Explore our ongoing work, prototypes and the research behind Chichewa speech and conversational technology. Each project is labeled with its current status.
Read the scope and status of each project, including what has not yet been released publicly.
An ongoing Chichewa speech collection effort associated with Masakhane targeting 500 hours of transcribed audio. The work covers recording protocols, speaker metadata, transcription guidance and quality assurance to address the shortage of documented speech resources.
View project detailsWork dating from 2022 combines speech dataset preparation, transcript normalization, model adaptation, training and evaluation, including fine-tuning Whisper large-v3. Earlier public models and code are credited to Dunstan Matekenya.
View project detailsChichewa conversational AI prototypes explore how people can interact with information services through WhatsApp. The work informs our bilingual customer support chatbot offering, emphasizing approved organizational content and human escalation.
View project detailsZaZo AI’s ongoing work is presented alongside earlier research by its technical lead, Dunstan Matekenya. Repository ownership, collaboration arrangements and project statuses remain distinct. A 500-hour target is not the same as a completed dataset.
Talk to us about applied research, collection protocols, Chichewa speech technology or conversational AI.