A simple server for OpenVoice

This project provides an OpenAI compatible text-to-speech server for the OpenVoice project.

OpenVoice is an Open Source project of Text-to-speech model and tools.

Features

• text-to-speech
• tone management
  • create tone by upload wave files
  • list tones
• openai-compatible '/v1/audio/speech' endpoint

Install

You need to build the docker images.

cd src
docker build . -t openvoice-server

The Dockerfile requires the download of all related models, hence the resulting docker image can run in an isolated environment.

Currently this project only supports CUDA.

Run

To run the server, use the docker image you just built:

docker run --rm -p 18080:8080  --device=nvidia.com/gpu=all openvoice-server:latest

Environment variables

• DATABASE_URL The server stores tone information in a SQLite database, you can control the position of the database by specifying the DATABASE_URL environment variable.

Use the API

GET /tones

Get a list of existing tones

POST /tones

Create a tone by uploading a .wav file.

• Request Content-Type: multipart/form-data

Fields in request:

• audiofile

• name

• desc

• Response Content-Type: application/json

See examples/create-tone.sh for the curl command.

GET /tones/{tone_id}

Get the details of a tone by tone_id.

• Response Content-Type: application/json

POST /tones/{tone_id}/{dialect}

Generate voice using the specified tone_id and dialect.

• Request Content-Type: application/json

Fields in request:

• text

• speed[optional]

• Response Content-Type: audio/x-wav

See examples/generate_speech.sh for the curl command

GET /cap

Get a list of supported languages and dialects.

POST /v1/audio/speech

The OpenAI compatible endpoint.

See examples/generate_speech_openai.sh for the curl command.

Note:

1. use dialect as model in request, find dialects from the /cap endpoint.
2. use tone_name as voice in request, find tones from the /tones endpoint.