Testing and fine-tuning environment for speech-to-text providers, tuned towards African languages (starting with Kiswahili, then Chichewa) for Ushahidi's transcription pipeline.
See whisper/README.md for the Whisper setup, baseline WER evaluation, fine-tuning, and model export workflow.