README.md

---
title: TWB Voice TTS Demo
emoji: 🌍
colorFrom: blue
colorTo: gray
sdk: gradio
sdk_version: 5.38.2
app_file: app.py
pinned: false
license: cc-by-4.0
short_description: 'Space to demo TTS models '
---

# TWB Voice - TTS Demo Space

This Gradio demo showcases neural Text-to-Speech models developed within the TWB Voice project by CLEAR Global. Currently it supports **Hausa** and **Kanuri** languages, developed as part of the first phase of the project.

## Features

- **Hausa TTS**: 3 speakers (1 female in Kenanci dialect, 2 male speakers from open.bible)
- **Kanuri TTS**: 1 female speaker  
- High-quality 24kHz audio synthesis
- Based on YourTTS architecture (VITS-based)

## Models

- 🤗 [Hausa Model](https://huggingface.co/CLEAR-Global/TWB-Voice-Hausa-TTS-1.0)
- 🤗 [Kanuri Model](https://huggingface.co/CLEAR-Global/TWB-Voice-Kanuri-TTS-1.0)

## Datasets

- 📊 [Hausa Dataset Samples](https://huggingface.co/datasets/CLEAR-Global/TWB-voice-TTS-Hausa-1.0-sampleset)
- 📊 [Kanuri Dataset Samples](https://huggingface.co/datasets/CLEAR-Global/TWB-voice-TTS-Kanuri-1.0-sampleset)

## Usage

1. Select your desired language (Hausa or Kanuri)
2. Choose a speaker from the available options
3. Enter text or use the example sentences
4. Click "Synthesize Speech" to generate audio

## Technical Details

- **Architecture**: YourTTS (VITS-based) fine-tuned from CML-TTS multilingual checkpoint
- **Sample Rate**: 24 kHz
- **Input**: Lowercase text with preserved diacritics
- **Framework**: Coqui TTS

## License

These models are released under **CC-BY-NC-4.0** license for non-commercial use only.

## Acknowledgments

Created by CLEAR Global with support from the Patrick J. McGovern Foundation.

Special thanks to:
- TWB Voice Project for high-quality voice data
- Idiap Coqui TTS for the YourTTS architecture
- CML-TTS Dataset for the multilingual base model
- Biblica open.bible for additional Hausa recordings