Continuous Speech Classification Systems for Voice Pathologies Identification

摘要

Voice pathologies identification using speech processing methods can be used as a preliminary diagnostic. The aim of this study is to compare the performance of sustained vowel /a/ and continuous speech task in identification systems to diagnose voice pathologies. The system recognizes between three classes consisting of two different pathologies sets and healthy subjects. The signals are evaluated using MFCC (Mel Frequency Cepstral Coefficients) as speech signal features, applied to SVM (Support Vector Machines) and GMM (Gaussian Mixture Models) classifiers. For continuous speech, the GMM system reaches 74% accuracy rate while the SVM system obtains 72% accuracy rate. For the sustained vowel /a/, the accuracy achieved by the GMM and the SVM is 66% and 69% respectively, a lower result than with continuous speech.