HOSAN

Automatic speech recognition (ASR) is increasingly used in the creative industry – from interactive installations and voicebots to media subtitling. With the project High-Quality Speech Recognition for All Varieties of Dutch (HOSAN), sector partners are working together to make this technology more inclusive. Research shows that ASR performs significantly worse for certain groups, such as people with an accent or dialect, older adults, children, and speakers for whom Dutch is a second language. As a result, these users risk being excluded from digital services that rely on ASR.

In this project, Sound & Vision will make extensive speech data from the NPO archive available to train new, representative Dutch speech models using SURF’s national supercomputer. In this first phase, the consortium will explore – based on concrete use cases – the technical and organizational requirements for improving ASR for all Dutch speakers.

€89.403,- will be used as a PPP program grant.