Project summary

Kusaal, Farefare, and Buli are spoken by millions in northern Ghana. None of them have working speech AI. No datasets. No ASR. No TTS. This project collects 100 hours of speech per language, transcribes it, and puts everything on HuggingFace for anyone to use.

What are this project's goals? How will you achieve them?

I want to build speech datasets for three northern Ghanaian languages that currently have nothing. 100 hours each, transcribed and reviewed by native speakers. Then train basic ASR and TTS models on top so developers have something to start from.

I am from Bawku. My family knows translators, Bible translation workers, and local radio staff there. That is how I get speakers to show up - through people who already trust each other. For Farefare and Buli I will do the same through church and community networks in Bolgatanga and Sandema. Speakers get paid. That part matters more than it sounds.

How will this funding be used?

Most of the money goes to paying speakers. Without that, people don't show up - I have seen this happen to other researchers.

The rest covers transport between Bawku, Bolgatanga, and Sandema, recording equipment, transcription review, cloud compute for training, and a small stipend so I can work on this full time.

If the full $35,000 isn't raised: $16,000 covers two languages, $8,000 covers Kusaal alone. Either way the output is complete and publishable.

Who is on your team? What's your track record on similar projects?

I am Prince Nasamu Alhassan, a student at University of Ghana and an NLP researcher with GhanaNLP. I built a Kusaal-English MT system - 34,568 sentence pairs, BLEU 27.57 -- and a Kusaal ASR dataset of 81 hours and 30,820 clips. Both are live on HuggingFace with downloads. I ran this work on Kaggle and during the GhanaNLP H200 hackathon. I presented it at a Tech-Talk where Google researchers attended and gave good feedback.

What are the most likely causes and outcomes if this project fails?

The most likely problem is Buli -- it is the least documented of the three and transcription could take longer than expected. If that happens I shift effort to Kusaal and Farefare, which are more de-risked. The fallback funding tiers exist so partial funding still produces something finished.

My time is the other risk. I am a student. I plan collection around vacation periods to manage this.

How much money have you raised in the last 12 months, and from where?

I have not received grant funding for this project yet.