Comparison of different time-frequency distributions in pitch detection

The characteristics of time-frequency distribution (TFD) bring good performance in analyzing nonstationary signals, especially signals with transition frequency following time evolutions. In this paper, the TFDs will be applied to pitch detection of speech signals. The basic idea of TFDs will be introduced first. Then some TFD algorithms will be developed to identify the onset and end positions of each pitch pulse and the performance on tracking pitch pulses is described. A speech signal segment will be processed using the novel time-frequency methods and the traditional autocorrelation method; the detection results will be compared.