Speech Recognition for People with Severe Dysarthria

This is the home page of the STARDUST project.

STARDUST is funded by the UK NHS NEAT programme from September 2000 for three years.

STARDUST is a collaboration between

  • The SChool of  Health And Related Research (SCHARR), University of Sheffield
  • The SPeech AND Hearing Research Group, Department of Computer Science, University of Sheffield (SPandH).
  • Barnsley District General Hospital


Dysarthria is a common speech disorder which, in its severest form, results in unintelligible speech. This disorder is commonly associated with other general neuromotor disabilities. People with these disabilities may have difficulty in making themselves understood as well as in reliably controlling environmental and communication aids. We propose to apply state-of-the-art Speech Technology research to the recognition of severely dysarthric speech. Off-the-shelf automatic speech recognition (ASR) systems function poorly for these users because of the increased variability of their articulations compared to normal speech and the difficulty of finding suitable corpora for training the recogniser. To overcome these problems we propose to develop a software application which integrates three functions:
  • Real-time visual feedback to assist dysarthric speakers to improve the consistency of their vocalisations.
  • Speech recognition capability with improved tolerance to speech variability.
  • The ability to tailor the package to the needs of the client: to modify the feedback as training proceeds and to construct new recognisers incrementally as data is collected.
The speech recognition software application will then be deployed to control electronic assistive technology, either a communication aid or an environmental control system.