SONDER is NTGent’s next production, directed by Yves Degryse, premiering in February 2027. It uses technology to try to understand what happens to a person after they’re gone. On stage, Alva Ishii, NTGent’s digital house artist, speaks in a West-Flemish voice.
There are no language models for West-Flemish. So northstar builds the voice: a system that listens to Dutch, answers like a person would, and speaks the answer in dialect.
Why it matters
Most speech technology is built for big languages. Dialects like West-Flemish are left out. SONDER is a chance to show that small, local speech models can be built here, run on local hardware, and be good enough for a theatre stage.
It’s also a production about memory and loss, so the voice has to be honest. The system is built to say when it doesn’t know, instead of making things up.
Where it’s going
The first prototype was built during a residency at Wintercircus in February. Since then we’ve recorded speakers in West Flanders and trained our own speech models. The voice is getting close. The dialect isn’t there yet, because West-Flemish changes from village to village and very little of it has ever been recorded.
The months before the premiere are about collecting the right speech and testing it with native listeners. We work on it every week with Yves Degryse and Marieke Cardinaels at NTGent, and with Joni De Borger, who develops Alva.