Pathological Voice Recognition based on Nonlinear Method
Zhang Xiao-jun · Information Security and Communications Privacy · 2014
This paper proposes a special identification method based on nonlinear characteristics to extract the characteristic parameters of the vocal nodules class of pathological voice. Firstly,by using the method of filtering segmentation, these voice signals are divided into two-channel to process. The low-frequency voice signals use the Bark filter bank meeting the human auditory characteristics to reconstruct the signals and extract the speech features, while the high-frequency ones use the largest Lyapunov exponent based on the nonlinear dynamics. Finally,The two should be integrated into a sequence of speech features for voice recognition. Recognition experiment is based on the database of pathological voices of the U.S. MEEI's company. The experimental results indicate that this method can effectively improve the recognition rate of pathological voice,making the recognition rate up to 99.4%.