Microsoft Research Africa, Nairobi

Join Mercy Muchai (opens in new tab) on Thursday Nov 5th at 3pm EAT as she presents:

Paza: How Architecture, Language Typology, and Adaptation Shape Speech Recognition in African Languages

Despite rapid and meaningful advances in speech recognition, building robust ASR systems for low-resource languages remains challenging. Key questions persist: how much data should be collected and in what forms, which model architectures offer the best opportunities for training and deployment, and what adaptation strategies are most effective for new domains and languages? 

Drawing on research work that informed the Paza Speech Playbook and other publications, Muchai will present an architecture-aware evaluation of 53 speech models and their accuracy-efficiency trade-offs across 39 African languages. She will also share findings from a systematic study fine-tuning three models from distinct architectural paradigms, showing how architecture, language typology, and adaptation strategy shape recognition performance and real-world deployment.

Mercy Muchai is a Research Engineer at Microsoft Research Africa in Nairobi. She leads the end-to-end AI/ML engineering efforts within the lab. Her work focuses on building globally equitable AI tools and models that can reason over multimodal data, with a particular emphasison improving the linguistic and cultural understanding of Generative AI for underrepresented languages and communities. 

Her recent work includes leading the research and development of automatic speech recognition models for African languages; including PazaBench, the Paza Speech Models and co-creating the Gecko Playbooks – a collection of toolkits for building more equitable AI systems. Muchai is also a Computer Science graduate student at the University of Texas at Austin.