The WAXAL project was built with a focus on the African AI ecosystem, with data collection led by African academic and community organizations and guided by Google experts on data collection practices. The project says partners retain ownership of the collected data and that all datasets are intended to be openly available for the broader community.
Partners named in the source include Makerere University, which collected ASR and/or TTS data for nine different languages, and the University of Ghana, which focused on eight languages using the ASR image-prompted data collection methodology. Digital Umuganda, in partnership with Addis Ababa University, led ASR collection for several regional languages.
For studio-recorded voices, Media Trust, Loud n Clear and African Institute for Mathematical Sciences Senegal led TTS recordings across various regional languages.
The collaborative framework has already supported follow-on research and publications. One study produced a cookbook for community-driven collection of impaired speech and led to the first open-source dataset for Akan speakers with conditions like cerebral palsy and stammering. The source says this work showed that in-person, image-prompted elicitation is more effective than text-based prompts for these populations.
Another study introduced a 5,000-hour speech corpus for five Ghanaian languages — Akan, Ewe, Dagbani, Dagaare, and Ikposo. The source says it used a controlled crowdsourcing approach to capture natural, spontaneous intonations and to support robust ASR and TTS systems for West Africa.
Further research benchmarked four models — Whisper, XLS-R, MMS, and W2v-BERT — across 13 African languages. The study examined how performance scales with more training data and said the benefits depend on linguistic complexity and domain alignment.
A systematic literature review cataloged 74 datasets across 111 African languages. The review highlighted the need for multi-domain conversational corpora and linguistically informed metrics such as Character Error Rate (CER) for morphologically rich and tonal language contexts.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
