Investigation and implementation of computer-based aids for the speech impaired
Thesis DisciplineElectrical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
A computer-based visual speech aid has been developed to help those with speech defects. By transforming speech into visual patterns, the speech-impaired are shown visual representations of target speech features and of their attempts to reproduce these features. Real-time digital signal processing allows a speech therapist and client to use the speech aid interactively. The aid has been tested in the clinics of several therapists. Two special tests have been designed to assess the speech aid's ability to separate acceptable and unacceptable speech, and to display crucial features of the speech. In addition, two new speech analysis algorithms have been developed for the aid. The first is a recognition algorithm for isolated fricative sounds of English. It is based on spectral classification. The second is an algorithm for reconstructing the shape of the vocal tract and it uses Wiener filtering to deconvolve the glottal pulse from the speech signal.