Erog0291/stew-yanga

Stew Yanga v0.1.1 - Nigerian Multilingual Speech Intelligence

Stew Yanga is Nigeria's unified speech model: it hears Igbo, Yoruba, Hausa, Pidgin and Naija English. v0.1.1 ears trained on full FLEURS Nigerian data with noise robustness.

Download the model

pip install -U huggingface_hub
huggingface-cli download Erog0291/stew-audio-0.1-asr-ig
huggingface-cli download Erog0291/stew-audio-0.1-asr-yo
huggingface-cli download Erog0291/stew-audio-0.1-asr-ha

Use it (ears)

from stew_yanga import StewYanga
sy = StewYanga()
print(sy.transcribe("my_clip.wav", "ig"))   # igbo, yoruba, hausa

Voice (TTS)

Natural Nigerian voices via the YarnGPT voice layer - recipe and samples on the umbrella repo: https://huggingface.co/Erog0291/stew-audio-0.1

Benchmarks (WER, lower is better)

Igbo 53.5 | Yoruba 61.0 | Hausa 38.7 (vs generic baseline 193.8 / 100.4 / 127.2)

Test live: https://huggingface.co/spaces/Erog0291/stew-audio-0.1