f5-ttsgrpc

Utilizes https://github.com/SWivid/F5-TTS,

This is a simple grpc tts server that can run inference for a single voice and prompt at a time. Each generation is returned back as bytes of a wav pcm file.

This is designed to be used in a pipeline with multiple instances ran with different configurations of voices and ports.

The server probably can handle a few client connections at a time. Voice generation is serialized between client connections so the more clients the more lag between generations may be experienced.

Future plans/Wish list (please contribute)

• potentially a docker container with multiple instances
• example integration
• a plexer that wraps all of the python processes and exposes a hgher level abstraction via grpc that will support multiple voices. (go implementation)

GRPC Services

See protobuf definition

User Installation

Download and init repository

git clone https://github.com/beeblebrox/ff5-ttsgrpc
git submodule update --recursive

Setup python 3.10 conda environment, or your favorite env manager

Install requirements

pip install -r requirements.txt

To Run

Edit setup-example.py and move to setup.py (or whatever name you want). Create one for each different instance you need.

ex:

python serve.py setup

python serve.py setup2

Dev Information

Generate grpc

From the base directory after venv is setup and sourced:

python generateproto.py

IDE info

Make sure you add the directory /F5-TTS to your system path or marked as source in project.

PyCharm