For many people around the globe, speaking to technology feels effortless. We dictate messages, ask virtual assistants for updates, and navigate the world using voice commands. But this ease quickly vanishes when technology doesn’t understand your language. For hundreds of millions of people—particularly in Sub-Saharan Africa, home to more than 2,000 languages—this remains a daily challenge. One of the biggest obstacles to building effective voice-based tools in the region has been the scarcity of open, high-quality speech data.
This gap is what inspired the creation of WAXAL. Named after the Wolof word meaning “to speak,” WAXAL is a new open dataset developed over the course of three years with a single goal: to support researchers and technologists in building inclusive, language-aware tools for Africa. The dataset spans 21 languages, such as Acholi, Hausa, Luganda, and Yoruba, and features more than 11,000 hours of audio drawn from nearly two million recordings. Within this collection are roughly 1,250 hours of transcribed speech for automatic speech recognition (ASR), along with over 20 hours of professionally recorded studio audio for text-to-speech (TTS) systems.
A community-led effort
WAXAL is the result of close collaboration with African institutions and organizations whose local knowledge and leadership were essential to its success. Makerere University in Uganda and the University of Ghana oversaw data collection for a total of 13 languages, while Digital Umuganda in Rwanda led work on five widely spoken languages. To produce studio-quality voice recordings, the project partnered with Media Trust and Loud n Clear, drawing on their regional expertise. Multilingual data for future releases was developed in collaboration with the African Institute for Mathematical Sciences (AIMS).
This partnership model ensures that contributors maintain ownership of the data they gathered, while collectively working toward a shared vision: making African language resources accessible to the global research community.
Ethical, real-world speech capture
Authenticity was a core priority throughout the project. Participants were invited to describe images in their native languages, allowing the dataset to reflect natural, everyday speech patterns. In parallel, trained voice professionals were recorded in studio environments to generate the clarity and consistency required for high-quality text-to-speech applications.
Beyond advancing speech technology, WAXAL also aims to support the long-term digital preservation of African languages. The full dataset has been released under an open license and is now freely available on Hugging Face, ready for researchers and developers to explore and build upon.
