Lancercmd/MoeGoe

Executable file for VITS inference, repacked as a FastAPI application.

MoeGoe Repacked

Repacked as a FastAPI application. api.py is the entry point.

The original README.md is here.

1. Put your checkpoints and config files in the statics directory with specific names.

2. Add them to the models varieable (here) with the following format:

models = {
    "hamidashi": ["hamidashi_604_epochs.pth", "hamidashi_config.json"],
    "yuzusoft": ["yuzusoft_365_epochs.pth", "yuzusoft_config.json"],
}

3. Create a subclass of BaseModel and its implementation using your models as Hamidashi or Yuzusoft.

4. Put your Model.syn() function before or after the general syn() function (here).

5. Configure the listening host and port here.

6. Run the application.

Usage

The application will run the sample for audition at startup.

Use http://localhost:10721 to access the application by default.

• /text

  text to synthesize, return the synthesized WAV file directly.

  http://localhost:10721/让宁宁说こんにちは -> 宁宁: こんにちは

  The application will use the file that is already existing/synthesized in the output directory as cache.

• /text?local

  text to synthesize, return the local absolute path of synthesized WAV file.

  http://localhost:10721/让芳乃说こんにちは?local -> ./.../filename.wav

  Avaliable only when accessed by localhost. Act as /text otherwise.

• /text?remote

  text to synthesize, return the remote path snippet of synthesized WAV file.

  http://localhost:10721/让妃爱说こんにちは?remote -> /output/filename.wav

• /output/filename

  filename that is already synthesized, return the synthesized WAV file directly.

  Use /text?remote to get the remote path snippet.

  http://localhost:10721/output/filename.wav -> 妃爱: こんにちは