Statistical approach for the emotional speeches of Parkinson Dysarthria

摘要

Dysarthria is a condition where the speech muscles becomeweak and become difficult to control. It leads to majordeficiencies in human speech. The dysarthria speakers withParkinson disease (PD) suffer from very slow and slurredvoice. It becomes difficult to understand by other people.In this study we conduct a statistical analysis on the sixbasic emotions (anger, disgust, fear, sadness, happiness,surprise) from the speech records of various dysarthriaspeakers. As a method to extract the features, we use shortterm spectral analysis (STSA) to extract the major prosodicfeatures from the speech samples. This method is widelyused in the field of speech recognition and featureextraction.We derive hidden Marcov models (HMMs) for eachprosodic feature. Using the HMMs, we determine themaximum log likelihood probability coefficients in order toconduct statistical analysis. The aim of this statisticalanalysis is to validate the para-linguistic information ofevery speech samples of PD with the speech samples ofhealthy controls (HC).