Time-frequency approach in continuous speech for detection of Parkinson's disease

In this paper low-frequency analysis is addressed in order to explore components of continuous speech signals, trying to making evident the changes in the spectrum, which could be associated to the tremor in speech of people with Parkinson's disease. Four time-frequency (TF) techniques based on WignerVille distribution (WVD) are used for the characterization of the low frequency content of the speech signals. The set of features includes centroids and the energy content of different frequency bands, due to the assumptions of non-stationary was taken into a account using enough time frameworks. The discrimination capability of the estimated features is evaluated using a support vector machine (SVM). The results show that the low frequency components are able to discriminate between pathological and healthy speakers with an accuracy of 72%.