Towards the improvement of automatic recognition of dysarthric speech

Dysarthria is a motor speech disorder that is often associated with irregular phonation (e.g. vocal fry) and amplitude, in coordination of articulators, and restricted movement of articulators, among other problems. The aim of this study is to raise dysarthic speech recognition rate through producing intelligibility enhanced speech using a procedure in which formants and energies are estimated from dysarthic speech and modified to more closely approximately desired normal targets. The modified parameters are taken to formant synthesizer to get final transformed speech, tested through perceptual tests to ensure quality and intelligibility. Then, we passed the modified dysarthric speech through an automatic speech recognition engine based on the HTK hidden Markov model toolkit. Speech recognition tests results indicate that the applied conversion algorithm raises the recognition rate of the dysarthric speech from 28% to 71.4%.