Wolof-TTS: Text-to-Speech Synthesis for Wolof

WolofTTs is an open-source Text-to-Speech (TTS) system designed to generate a synthetic voice speaking in Wolof from any textual input in the same language. This project is built on top of the xTTS v2 model and provides a Dockerized deployment for serving the model and performing inference through an API.

This implementation is based on a fine-tuned version of xTTS v2 by GalsenAI. The goal of this repo is to make Wolof TTS more accessible by providing an easy-to-use deployment container.

Important Notice

xTTS is not for commercial use. It is strictly intended for research, education, and non-profit applications. Any commercial usage is prohibited.

Features

• High-quality Wolof speech synthesis
• Fast and lightweight inference
• Supports text normalization and silence removal
• API integration using Flask and Waitress
• Dockerized for easy deployment

Installation

Prerequisites

• Docker
• CUDA (for GPU acceleration, optional)
• Model Checkpoint (must be downloaded manually, see below)

Download Model Checkpoint

Before building the Docker container, download the model checkpoint manually from:

https://drive.google.com/uc?id=1jsGAMBo354uRhwVKNnuJpLtq2uJYcVHN

or

!gdown 1jsGAMBo354uRhwVKNnuJpLtq2uJYcVHN

Place the downloaded file in the same directory as the Dockerfile.

NB : why we are not downloading the check-point programmatically inside Dockerfile is because of sometime we get blocked due to request limit of google drive so we decide do it manually and we have to wait for a long time.

Build and Run with Docker

Build and run the Docker container:

docker build -t wolof-tts .

docker run --gpus all -p 8080:8080 wolof-tts

Usage

Running the Flask API Server

The Flask server runs automatically inside the container.

Synthesizing Speech via API

You can send a POST request to the /predict endpoint with the text to synthesize:

curl -X POST http://localhost:8080/predict -H "Content-Type: application/json" -d '{"text": "Màngi tuddu Aadama, di baat bii waa Galsen A.I defar ngir wax ak yéen ci wolof!"}' --output output.wav

This will return a output.wav file containing the synthesized speech.

Checking API Health

You can verify the API status using:

curl http://localhost:8080/health

Example Response:

{
 "status": "healthy",
 "device": "cuda:0"
}

API Endpoints

Endpoint	Method	Description
/predict	POST	Generate speech from text
/health	GET	Check service health and device status

Model Details

• Base Model: xTTS V2 (fine-tuned for Wolof by GalsenAI)
• Training Data: Wolof-TTS (https://huggingface.co/datasets/galsenai/anta_women_tts)

Contributing

Contributions are welcome! For major changes, please open an issue first to discuss what you'd like to change.

License

This project is licensed under the MIT License.