Acoustic features of vocal tract function are used widely in the study of pathological voices detection. Classification of normal and pathological voices by acoustic parameters is a useful way to diagnose voice diseases. In this aspect, mel-frequency cepstral coefficients are proved to be effective with traditional classifiers such as Gaussian Mixture Model (GMM). However, the accuracy of the classification method can be further improved.

WANG Xiang, ZHAN Jianping and YAN Yonghong of Thinkit Speech Lab, Institute of Acoustics, Chinese Academy of Science carried out a series of studies and compared a Gaussian mixture model supervector kernel-support vector machine (GMM-SVM) classifier with GMM classifier for the detection of voice pathology.

They found that a sustain vowel phonation can be classified as normal or pathological with an accuracy of 96.1%. Voice recordings are selected from the Kay database to carry out the experiments. Experimental results show that equal error rates decrease from 8.0% for GMM to 4.6% for GMM-SVM.

This research result was published on the recently issued Journal of Voice (2011, Vol. 25, Issue.1, Jan. 2011, Pages 38-43).