YorubaVoice

YorubaVoice is an open speech-to-text model and dataset for Yoruba, built with community contributors.

Voice interfaces exclude tens of millions of Yoruba speakers because no usable open model exists.

A community recording app collects consented, labelled speech, and a fine-tuned model is released openly with benchmarks.

Developers building voice products for West African markets.

Contributors read prompts in the app, recordings are validated by peers, and validated audio feeds a fine-tuning pipeline with public evaluation.

Expand to Igbo and Hausa, and publish a hosted inference API.