pyVideoTrans

Appearance

TTS Channel: GPT-SoVITS ​

What is GPT-SoVITS ​

GPT-SoVITS is a powerful open-source multilingual text-to-speech (TTS) project supporting Chinese, English, Japanese, Korean, and more. Key features include:

  • Zero-shot TTS: Generate speech from just a 5-second voice sample
  • Few-shot TTS: Fine-tune the model with as little as 1 minute of training data for better voice similarity and naturalness
  • Cross-language support: Synthesize in languages different from the training data — currently English, Japanese, Korean, Cantonese, and Chinese

What's New in GPT-SoVITS v2 ​

  1. Added Korean and Cantonese support
  2. Improved text front-end processing
  3. Expanded base model training data to 5,000 hours
  4. Better quality synthesis from low-quality reference audio (e.g., clipped frequencies, muffled web audio)

GPT-SoVITS User Manual: https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e

Prerequisites ​

RequirementDetails
pyVideoTrans versionv3.66 or higher
GPT-SoVITS versionv2 recommended (natively supported)
HardwareNVIDIA GPU recommended (CPU also works)
Reference audio formatWAV, 5–10 seconds duration

Important: You must use the official GPT-SoVITS package or deploy from official source code. Third-party API implementations may be incompatible and cause errors in pyVideoTrans.

Deployment Steps ​

Step 1: Download the Package ​

Download the official GPT-SoVITS package: https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e/dkxgpiy9zb96hob4

image.png

Step 2: Start the API Service ​

  1. Navigate to the GPT-SoVITS folder
  2. Type cmd in the address bar and press Enter to open a terminal
  3. Run the following command to start the API service:
bash
.\runtime\python api_v2.py

image.png

The API service must be running to use it in pyVideoTrans! The default port is 9880.

Step 3: Configure pyVideoTrans ​

1. Enter the API Address ​

Open pyVideoTrans and go to: Menu → TTS Settings → GPT-SoVITS

Enter in the API text box:

http://127.0.0.1:9880

image.png

Note: The default port is 9880. If you changed the port, update the API address accordingly. For local deployment, use 127.0.0.1 instead of 0.0.0.0.

2. Configure Reference Audio ​

Reference audio must be WAV format, 5–10 seconds duration, or you'll get a 400 Client error.

Reference audio is the source audio GPT-SoVITS uses to clone the voice. Steps:

  1. Prepare a WAV audio file (e.g., 1.wav, 5 seconds long, containing "What a lovely day, pouring rain falling down")
  2. Copy this file to the GPT-SoVITS folder, in the same directory as api_v2.py
  3. In the "Reference Audio" text box, enter the following format:
filename.wav#text corresponding to the audio#language code

image.png

Language codes: zh for Chinese, en for English, ja for Japanese, ko for Korean.

If you store reference audio files in the wavs subfolder within the GPT-SoVITS directory, the path should be:

wavs/1.wav#What a lovely day, pouring rain falling down#zh

image.png

3. Check the api_v2? Option ​

If you're running api_v2.py, make sure the "api_v2?" checkbox is selected.

image.png

4. Test the Connection ​

Click "Test" — if no errors appear, configuration is successful.

Using in pyVideoTrans ​

After configuration, select "GPT-SoVITS" from the "TTS Channel" dropdown on the main screen.

Supported features:

  • Single video dubbing: Select GPT-SoVITS as the TTS channel on the main screen
  • Batch dubbing: Select GPT-SoVITS in the batch dubbing interface
  • Multi-character dubbing: Assign different reference audio to different characters

Reference Audio Requirements ​

ItemRequirement
FormatWAV only
Duration5–10 seconds (ideal); under 3 seconds will cause errors
ContentClear pronunciation, no background noise
LocationSame directory as api_v2.py, or in the wavs subdirectory

Source code verification (_gptsovits.py): Reference audio over 10 seconds is auto-truncated to 9990ms. Under 3 seconds, it's skipped with an error: "Reference audio duration less than 3 seconds."

Troubleshooting ​

1. 404 Error During Testing ​

This is caused by using a third-party package whose API is incompatible with the official version. Download and use the official package.

2. "Remote Computer Actively Refused" or "Check if API Service Is Running" ​

Possible causes:

  • API service not started → Run .\runtime\python api_v2.py first
  • Blocked by firewall → Disable the firewall or add GPT-SoVITS to the whitelist
  • Wrong port number → Confirm you're using the default port 9880

3. 400 Client error ​

The reference audio format or duration doesn't meet requirements. Check:

  • Is it in WAV format?
  • Is the duration between 5–10 seconds?

4. Cannot Connect to GPT-SoVITS ​

Make sure:

  • The GPT-SoVITS API service is running (terminal window must stay open)
  • The API address is correct (http://127.0.0.1:9880)
  • You're using the official version, not a third-party package