Google Research and Zindi LinksThe award All news
I’m happy to share that I’ve won 2nd place and the Best African Participant prize at the Google WAXAL ASR Challenge.
Organized by Zindi and Google Research, the challenge was to use the WAXAL dataset to build multilingual automatic speech recognition (ASR) systems that can generalize to previously unseen speech data, with a focus on 3 low-resource languages: Lingala, Shona and Luganda.
WAXAL (from the Wolof word for “speak”) was developed through a multi-year collaboration between Google Research and multiple African academic and community organizations. It is one of the largest openly accessible speech resources for African languages, covering 27 languages spoken by more than 100 million people, and includes thousands of hours of natural speech and high-quality recordings for both speech recognition and speech generation research.
For the past 3 years, I’ve contributed to developing machine translation, ASR and text-to-speech models at GO AI Corporation for Mooré, my mother tongue and the most widely spoken language in my lovely home country, Burkina Faso. Working on 3 new languages (that I don’t speak a word of) was particularly challenging, but it helped me discover new strategies to improve the robustness and generalization of ASR systems.
Credit to the many organizations, engineers and researchers who pioneered the open-source datasets, tools and foundation models for African NLP, and thanks to Zindi and Google Research for hosting the challenge!
Read the announcement · the final leaderboard