Together with Roularta, VRT aims to develop a new and scalable Flemish text-to-speech model that can generate qualitative, realistic speech based on written input. Through an automatic context analysis, the system needs to understand the context of the text in order to improve the general prosody (stress, rhythm, intonation) as well as the emotion of the TTS system in order to obtain a human character. In particular, the system should allow new, unique custom Flemish voices to be generated in a scalable way and linked to a specific brand.
On the long run, this TTS system will allow a large number of applications, such as the generation of new audio formats based on non-audio content, automated voice-overs and automated podcast readers. The innovation project will generate significant advantages in terms of cost, speed, and flexibility in the creation of new voices and innovative audio content.
This project has received funding from the Flemish government, as part of the digital transformation programme for the Flemish media sector.