Africa TTS

Text-to-speech for 92 African languages and English, with voice cloning, built on Kyutai's Pocket TTS. Every language has a default voice, so all you need is text and a language code.

🎙️ Try it live · 🔊 Samples · 🤗 Model

An ongoing project: native-speaker feedback is what improves it most. See CONTRIBUTING.md.

⚡ Runs on a CPU, no GPU needed

Africa TTS speaks 2.4× faster than real time on 4 CPU cores, and starts streaming audio 0.17 s after it gets the text. A 10-second sentence takes about 4 seconds to generate, and playback can start almost at once. It has 110M parameters, and the on-device bundle (int8) is 209 MB, so it runs on laptops, servers and phones.

Measured on a dedicated 4-vCPU cloud machine (Modal); your laptop may be faster. Measure your own machine with benchmarks/speed.py; details in docs/BENCHMARKS.md.

🎯 Intelligible

On held-out sentences in 47 languages, transcribed by Omnilingual ASR, Africa TTS has a lower character error rate than MMS-TTS (9.0% vs 10.3%, better in 21 of 27 languages) and the 3.3B Sunbird Orpheus (5.9% vs 10.3%), and is close to the 813M OmniVoice (8.8% vs 8.4%) at a seventh of its size: benchmark.

Install

# CPU (laptops, servers): small CPU-only PyTorch
pip install africa-tts --extra-index-url https://download.pytorch.org/whl/cpu

# NVIDIA GPU (Linux): PyPI's PyTorch is the CUDA build
pip install africa-tts

English also needs espeak-ng (apt install espeak-ng / brew install espeak-ng). On macOS and Windows, plain pip install africa-tts gives the CPU build. The first time, it asks before downloading the model (about 440 MB) from Hugging Face, then caches it and works offline. In scripts, approve the download with --yes, AfricaTTS.load(download=True), or africa-tts --download beforehand.

Use

africa-tts "Akwaaba! Yɛfrɛ me Kofi." --lang twi -o twi.wav
africa-tts --list-languages

from africa_tts import AfricaTTS

tts = AfricaTTS.load()                                  # or AfricaTTS.load(device="cuda") on a GPU
audio = tts.generate("Akwaaba! Yɛfrɛ me Kofi.", lang="twi")              # default Twi voice
audio = tts.generate("Hello Ghana!", lang="eng", voice="my_voice.wav")  # clone a 3-10 s clip
tts.save(audio, "out.wav")

• Languages: ISO 639-3 codes (twi, hau, yor, swh, zul, amh, eng, ...); see africa-tts --list-languages.
• Numbers: spell them out ("twenty", not "20").

Run it elsewhere

More

• Benchmarks: speed, and intelligibility per language
• Model, how it was made, data and limitations
• Training recipe · ONNX export · training code
• Status and roadmap · Contributing

License

Code: MIT (LICENSE). Model and default voices: CC-BY-NC-4.0. espeak-ng (English) is GPL-3.0-or-later. Built by AfriSpeech / Ghana NLP.