xtts-api-server 0.2.5

A simple FastAPI server to host XTTSv2

A simple FastAPI Server to run XTTSv2

The project is inspired by silero-api-server repo uses XTTSv2

Created a PR for SillyTavern: here

The TTS module or server can be used any way you wish.

Installation

pip install xtts-api-server

I strongly recommend installing pytorch with CUDA so that the entire process is on the video card

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Starting Server

python -m xtts-api-server will run on default ip and port (0.0.0.0:8020)

usage: xtts-api-server [-h] [-o HOST] [-p PORT] [-sf SPEAKER_FOLDER] [-o OUTPUT]

Run XTTSv2 within a FastAPI application

options:
  -h, --help            show this help message and exit
  -o HOST, --host HOST
  -p PORT, --port PORT
  -sf SPEAKER_FOLDER, --speaker_folder The folder where you get the samples for tts
  -o OUTPUT, --output Output folder

The first time you run or generate, you may need to confirm that you agree to use XTTS. The model will be loaded into memory after the first generation.

API Docs

API Docs can be accessed from http://localhost:8020/docs

Voice Samples

You can find the sample in this repository, also by default samples will be saved to /output/output.wav or you can change this, more details in the API documentation

Selecting Folder

You can change the folders for speakers and the folder for output via the API.

Get Speakers

Once you have at least one file in your speakers folder, you can get its name via API and then you only need to specify the file name.