Voice disorders identification using discrete wavelet based features

Voice disorders are currently one of the most common diseases. This study aims to determine whether a person has voice pathology by analyzing his/her sound samples. For this purpose, co-utilizing the discrete wavelet transform based the linear predictive cepstral coefficients and their statistical parameters is proposed as feature vector. In the experiments, five different vocal fold disease groups and healthy individuals out of 304 people were employed using sustained /a/, /i/ and /u/ sounds. Experimental results show over 99% correct recognition performance.