Nonlinear feature extraction of pathological voice based on the human ear filter bank

Inhalt:
The nonlinear features of voices can effectively distinguish normal and pathological voices; however, feature extraction is a time-consuming process, which is unfavorable for real-time processing. This study aims to reduce the time of feature extraction based on nonlinear features to either maintain or improve the rate of pathological voice recognition. A method based on the human ear filter bank is proposed to optimize the extraction of nonlinear characteristics of pathological voice. In the extraction process, feature optimization is achieved by extracting the nonlinear features of a certain frequency band. The results show that the method based on the human ear filter bank greatly shortens the extraction time and typically acquires a higher rate of parameter identification than the traditional methods.