Speech rehabilitation in persons with a (chronic) speech disorders like dysarthria is very demanding and requires a lot of time and financial means. Speech technology based solutions offer an interesting supplementary opportunity. In this project which started within a European H2020 project, a Virtual Articulation Therapist (VAT) is build and further clinically operationalised. Different neural networks were trained in collaboration with partners from the TAPAS consortium (https://www.tapas-etn-eu.org/) to allow a precise phoneme identification and detailed phoneme error detection. The developed algorithms were integrated in a friendly user interface to allow patients to train themselves in their own environment. The VAT will be further evaluated with samples of healthy speakers and speakers with dysarthria on feasibility, acceptability and effectiveness.
Next to the development of the VAT, acoustic features of accent production are explored to support perceptual evaluation of dysarthric speech. Both articulation and prosodic features like word accent play an important role in speech intelligibility.