摘要：This paper proposes a recognition method forextracting nonlinear characteristic parameters of pathologicalvoices using Bark wavelet sub-band filtering. First, the speechsignal was processed through 24 Bark filter banks. According tothe signal obtained from each channel, the featuremulti-frequency band nonlinear coefficient was extractedaccording to the frequency division factor, . We used 53normal voices and 117 pathological voices from the MEEI’spathological voice experimental database, and 14 machinelearning methods were used to perform the recognitionexperiments. The experimental results showed that theproposed method effectively improved the recognition rate.Moreover, the proposed feature was optimal when = 18, andthe highest recognition rate was achieved when the supportvector machine learning algorithm was used.
关键词：frequency division; nonlinear; pathological voice speech recognition; frequency division factor
