| Dysarthria is a general term for a speech disorder in which speech is slow, weak, imprecise or uncoordinated. Commercially available automatic speech recognition (ASR) systems cannot reliably recognize dysarthric speech due to the inherent variability in such utterances. People with dysarthria generally lack articulatory precision. Simple phonemes like vowels are physically the easiest sounds to produce, since they do not require dynamic movement of the vocal system. This research is primarily a feasibility study investigating the reliability of vowel-based phoneme recognition of dysarthric speech. The goal is to evaluate if ASR algorithms could be used to reliably differentiate among the different vowel sounds produced by dysarthric speakers. The intended purpose is to provide personal computer based access methods for people with dysarthric speech. In this work, the hidden Markov model (HMM) is the basic technological approach adopted in developing the speech recognition algorithms, and all the experimental results quantifying the feasibility of these algorithms are presented. |
