New on Mozilla Data Collective: unscripted, code-switched Sheng speech from Nairobi, recorded on phones.
Safi Data recruited via WhatsApp and Telegram. 70+ contributors recorded natural mobile money conversations and were paid by mobile money. 5 hours in one day, vs. 4 to 8 weeks in a studio.
1-hr sample now (WAV, transcripts, metadata). Full 50 hrs coming. SDL 1.0; no speaker ID or voice cloning.
https://mozilladatacollective.com/datasets/cmt7e4pfp004nnz08j3shtpcx