Contribute to Voice Core

Contribute to Voice Core

Contributors can enhance the Voice Core in two ways:

  • Improve the TTS model

  • Contribute new Voice Data

No contribution to the STT model is needed, as it currently utilizes Azure services. The accuracy is already optimal, and no further enhancements are required.

Improvement on the TTS model

All files related to the model must be submitted. There are several service providers supported by the ecosystem. Below is a list of them and their requirements:

  • sovits.pth: This is your main model file. Ensure it's named "sovits.pth" as specified.

  • reference1.wav: Reference audio file in .wav format. Make sure the name of the file matches the reference in the "config.json" file.

  • gpt.ckpt: The checkpoint file for the model. Confirm it is named "gpt.ckpt".

  • config.json: The configuration file for your model. It must be named "config.json".

Below is the structure for a sample folder submission of a complete model.

AudioModelSubmission/
├── sovits.pth                # The main model file
├── reference1.wav            # Reference audio file (name as per config.json)     
├── gpt.ckpt                  # Checkpoint file for the model
└── config.json               # Configuration file for the model

A sample config.json file is as below.

{
    "refFile": "Olyn.wav",
    "refText": "yet still, I stand, a testiment to the resilience of human spirit"
}

To submit a Voice Model, select Voice Core

Then "I got a Voice Model" and upload the model file following the guidelines provided above.

New Voice Data Contribution

  • Voice data submitted must be legally acquired with the right to share.

  • Voice data acquired must be acquired from authentic sources.

  • Voice data, data should be free from background noise and only voices to be trained are kept in the audio.

  • Voice data must be produced in format of .wav.

Last updated