Maarallee has already collected a large amount of Flemish speech from enthusiastic volunteers. This speech data was processed anonymously by an AI model developed at KU Leuven. However, the transcripts generated by the model still contain errors that need to be corrected by people.

To support this task, Maarallee is looking for volunteers who are willing to review and correct transcribed speech recordings. This combines data collection and data quality control in a single effort.

Through Zooniverse.org, the world's largest citizen science platform, participants can listen to audio recordings and improve AI-generated transcripts. These small corrections make a big difference. They help train the speech technologies of the future to better understand how people in Flanders really speak.

As the recordings and transcripts are in Dutch, participation is currently only open to Dutch-speaking volunteers.

To ensure that the geographical spread of speakers is represented as accurately as possible, Maarallee is looking for contributors from all Flemish provinces. Every accent counts!

You can contribute by:

• creating an account on Zooniverse.org;
• listening to audio fragments;
• correcting AI-generated transcripts;
• helping to build the Maarallee dataset of spoken Flemish.

And of course:

In addition to building a high-quality dataset, Maarallee is also looking for more and longer speech recordings. Do you have some time to spare? Download the app, answer a few questions and help speech technology learn from real people.

Because the more diverse the voices, the better.

A big thank you. So, are you joining us?