG,UAAU,GAAIN,GAAe,GAE3B/F,KAAKuG,iBAAiBL,EAAMpF,EAAQ0F,GAC7C,GAEA,8BAKA,SAAyBrC,EAAoBrD,EAAsB6B,GACjE,IAAK,IAAIJ,EAAI,EAAGA,EAAIzB,EAAO0B,OAAQD,IAAKI,GAAU,EAAG,CACnD,IAAM8D,EAAMzD,KAAKwC,KAAK,EAAGxC,KAAK0D,IAAI,EAAG5F,EAAOyB,KAC5C4B,EAASwC,SAAShE,EAAQ8D,EAAM,EAAU,MAANA,EAAqB,MAANA,GAAc,EACnE,CAEA,OAAOtC,CACT,GAEA,0BAKA,SAAqBA,EAAoBxB,EAAgBiE,GACvD,IAAK,IAAIrE,EAAI,EAAGA,EAALine truncated
+{"version":3,"file":"crunker.esm.js","mappings":"AACA,IAAIA,EAAsB,CCA1BA,EAAwB,CAACC,EAASC,KACjC,IAAI,IAAIC,KAAOD,EACXF,EAAoBI,EAAEF,EAAYC,KAASH,EAAoBI,EAAEH,EAASE,IAC5EE,OAAOC,eAAeL,EAASE,EAAK,CAAEI,YAAY,EAAMC,IAAKN,EAAWC,IAE1E,ECNDH,EAAwB,CAACS,EAAKC,IAAUL,OAAOM,UAAUC,eAAeC,KAAKJ,EAAKC,I,8jDCoBlF,IAGqBI,EAAO,WAU1B,aAAqE,6DAAJ,CAAC,EAApDC,EAAU,EAAVA,WAAU,UACtBC,KAAKC,SAAWD,KAAKE,eAAeH,GAEpCA,IAAAA,EAAeC,KAAKC,SAASF,YAE7BC,KAAKG,YAAcJ,CACrB,C,UAoZC,O,EAlZD,E,EAAA,6BAKA,WAAkE,IAA3CA,EAAqB,UAAH,6CAAG,MAE1C,OADAK,OAAOC,aAAeD,OAAOC,cAAiBD,OAAeE,oBAAuBF,OAAeG,gBAC5F,IAAIF,aAAa,CAAEN,WAAAA,GAC5B,GAEA,mBAIA,WACE,OAAOC,KAAKC,QACd,GAEA,wBAGAO,iBAA4E,kCAAxDC,EAAS,yBAATA,EAAS,gBAC3B,aAAaC,QAAQC,IACnBF,EAAUG,KAAIJ,eAAOK,GACnB,IAAIC,EAkBJ,OAfEA,EADED,aAAoBE,MAAQF,aAAoBG,WACnCH,EAASI,oBAETC,MAAML,GAAUM,MAAK,SAACC,GASnC,OARIA,EAASC,QAAQC,IAAI,kBAAoBF,EAASC,QAAQ7B,IAAI,gBAAiB+B,SAAS,WAC1FC,QAAQC,KAAK,oEAAD,OAERL,EAASC,QAAQ7B,IAAI,gBAAiBkC,MAAM,KAAK,GAAE,sDACPb,EAAQ,OAInDO,EAASH,aAClB,UAGW,EAAKhB,SAAS0B,gBAAgBb,EAC7C,IAEJ,GAEA,wBAOA,SAAWc,GACT,IAAMC,EAAS7B,KAAKC,SAAS6B,aAC3B9B,KAAK+B,qBAAqBH,GAC1B5B,KAAKG,YAAcH,KAAKgC,aAAaJ,GACrC5B,KAAKG,aAgBP,OAbAyB,EAAQK,SAAQ,SAACnB,GACf,IAAK,IAAIoB,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IAAiB,CAIpF,IAHA,IAAME,EAAaP,EAAOQ,eAAeH,GACnCI,EAAaxB,EAAOuB,eAAeH,GAEhCK,EAAIzB,EAAOuB,eAAeH,GAAeM,OAAS,EAAGD,GAAK,EAAGA,IACpEH,EAAWG,IAAMD,EAAWC,GAG9BV,EAAOQ,eAAeH,GAAeO,IAAIL,EAC3C,CACF,IAEOP,CACT,GAEA,yBAOA,SAAYD,GACV,IAAMC,EAAS7B,KAAKC,SAAS6B,aAC3B9B,KAAK+B,qBAAqBH,GAC1B5B,KAAK0C,aAAad,GAClB5B,KAAKG,aAEHwC,EAAS,EAUb,OARAf,EAAQK,SAAQ,SAACnB,GACf,IAAK,IAAIoB,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IACnEL,EAAOQ,eAAeH,GAAeO,IAAI3B,EAAOuB,eAAeH,GAAgBS,GAGjFA,GAAU7B,EAAO0B,MACnB,IAEOX,CACT,GAEA,sBAUA,SAASf,GAA6E,IAAxD8B,EAAmB,UAAH,6CAAG,EAAGC,EAAkB,UAAH,6CAAG,EACpE,GAAgB,IAAZA,EAAe,OAAO/B,EAE1B,GAAI8B,EAAW,EAAG,MAAM,IAAIE,MAAM,8DAClC,GAAID,EAAU,EAAG,MAAM,IAAIC,MAAM,6DAQjC,IANA,IAAMC,EAAgB/C,KAAKC,SAAS6B,aAClChB,EAAOqB,iBACPa,KAAKC,KAAKnC,EAAO0B,OAASK,EAAU/B,EAAOf,YAC3Ce,EAAOf,YAGAmC,EAAgB,EAAGA,EAAgBpB,EAAOqB,iBAAkBD,IAAiB,CACpF,IAAMgB,EAAcpC,EAAOuB,eAAeH,GAC1Ca,EACGV,eAAeH,GACfO,IAAIS,EAAYC,SAAS,EAAGH,KAAKC,KAAKL,EAAW9B,EAAOf,YAAc,GAAI,GAE7EgD,EACGV,eAAeH,GACfO,IACCS,EAAYC,SAASH,KAAKC,KAAKL,EAAW9B,EAAOf,YAAc,EAAGgD,EAAcP,OAAS,GACzFQ,KAAKC,MAAML,EAAWC,GAAW/B,EAAOf,YAE9C,CAEA,OAAOgD,CACT,GAEA,wBASA,SAAWjC,EAAqBsC,EAAeC,GAAmE,IAAtDC,EAAiB,UAAH,6CAAG,EAAGC,EAAkB,UAAH,6CAAG,EAChG,GAAIH,GAASC,EAAK,MAAM,IAAIP,MAAM,6EAMlC,IAJA,IAAMN,EAASQ,KAAKQ,OAAOH,EAAMD,GAASpD,KAAKG,aACzCwC,EAASK,KAAKQ,MAAMJ,EAAQpD,KAAKG,aACjCsD,EAAYzD,KAAKC,SAAS6B,aAAahB,EAAOqB,iBAAkBK,EAAQxC,KAAKG,aAE1EuD,EAAU,EAAGA,EAAU5C,EAAOqB,iBAAkBuB,IAIvD,IAHA,IAAMC,EAAY7C,EAAOuB,eAAeqB,GAClCtB,EAAaqB,EAAUpB,eAAeqB,GAEnCnB,EAAI,EAAGA,EAAIC,EAAQD,IAC1BH,EAAWG,GAAKoB,EAAUhB,EAASJ,GAG/BA,EAAIe,EAAStD,KAAKG,cACpBiC,EAAWG,IAAMA,GAAKe,EAAStD,KAAKG,cAIlCoC,EAAIC,EAASe,EAAUvD,KAAKG,cAC9BiC,EAAWG,KAAOC,EAASD,IAAMgB,EAAUvD,KAAKG,cAKtD,OAAOsD,CACT,GAEA,kBAGA,SAAK3C,GACH,IAAM8C,EAAS5D,KAAKC,SAAS4D,qBAM7B,OAJAD,EAAO9C,OAASA,EAChB8C,EAAOE,QAAQ9D,KAAKC,SAAS8D,aAC7BH,EAAOR,QAEAQ,CACT,GAEA,oBAUA,SAAO9C,GAAuE,IAAlDkD,EAAe,UAAH,6CAAG,YACnCC,EAAWjE,KAAKkE,YAAYpD,GAC5BqD,EAAWnE,KAAKoE,cAAcH,EAAUnD,EAAOqB,iBAAkBrB,EAAOf,YACxEsE,EAAY,IAAIrD,KAAK,CAACmD,GAAW,CAAEH,KAAAA,IAEzC,MAAO,CACLM,KAAMD,EACNE,IAAKvE,KAAKwE,WAAWH,GACrBI,QAASzE,KAAK0E,oBAAoBL,GAEtC,GAEA,sBAMA,SAASC,GAA6D,IAAjDK,EAAmB,UAAH,6CAAG,UAChCC,EAAIC,SAASC,cAAc,KAOjC,OALAF,EAAEG,MAAMC,QAAU,OAClBJ,EAAEK,KAAOjF,KAAKwE,WAAWF,GACzBM,EAAEM,SAAW,GAAH,OAAMP,EAAQ,YAAIL,EAAKN,KAAKtC,MAAM,KAAK,IACjDkD,EAAEO,QAEKP,CACT,GAEA,0BAOA,SAAgBQ,GACd,OAAOpF,KAAKqF,oBAAiBC,EAAYF,GAC3C,GAEA,mBAGA,WAEE,OADApF,KAAKC,SAASsF,QACPvF,IACT,GAEA,0BAKA,SAAqB4B,GACnB,OAAOoB,KAAKwC,IAAG,MAARxC,KAAI,EAAQpB,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAO2E,QAAQ,KAC5D,GAEA,kCAKA,SAA6B7D,GAC3B,OAAOoB,KAAKwC,IAAG,MAARxC,KAAI,EAAQpB,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAOqB,gBAAgB,KACpE,GAEA,0BAKA,SAAqBP,GACnB,OAAOA,EAAQhB,KAAI,SAACE,GAAM,OAAKA,EAAO0B,MAAM,IAAEkD,QAAO,SAACd,EAAGe,GAAC,OAAKf,EAAIe,CAAC,GAAE,EACxE,GAEA,0BAKA,WACE,MAAO,iBAAkBvF,QAAU,uBAAwBA,QAAU,oBAAqBA,MAC5F,GAEA,2BAOA,SAAsBU,EAAsB8E,EAAuB7F,GACjE,IAEM8F,EADiBC,EACJF,EAIbG,EALiBD,EAKDhF,EAAO0B,OACvBwD,EAFkB,GAEiBD,EAEnC9E,EAAc,IAAIgF,YALD,EAK8BD,GAC/CE,EAAO,IAAIC,SAASlF,GAgB1B,OAdAjB,KAAKoG,aAAaF,EAAM,EAAG,QAC3BA,EAAKG,UAAU,EAAGL,GAAgB,GAClChG,KAAKoG,aAAaF,EAAM,EAAG,QAC3BlG,KAAKoG,aAAaF,EAAM,GAAI,QAC5BA,EAAKG,UAAU,GAAI,IAAI,GACvBH,EAAKI,UAAU,GAAI,GAAG,GACtBJ,EAAKI,UAAU,GAAIV,GAAe,GAClCM,EAAKG,UAAU,GAAItG,GAAY,GAC/BmG,EAAKG,UAAU,GAAItG,EAAa8F,GAAY,GAC5CK,EAAKI,UAAU,GAAIT,GAAY,GAC/BK,EAAKI,UAAU,GAtBE,IAsBY,GAC7BtG,KAAKoG,aAAaF,EAAM,GAAI,QAC5BA,EAAKG,UAAU,GAAIN,GAAe,GAE3B/F,KAAKuG,iBAAiBL,EAAMpF,EAAQ0F,GAC7C,GAEA,8BAKA,SAAyBrC,EAAoBrD,EAAsB6B,GACjE,IAAK,IAAIJ,EAAI,EAAGA,EAAIzB,EAAO0B,OAAQD,IAAKI,GAAU,EAAG,CACnD,IAAM8D,EAAMzD,KAAKwC,KAAK,EAAGxC,KAAK0D,IAAI,EAAG5F,EAAOyB,KAC5C4B,EAASwC,SAAShE,EAAQ8D,EAAM,EAAU,MAANA,EAAqB,MAANA,GAAc,EACnE,CAEA,OAAOtC,CACT,GAEA,0BAKA,SAAqBA,EAAoBxB,EAAgBiE,GACvD,IAAK,IAAIrE,EAAI,EAAGA,EAALine truncated

system/tts_generator/tts_generator.html

@@ -532,6 +532,7 @@
                 <option value="en" selected>English</option>
                 <option value="fr">French</option>
                 <option value="de">German</option>
+                <option value="hi">Hindi</option>
                 <option value="hu">Hungarian</option>
                 <option value="it">Italian</option>
                 <option value="ja">Japanese</option>
@@ -921,18 +922,21 @@
             const playType = document.getElementById('playType').value; // Play type (chunk or stream)
             playModeValue = document.getElementById('playMode').value;
 
-            let cleanedText = textInput.replace(/ \- | \– /g, ' ');
+            let cleanedText = textInput.replace(/ \\- | \\– /g, ' ');
             cleanedText = cleanedText.replace(/%/g, " percent");
-            cleanedText = cleanedText.replace(/\.\s'/g, ".'");
+            cleanedText = cleanedText.replace(/\\.\\s'/g, ".'");
-            cleanedText = cleanedText.replace(/(?<!\w)[']+(?!\w)/g, '');
+            cleanedText = cleanedText.replace(/(?<!\\w)\['\]+(?!\\w)/g, '');
             cleanedText = cleanedText.replace(/"/g, "'");
-            cleanedText = cleanedText.replace(/[^a-zA-Z0-9\s\.,;:!?\-\'"À-ÿ\u0400-\u04FF\u0150\u0151\u0170\u0171]\$/g, '');
+            cleanedText = cleanedText.replace(/[^a-zA-Z0-9\s.,;:!?\-'"$À-ÿ\u0400-\u04FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u2026\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]/g, '');
-            // Replace dash followed by space with comma and space
+            cleanedText = cleanedText.replace(/-\\s/g, '; ');
-            cleanedText = cleanedText.replace(/-\s/g, '; ');
-            // Replace double dash "--" with semicolon "; "
             cleanedText = cleanedText.replace(/--/g, '; ');
-
+            // Split by Fullwidth/CJK punctuation regardless of if there are spaces behind them
-            let sentences = cleanedText.split(/(?<=[.!?:])\s+/);
+            // Please test if it works properly. In my test they it should be 
+            // (?<=[.!?:])\s+ instead of 
+            // (?<=\[.!?:\])\\s+ 
+            // so I don't know whats going on here
+            // let sentences = cleanedText.split(/(?<=[\[.!?:\]])\s+|(?<=[\u3002\uFF01\uFF1A\uFF1F])/);
+            let sentences = cleanedText.split(/(?<=[\[.!?:\]]['"\u2018\u2019\u201C\u201D]*)\s+|(?<=[\u3002\uFF01\uFF1A\uFF1F])/);
 
             if (!sentences) {
                 console.error("No sentences found in the input text.");
@@ -2208,4 +2212,4 @@
 </script>
 </body>
 
-</html>
+</html>

tts_server.py

@@ -476,6 +476,11 @@ async def deepspeed(request: Request, new_deepspeed_value: bool):
 ########################
 #### TTS GENERATION ####
 ########################
+debug_generate_audio = False
+tts_stop_generation = False # Called to stop generation of the current text at whatever stage its at. currently only set for streaming.
+tts_generation_lock = False # Tracks locking of the generaetion process.
+tts_narrator_generatingtts = False # Tracks if the current tts processes are narrator based, to avoid moving model on each chunk of text generated.
+# Not in use tts_queue_behaviour = False # True is to queue the current generation request. False is to cancel the current generation and start a new one.
 
 # TTS VOICE GENERATION METHODS (called from voice_preview and output_modifer)
 async def generate_audio(text, voice, language, temperature, repetition_penalty, output_file, streaming=False):
@@ -486,9 +491,11 @@ async def generate_audio(text, voice, language, temperature, repetition_penalty,
         return response
     async for _ in response:
         pass
-    
+
+
 async def generate_audio_internal(text, voice, language, temperature, repetition_penalty, output_file, streaming):
-    global model
+    global model, tts_stop_generation, tts_generation_lock
+    tts_generation_lock = True
     if params["low_vram"] and device == "cpu":
         await switch_device()
     generate_start_time = time.time()  # Record the start time of generating TTS
@@ -539,6 +546,12 @@ async def generate_audio_internal(text, voice, language, temperature, repetition
             yield wav_buf.read()
 
             for i, chunk in enumerate(output):
+                if tts_stop_generation:
+                    print(f"[{params['branding']}TTSGen] Stopping audio generation.")
+                    file_chunks.clear()  # Clear the file_chunks list
+                    tts_stop_generation = False
+                    tts_generation_lock = False
+                    break
                 file_chunks.append(chunk)
                 if isinstance(chunk, list):
                     chunk = torch.cat(chunk, dim=0)
@@ -547,6 +560,7 @@ async def generate_audio_internal(text, voice, language, temperature, repetition
                 chunk = np.clip(chunk, -1, 1)
                 chunk = (chunk * 32767).astype(np.int16)
                 yield chunk.tobytes()
+                print(f"[{params['branding']}Debug] Stream audio generation: Yielded audio chunk {i}.") if debug_generate_audio else None   
         else:
             # Non-streaming-specific operation
             torchaudio.save(output_file, torch.tensor(output["wav"]).unsqueeze(0), 24000)
@@ -565,9 +579,9 @@ async def generate_audio_internal(text, voice, language, temperature, repetition
             file_path=output_file,
             speaker_wav=[f"{this_dir}/voices/{voice}"],
             language=language,
-            temperature=temperature,
+            temperature=float(temperature),
-            length_penalty=model.config.length_penalty,
+            length_penalty=float(model.config.length_penalty),
-            repetition_penalty=repetition_penalty,
+            repetition_penalty=float(repetition_penalty),
             top_k=model.config.top_k,
             top_p=model.config.top_p,
         )
@@ -589,12 +603,11 @@ async def generate_audio_internal(text, voice, language, temperature, repetition
     # Print Generation time and settings
     generate_end_time = time.time()  # Record the end time to generate TTS
     generate_elapsed_time = generate_end_time - generate_start_time
-    print(
+    print(f"[{params['branding']}TTSGen] \033[93m{generate_elapsed_time:.2f} seconds. \033[94mLowVRAM: \033[33m{params['low_vram']} \033[94mDeepSpeed: \033[33m{params['deepspeed_activate']}\033[0m")
-        f"[{params['branding']}TTSGen] \033[93m{generate_elapsed_time:.2f} seconds. \033[94mLowVRAM: \033[33m{params['low_vram']} \033[94mDeepSpeed: \033[33m{params['deepspeed_activate']}\033[0m"
-    )
     # Move model back to cpu system ram if needed.
-    if params["low_vram"] and device == "cuda":
+    if params["low_vram"] and device == "cuda" and tts_narrator_generatingtts == False:
         await switch_device()
+    tts_generation_lock = False
     return
 
 
@@ -854,6 +867,13 @@ async def tts_generate_streaming(request: Request, text: str = Form(...), voice:
         print(f"An error occurred: {e}")
         return JSONResponse(content={"error": "An error occurred"}, status_code=500)
 
+@app.put("/api/stop-generation")
+async def stop_generation_endpoint():
+    global tts_stop_generation, tts_generation_lock
+    if tts_generation_lock and not tts_stop_generation:
+        tts_stop_generation = True
+    return {"message": "Generation stopped"}
+
 ##############################
 #### Standard Generation ####
 ##############################
@@ -890,7 +910,7 @@ class JSONInput(BaseModel):
     narrator_enabled: bool = Field(..., description="narrator_enabled needs to be true or false.")
     narrator_voice_gen: str = Field(..., pattern="^.*\.wav$", description="narrator_voice_gen needs to be the name of a wav file e.g. mysample.wav.")
     text_not_inside: str = Field(..., pattern="^(character|narrator)$", description="text_not_inside needs to be 'character' or 'narrator'.")
-    language: str = Field(..., pattern="^(ar|zh-cn|cs|nl|en|fr|de|hu|it|ja|ko|pl|pt|ru|es|tr)$", description="language needs to be one of the following ar|zh-cn|cs|nl|en|fr|de|hu|it|ja|ko|pl|pt|ru|es|tr.")
+    language: str = Field(..., pattern="^(ar|zh-cn|cs|nl|en|fr|de|hu|hi|it|ja|ko|pl|pt|ru|es|tr)$", 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.")
     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.")
@@ -1029,12 +1049,17 @@ async def tts_generate(
             "autoplay_volume": autoplay_volume,
             "streaming": streaming,
         }
+        global tts_narrator_generatingtts
         JSONresult = TTSGenerator.validate_json_input(json_input_data)
         if JSONresult is None:
             pass
         else:
+            print(f"[{params['branding']}API] \033[91mError with API request:\033[0m", JSONresult)
             return JSONResponse(content={"error": JSONresult}, status_code=400)
         if narrator_enabled:
+            if params["low_vram"] and device == "cpu":
+                await switch_device()
+            tts_narrator_generatingtts = True
             processed_parts = process_text(text_input)
             audio_files_all_paragraphs = []
             for part_type, part in processed_parts:
@@ -1054,9 +1079,11 @@ async def tts_generate(
                     voice_description = "\033[36mCharacter (Text-not-inside)\033[0m" if text_not_inside == "character" else "\033[92mNarrator (Text-not-inside)\033[0m"
                     print(f"[{params['branding']}TTSGen] {voice_description}")
                 # Replace multiple exclamation marks, question marks, or other punctuation with a single instance
-                cleaned_part = re.sub(r'([!?.])\1+', r'\1', part)
+                cleaned_part = re.sub(r'([!?.\u3002\uFF1F\uFF01\uFF0C])\1+', r'\1', part)
+                # Replace "Chinese ellipsis" with a single dot
+                cleaned_part = re.sub(r"\u2026{1,2}", ". ", cleaned_part)
                 # Further clean to remove any other unwanted characters
-                cleaned_part = re.sub(r'[^a-zA-Z0-9\s\.,;:!?\-\'"\u0400-\u04FFÀ-ÿ\u0150\u0151\u0170\u0171]\$', '', cleaned_part)
+                cleaned_part = re.sub(r'[^a-zA-Z0-9\s.,;:!?\-\'"$\u0400-\u04FF\u00C0-\u00FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]', '', cleaned_part)
                 # Remove all newline characters (single or multiple)
                 cleaned_part = re.sub(r'\n+', ' ', cleaned_part)
                 output_file = this_dir / "outputs" / f"{output_file_name}_{uuid.uuid4()}_{int(time.time())}.wav"
@@ -1066,6 +1093,10 @@ async def tts_generate(
                 audio_files_all_paragraphs.append(audio_path)
             # Combine audio files across paragraphs
             output_file_path, output_file_url, output_cache_url = combine(output_file_timestamp, output_file_name, audio_files_all_paragraphs)
+            tts_narrator_generatingtts = False
+            # Move model back to cpu system ram if needed.
+            if params["low_vram"] and device == "cuda":
+                await switch_device()
         else:
             if output_file_timestamp:
                 timestamp = int(time.time())
@@ -1085,15 +1116,19 @@ async def tts_generate(
                 output_cache_url = f'http://{params["ip_address"]}:{params["port_number"]}/audiocache/{output_file_name}.wav'
             if text_filtering == "html":
                 cleaned_string = html.unescape(standard_filtering(text_input))
-                cleaned_string = re.sub(r'([!?.])\1+', r'\1', text_input)
+                cleaned_string = re.sub(r'([!?.\u3002\uFF1F\uFF01\uFF0C])\1+', r'\1', text_input)
+                # Replace "Chinese ellipsis" with a single dot
+                cleaned_string = re.sub(r"\u2026{1,2}", ". ", cleaned_string)
                 # Further clean to remove any other unwanted characters
-                cleaned_string = re.sub(r'[^a-zA-Z0-9\s\.,;:!?\-\'"\u0400-\u04FFÀ-ÿ\u0150\u0151\u0170\u0171]\$', '', cleaned_string)
+                cleaned_string = re.sub(r'[^a-zA-Z0-9\s.,;:!?\-\'"$\u0400-\u04FF\u00C0-\u00FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]', '', cleaned_string)
                 # Remove all newline characters (single or multiple)
                 cleaned_string = re.sub(r'\n+', ' ', cleaned_string)
             elif text_filtering == "standard":
-                cleaned_string = re.sub(r'([!?.])\1+', r'\1', text_input)
+                cleaned_string = re.sub(r'([!?.\u3002\uFF1F\uFF01\uFF0C])\1+', r'\1', text_input)
+                # Replace "Chinese ellipsis" with a single dot
+                cleaned_string = re.sub(r"\u2026{1,2}", ". ", cleaned_string)
                 # Further clean to remove any other unwanted characters
-                cleaned_string = re.sub(r'[^a-zA-Z0-9\s\.,;:!?\-\'"\u0400-\u04FFÀ-ÿ\u0150\u0151\u0170\u0171]\$', '', cleaned_string)
+                cleaned_string = re.sub(r'[^a-zA-Z0-9\s.,;:!?\-\'"$\u0400-\u04FF\u00C0-\u00FF\u0150\u0151\u0170\u0171\u0900-\u097F\u2018\u2019\u201C\u201D\u3001\u3002\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\uF900-\uFAFF\u0600-\u06FF\u0750-\u077F\uFB50-\uFDFF\uFE70-\uFEFF\uAC00-\uD7A3\u1100-\u11FF\u3130-\u318F\uFF01\uFF0c\uFF1A\uFF1B\uFF1F]', '', cleaned_string)
                 # Remove all newline characters (single or multiple)
                 cleaned_string = re.sub(r'\n+', ' ', cleaned_string)
             else: