TTS Channel: clone-voice

The model used in this project is xtts_v2 from coqui.ai. The model is released under the Coqui Public Model License 1.0.0. Please follow this license when using the project. Full license text: https://coqui.ai/cpml.txt

What is clone-voice

clone-voice is a voice cloning tool that uses any human voice to synthesize speech from text, or convert one voice into another using that voice's characteristics.

Key features:

Easy to use: Works without an NVIDIA GPU. Download the pre-compiled version, double-click app.exe to open the web interface, and use it with just mouse clicks
Multi-language support: Supports 16 languages including Chinese, English, Japanese, Korean, French, German, Italian, and more
Online recording: Record voice directly from the microphone as reference audio
Quality: Excellent for English; passable for Chinese

Model limitation: The xtts model can only be used for learning and research purposes, not for commercial use.

Prerequisites

Requirement	Details
Operating System	Windows (pre-compiled), Linux, macOS (source deployment)
Disk Space	Pre-compiled: ~4.7GB (main: 1.7G + model: 3G)
Python	Source deployment requires Python 3.9–3.11
GPU	Optional; NVIDIA GPU enables CUDA acceleration

Windows Pre-compiled Version

Step 1: Download

Open the Releases download page and download:

Pre-compiled main file (~1.7GB)
Model file (~3GB)

Step 2: Extract

Extract to a location, e.g., E:/clone-voice

Step 3: Extract the Model

Extract the downloaded model into the tts folder inside the software directory. After extraction:

The tts folder should contain 3 subdirectories.

Step 4: Launch

Double-click app.exe
Wait for the web window to open automatically
Read the text prompts in the cmd window carefully — any errors will be displayed there

On first launch, the model needs to cold-load, which takes some time. Wait for http://127.0.0.1:9988 to appear and the browser page to open, then wait another 2–3 minutes before performing conversions.

Step 5: Use

Text → Voice

Select the "Text → Voice" button
Enter text in the text box, or click to import an SRT subtitle file
Click "Start Now"

Voice → Voice

Select the "Voice → Voice" button
Click or drag the audio file to convert (mp3/wav/flac)
Select the voice to clone from the "Voice File to Use" dropdown
If no suitable option, click "Upload" to select a 5–20s pre-recorded voice file
Or click "Start Recording" to record 5–20s of voice online
Click "Start Now"

If your machine has an NVIDIA GPU with a properly configured CUDA environment, CUDA acceleration will be used automatically.

Source Deployment (Linux / Mac / Windows)

Deployment Steps

Install Python and Git: Python 3.9–3.11 required. Install Git

Clone the source code:

bash

# Create an empty directory, e.g., E:/clone-voice
# Open cmd in that directory and run:
git clone [email protected]:jianchang512/clone-voice.git .

Create a virtual environment:
bash
```
python -m venv venv
```
1
Activate the environment (Windows):
bash
```
E:/clone-voice/venv/scripts/activate
```
1

Install dependencies:

bash

pip install -r requirements.txt --no-deps

Enable CUDA acceleration (optional, requires NVIDIA GPU):

bash

pip uninstall -y torch
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu121

Install FFmpeg:
- Windows: Extract ffmpeg.7z and place ffmpeg.exe in the same directory as app.py
- Linux/macOS: Download from FFmpeg official site and place the ffmpeg executable in the root directory
First run to download model:
bash
```
python code_dev.py
```
1
- Enter y when prompted to accept the license
- Wait for the model download to complete
Start the service:
bash
```
python app.py
```
1
Train the model (optional):
bash
```
python train.py
```
1
Training parameters can be adjusted in param.json.

Using in pyVideoTrans

Start clone-voice's web interface (default: http://127.0.0.1:9988)
Open pyVideoTrans, go to Menu → TTS Settings → clone-voice
Enter the API address: http://127.0.0.1:9988
Click "Test" — if no errors, configuration is successful
Select "clone-voice" from the "TTS Channel" dropdown on the main screen

Reference Audio Requirements

Item	Requirement
Format	WAV, MP3, or FLAC
Duration	5–20 seconds (ideal) for best synthesis quality
Content	Clear, accurate pronunciation, no background noise
Languages	Supports 16 languages

Troubleshooting

1. Model Download Stalls or Fails

The model is hosted on huggingface.co. If the download fails repeatedly, try using a network proxy or mirror.

2. Model Check/Update on Every Launch

Each startup connects to check or update the model. To disable this, edit the dependency package:

Open \venv\Lib\site-packages\TTS\utils\manage.py, around line 389, in the def download_model method, comment out the block from if md5sum is not None: to the else.

3. Stuck After Launch

On first launch, the model needs to cold-load, which takes time. Wait for http://127.0.0.1:9988 to appear and the browser page to open, then wait another 2–3 minutes before performing conversions.

4. cmd Window Freezes, Requires Enter to Continue

Click the icon in the top-left corner of the cmd window, select "Properties", and uncheck "QuickEdit Mode" and "Insert Mode".

5. Voice-to-Voice Thread Startup Failure

First, verify the model is correctly placed (the tts folder should contain 3 subdirectories):

If correctly placed but still failing, download extra-to-tts_cache.zip and copy the 2 extracted files to the tts_cache folder in the software root directory.

If the above doesn't work, configure a stable network proxy.

6. Text Length Exceeds Limit

The text length exceeds the character limit of 182/82 for language

Sentences separated by periods are too long. Break long sentences with periods rather than commas, or manually edit clone/character.json to adjust the limit.

7. Error `symbol not found __svml_cosf8_ha`

Download the DLL from https://www.dll-files.com/svml_dispmd.dll.html and copy it to C:\Windows\System32.

CUDA Acceleration

Installing CUDA Tools

Detailed installation guide

Update your GPU driver to the latest version
Install CUDA Toolkit 11.8
Install cuDNN for CUDA 11.X

Verify Installation

Press Win + R, type cmd, press Enter, then run:

bash

nvcc --version

Confirm version information is displayed:

Then run:

bash

nvidia-smi

Confirm output is displayed and the CUDA version is visible:

If both commands produce output, CUDA is installed correctly and can be used for acceleration. Otherwise, reinstall.

TTS Channel: clone-voice ​

What is clone-voice ​

Prerequisites ​

Windows Pre-compiled Version ​

Step 1: Download ​

Step 2: Extract ​

Step 3: Extract the Model ​

Step 4: Launch ​

Step 5: Use ​

Text → Voice ​

Voice → Voice ​

Source Deployment (Linux / Mac / Windows) ​

Deployment Steps ​

Using in pyVideoTrans ​

Reference Audio Requirements ​

Troubleshooting ​

1. Model Download Stalls or Fails ​

2. Model Check/Update on Every Launch ​

3. Stuck After Launch ​

4. cmd Window Freezes, Requires Enter to Continue ​

5. Voice-to-Voice Thread Startup Failure ​

6. Text Length Exceeds Limit ​

7. Error symbol not found __svml_cosf8_ha ​

CUDA Acceleration ​

Installing CUDA Tools ​

Verify Installation ​

TTS Channel: clone-voice

What is clone-voice

Prerequisites

Windows Pre-compiled Version

Step 1: Download

Step 2: Extract

Step 3: Extract the Model

Step 4: Launch

Step 5: Use

Text → Voice

Voice → Voice

Source Deployment (Linux / Mac / Windows)

Deployment Steps

Using in pyVideoTrans

Reference Audio Requirements

Troubleshooting

1. Model Download Stalls or Fails

2. Model Check/Update on Every Launch

3. Stuck After Launch

4. cmd Window Freezes, Requires Enter to Continue

5. Voice-to-Voice Thread Startup Failure

6. Text Length Exceeds Limit

7. Error `symbol not found __svml_cosf8_ha`

CUDA Acceleration

Installing CUDA Tools

Verify Installation