17 October 2025

Researchers have launched a citizen science platform to enable AI to better understand Flemish dialects and accents.

The Maarallee platform, developed by KU Leuven and the Flemish Knowledge Centre for Citizen Science (Scivil), invites people to submit short voice recordings to increase the data available for training AI models.

Current speech recognition technology is mostly based on data from the Netherlands, meaning that AI models don’t recognise the Flemish accent very well.

“Anyone who wants their GPS to understand them when they say ‘Turnhout’ with, for example, a Kempen accent, can help find a solution,” says Isaak Vandermaesen, AI and Education project officer at Scivil. “By recording your voice on maarallee.be, you are helping to build an AI that is truly tailored to all Flemish people.”

"We are taking a step towards AI that can correctly understand every Flemish person"

Anyone can record short sentences, which are automatically transcribed and then corrected by volunteers. This provides data that KU Leuven can use to develop a robust automatic speech recognition model for Flemish.

Limburgers do not speak more slowly, KU Leuven research shows

Contrary to popular belief, speakers from the province of Limburg do not speak more slowly than other Flemish people. This is the conclusion of...

belganewsagency.eu

The partners say the collected data has both scientific and social value. Speech recognition is playing an increasingly important role in education, media, healthcare, industry and services, in areas such as automatic subtitling, hands-free systems and voice-controlled tools for people with disabilities.

“Without broad and diverse speech data, AI remains a technology for the few,” says Vandermaesen. “With Maarallee.be, we are taking a step towards AI that can correctly understand every Flemish person.”