Speech recognition and natural voices in Yoruba, Hausa, Igbo and Pidgin for Nigerian call centres
Banks, telecoms and insurers in Nigeria run call centres where many callers speak Yoruba, Hausa, Igbo or Pidgin, but their phone menus and speech analytics only work in standard English. Callers press zero and wait, agents rarely log what was said, and quality teams can review fewer than 2% of calls by hand.
Ohun offers streaming speech recognition, transcription and text-to-speech APIs trained on 6,000 hours of consented Nigerian call-centre and broadcast speech, with tone-aware Yoruba and Igbo models. Call centres use it to route calls by spoken intent, transcribe and tag every call, and replace English-only phone menus with natural local-language voices.
Nigerian call centres for banking, telecoms, insurance and public services run more than 40,000 seats, and the same languages are spoken by over 150 million people across West Africa.
Usage-based API pricing per audio minute, from US$0.006 for transcription, with committed monthly contracts and on-premise deployment for regulated customers.
Tolulope BankoleCEO and co-founder
Ibrahim MusaCTO and co-founder
Chiamaka EzeHead of DataTolulope and Ibrahim have built call-centre software before and know how these buyers think. Four paid pilots at pre-seed is a good start. Banks will want on-premise deployment and long security reviews, so I'd expect sales cycles to be slower than the plan suggests.
Restoring Yoruba tone marks before synthesis, and modelling tone in recognition, is the detail that separates this from generic multilingual models. Halving the word error rate since January is real progress. Igbo is clearly behind the other languages for now, and the team says so openly.
Reply from Tolulope Bankole, CEO
Igbo is our main data priority this quarter, with a linguist joining the team in November.
Our microfinance call centre in Ibadan is one of the pilots. Two thirds of our callers speak Yoruba, and once Yoruba prompts replaced the English menu, far fewer callers pressed zero to wait for an agent. The team was careful about consent for its training audio, which our compliance lead checked. Transcripts still need a manual pass when callers read out loan account numbers quickly.
Coming soon. Testimonials are invited by the founder and shown separately from open reviews.
Coming soon. Follow Ohun Labs to hear about product and funding updates when this launches.
