The project will develop and implement new models for linguistic prominence (i.e. how speakers highlight important stretches in speech) in text-to-speech synthesis.

Incorrectly placed or missing prominence in synthetic speech has a highly negative effect on intelligibility and naturalness and makes listening to long stretches of synthetic speech tiring. The prominence models will be systematically tested and deployed in the context of Wikispeech – a recently inaugurated project to create an open source text-to-speech system and online service to make Wikipedia (and other Wikimedia projects) accessible to people that have difficulties reading.

Key people

Belongs to: Information and Communication Technology - The Next Generation
Last changed: Apr 27, 2019