Google’s WAXAL Initiative Expands AI Access Across Africa

A new large-scale speech dataset, WAXAL, developed by Google in collaboration with African research institutions, is poised to revolutionize artificial intelligence (AI) accessibility across Sub-Saharan Africa. The initiative aims to address the significant gap in AI technology that understands and responds to African languages, empowering over 100 million speakers across more than 26 countries. [Google Research]

Addressing the AI Divide in Africa

Voice-enabled technologies have largely favored high-resource languages, leaving millions in Africa unable to access essential tools in their native tongues. Sub-Saharan Africa, home to over 2,000 distinct languages, has been particularly underserved. WAXAL, named from the Wolof word for “speak,” seeks to rectify this imbalance by providing a critical foundation for African speech technology. [Rest of World]

WAXAL: A Collaborative Effort

The WAXAL project, initiated in 2021, is a collaborative achievement powered by leading African organizations. Key partners include Makerere University (Uganda), the University of Ghana, Digital Umuganda (Rwanda) in partnership with Addis Ababa University, and the African Institute for Mathematical Sciences. These institutions led the data collection efforts, guided by Google’s expertise in data collection practices. [BusinessDay]

A crucial aspect of WAXAL is the principle of local ownership. African partners retain ownership of the data they collected, while collaborating with Google to make the resources openly available to the global research community. [Rest of World]

Dataset Details and Licensing

WAXAL’s initial release features approximately 1,846 hours of transcribed natural speech for automatic speech recognition (ASR) and over 565 hours of high-fidelity recordings for text-to-speech (TTS) across 27 Sub-Saharan African languages. [Google Research] The dataset covers 21 languages, including Acholi, Hausa, Luganda, and Yoruba. [Rest of World] The resources are released under a Creative Commons license (CC-BY-4.0) to encourage research and the development of inclusive voice-enabled technologies. [Google Research]

Impact and Future Research

WAXAL is already catalyzing research, including the development of a cookbook for community-driven collection of impaired speech and the creation of the first open-source dataset for Akan speakers with conditions like cerebral palsy, and stammering. A major study also introduced a 5,000-hour speech corpus for five Ghanaian languages – Akan, Ewe, Dagbani, Dagaare, and Ikposo. [Google Research] research has benchmarked four state-of-the-art models across 13 African languages, offering insights into data efficiency and linguistic complexity. [Google Research]

WAXAL represents a significant step towards digital sovereignty for Africa, empowering local entrepreneurs and researchers to build AI tools tailored to the continent’s unique linguistic diversity. [Rest of World]