WAXAL: A Massive Open Speech Dataset for 27 African Languages

Google Research just dropped WAXAL, and honestly, this is the kind of thing that makes me optimistic about where speech tech is headed. It’s a massive open dataset covering 27 Sub-Saharan African languages spoken by over 100 million people across 26+ countries. And they’re releasing it under a CC-BY-4.0 license, which means anyone can use it without jumping through hoops.

Voice assistants and transcription tools have been a game-changer for a handful of languages. But if you speak Yoruba, Wolof, or Luganda? Good luck finding a decent speech recognition system. That’s not just an inconvenience—it’s a barrier. Sub-Saharan Africa alone has over 2,000 languages, and most of them have been completely ignored by the big tech companies. WAXAL is trying to change that.

The project started back in 2021, and it’s been a multi-year collaboration with African academic institutions and community organizations. That’s key. Too often, datasets are collected by outsiders who don’t understand the linguistic nuances. Here, local communities were involved in everything from script writing to recording.

What’s in the dataset?

WAXAL comes in two flavors:

WAXAL-ASR (for automatic speech recognition): About 1,846 hours of transcribed natural speech. Instead of having people read scripts—which always sounds stiff and unnatural—participants described images from Google’s Open Images dataset. This approach captures real speech patterns, including tonal variations and code-switching. If you’ve ever tried to build an ASR system for a tonal language, you know how crucial this is.

WAXAL-TTS (for text-to-speech): Over 565 hours of high-fidelity recordings designed for synthetic voice generation. The process here was surprisingly hands-on: community members worked in pairs, drafting 10,000–20,000 word scripts and alternating between reading and recording. Some participants even used project funding to build custom studio boxes for professional-grade acoustics. That’s dedication.

Why this matters

Data scarcity has been the single biggest bottleneck for African language speech tech. Most existing datasets are tiny, restricted, or don’t capture natural speech patterns. WAXAL changes that by providing a foundation that researchers and startups can actually build on.

The permissive license is a big deal too. CC-BY-4.0 means you can use it for commercial projects, modify it, share it—just give credit. No legal headaches, no negotiating with a corporate legal team. This is how open research should work.

I do have one criticism: 27 languages is a start, but it’s a drop in the ocean. The continent has over 2,000 languages, and even within those 27, coverage varies. Some languages have hundreds of hours of data, others much less. Google says they intend to expand the collection over time, but I’ll believe it when I see it. Corporate commitments to “continuous expansion” have a way of fizzling out.

Still, this is a genuinely impressive effort. The methodology is sound, the collaboration model is right, and the licensing is generous. If you’re working on speech tech for African languages, WAXAL is the resource you’ve been waiting for.

Check out the dataset and the paper if you want to dig into the details. This is one of those rare releases that could actually move the needle.