Pathological voice detection based on improved EWT
Feature extraction is a crucial step in the detection of pathological voice signals.To address the issueof frequency band partitioning when dealing with complex spectral signalsin Empirical Wavelet Transform(EWT),an improved EWT based on the cepstral envelope is proposed.It adaptively partitions the first and second resonance peak frequency bands of the vowel/a/,and obtains the EWTPCC(Empirical Wavelet Transform Pearson Correlation Coefficient)feature by calculating the Pearson correlation coefficient between different frames within the first and second resonance peak frequency bands.The experimental results demonstrate that the method combining the EWTPCC feature with Support Vector Machines(SVM)achieves a recognition rate of 87.65%on the Saarbrücken Voice Database(SVD).This approach effectively distinguishes between normal and pathological voices.
万方数据
维普
