Google and African researchers start the WAXAL speech dataset

Voice tech finally stopped ghosting African languages after a massive open dataset tackled the data drought head-on.

Why voice tech kept failing locally

• Many devices choke on African languages.
• Over 2,000 tongues lack usable speech data.
• Sub-Saharan users get locked out of convenience.
• Data scarcity stayed the core blocker.

What WAXAL brings to the table

• WAXAL rolled out as a large-scale speech dataset.
• Named after the Wolof word for speak.
• Covers 21 African languages.
• Built to unlock inclusive voice systems.

What the dataset actually contains

• Nearly two million recordings fuel the corpus.
• Total audio clears 11,000 hours.
• Roughly 1,250 hours are fully transcribed.
• Studio speech supports text-to-speech work.

Who built it together

• Makerere University gathered language data.
• The University of Ghana supported 13 languages.
• Digital Umuganda was led in five languages.
• The African Institute for Mathematical Sciences added multilingual datasets.

Studio and quality control work

• Media Trust helped produce clean voice recordings.
• Loud n Clear handled professional audio capture.
• Everyday speech balanced with studio voices.
• Ethical collection stayed a priority.

Data ownership and access rules

• Contributors keep rights to their data.
• Researchers worldwide get open access.
• Sharing does not erase local control.
• Collaboration stays balanced.

Why it matters long term

• Voice tools can finally serve local users.
• Language preservation gains a digital backup.
• AI systems learn from real speech.
• The dataset is live under an open license.