Chinese dialect identification based on gender classification

This paper designs a novel Chinese dialect identification system based on gender classification. RelAtive SpecTrAl Perceptual Linear Predictive (RASTA-PLP) coefficients are applied to construct gender dependent GMM models. Gender classification is conducted before testing. The gender classification criterion is the average pitch of the segment of testing speech. Experiments are evaluated on Chinese dialect telephone speech corpus which was recorded by our laboratory. The results demonstrate significant improvement in Chinese dialect identification compare to the gender independent method with a slight increase in the processing computation. The improved performance can attain 3.77% at least.