WAXAL: The African Speech Dataset That Actually Matters

Let’s be real for a second: most speech technology is built for English, Mandarin, maybe Spanish if you’re lucky. Meanwhile, Sub-Saharan Africa alone has over 2,000 languages, and the people speaking them get left out of the voice assistant party. Google Research has been sitting on this problem for a few years, and now they’re finally dropping something substantial.

WAXAL is the name, and it’s not just another dataset dump. This thing covers 27 native languages spoken by over 100 million people across 26+ countries. The team started working on it back in 2021, partnering with African academic institutions and community orgs. That’s the kind of collaboration that actually makes a difference, not just parachuting in with a microphone and leaving.

The numbers: roughly 1,846 hours of transcribed natural speech for automatic speech recognition (ASR), plus over 565 hours of high-fidelity recordings for text-to-speech (TTS). Both are released under Creative Commons CC-BY-4.0, which is about as permissive as it gets. No weird licensing games, no “for research only” asterisks. They want people to build stuff with this.

What I find interesting is how they collected the ASR data. Instead of making people read scripts — which always sounds robotic and unnatural — they used image prompts. Participants described visual stimuli covering 50+ topics in their native language. This captures real linguistic quirks: tonal variations, code-switching, the way people actually talk when they’re not staring at a piece of paper. It’s a small shift in methodology that makes a huge difference in data quality.

The TTS side is equally thoughtful. Local community members worked in pairs, drafting scripts of 10,000 to 20,000 words, alternating between reading and recording. Some even used project funding to build custom studio boxes for proper acoustics. That’s not just data collection — it’s capacity building. People learned skills, set up infrastructure, and now have something to show for it beyond a paycheck.

Google says this is just the initial release and they plan to keep expanding. If they actually follow through, this could be the foundation for African language speech tech for years to come. No more building everything from scratch every time someone wants to add a new language. No more relying on sketchy datasets scraped from YouTube.

This is the kind of open resource that actually moves the needle. Not a press release, not a feel-good blog post — real data, real licenses, real impact. Let’s see what the community does with it.