Maarallee

With maarallee.be, we collect voice recordings to build a dataset of spoken Flemish. Participants can make short audio recordings through our Maarallee app, using their own way of speaking, with their own accent, dialect, or intonation—simply by answering an audible prompt.

These speech recordings are then converted into written text, which is subsequently corrected by volunteers. And all of this is done in a privacy‑friendly way, of course! This combination of corrected text and voice recordings can be used as training data to improve automatic speech recognition (also known as ASR). These are AI models that recognise sounds as words—and that’s not an easy task for a computer! By collaboratively building a large database of spoken Flemish, we create the conditions for developing ASR models that are specifically tailored to the Flemish language and that truly understand our Flemish sounds and way of speaking.

In this way, we gradually help AI understand Flemish better—from Acid and Willem Vermandere to Natalia and Belle Perez. That’s the plan! Speech technology that understands the Flemish way of speaking—whether you’re from Pelt, Ostend, Borgerhout, or anywhere else.