Languages supported

[!NOTE] All models are from the repository: snakers4/silero-models

Language	Model	Speakers
Russian	v4_ru	5: aidar, baya, kseniya, xenia, eugene
Ukrainian	v4_ua	1: mykyta
Uzbek	v4_uz	1: dilnavoz
English	v3_en	118: en_0, en_1, ..., en_117
Spanish	v3_es	3: es_0, es_1, es_2
French	v3_fr	6: fr_0, fr_1, fr_2, fr_3, fr_4, fr_5
German	v3_de	5: bernd_ungerer, eva_k, friedrich, hokuspokus, karlsson
Tatar	v3_tt	1: dilyara
Mongolian	v3_xal	2: erdni, delghir

Installation via docker

[!IMPORTANT] This requires docker installed and the docker daemon running

docker run --rm -p 8000:8000 twirapp/silero-tts-api-server

Clone the repository:

git clone https://github.com/twirapp/silero-tts-api-server.git && cd silero-tts-api-server

Build docker image:

docker build -f docker/Dockerfile -t silero-tts-api-server .

Run the container:

docker run --rm -p 8000:8000 silero-tts-api-server

Or use docker compose:

docker-compose -f docker/compose.yml up

Installation

[!IMPORTANT] Minimum requirement python 3.9

This project uses rye for dependency management, it assumes you have installed it

1. Clone the repository

   git clone https://github.com/twirapp/silero-tts-api-server.git && cd silero-tts-api-server
   

2. Install dependencies

   This will automatically create the virtual environment in the .venv directory and install the required dependencies

   rye sync
   

   python3 -m venv .venv && source .venv/bin/activate
   

   Install only the required dependencies:

   pip3 install --no-deps -r requirements.lock
   

3. Download silero tts models

   bash ./install_models.sh
   

4. Run the server

   litestar run
   

[!NOTE] The default will be localhost:8000

Documentation

You can view the automatically generated documentation based on OpenAPI at:

Provider	Url
Swagger	https://localhost:8000/schema/
ReDoc	https://localhost:8000/schema/redoc
Stoplight Elements	https://localhost:8000/schema/elements
RepiDoc	https://localhost:8000/schema/repidoc
OpenAPI schema yaml	https://localhost:8000/schema/openapi.yaml
OpenAPI schema json	https://localhost:8000/schema/openapi.json

Endpoints

• GET /generate - Generate audio in wav format from text. Parameters: text speaker sample_rate, pitch, rate
• GET /speakers - Get list of speakers

sample_rate can be set from 8 000, 24 000, 48 000 pitch and rate can be set from 0 to 100

Environment variables

• TEXT_LENGTH_LIMIT - Maximum length of the text to be processed. Default is 930 characters.
• MKL_NUM_THREADS - Number of threads to use for generating audio. Default number of threads: number of CPU cores.

Considerations for the future

This repository is dedicated to twir.app and is designed to meet its requirements.

TwirApp needs to generate audio using the CPU. If support for other devices such as cuda or mps is needed, please open an issue.