Pitch detection algorithm using a wavelet correlation model

A new algorithm for pitch detection of the speech signal is introduced. The technique is based on the discrete wavelet transform to classify the speech signal into voiced and unvoiced segments. The wavelet parameters of the voiced segments in two frequency bands are extracted and crosscorrelation is performed to generate a correlation function. Then, a peak detection is applied to extract the pitch period. The algorithm is highly immunized to noise. A comparison between the ordinary methods and this new one is presented. The pitch contour is varying through the utterance period rather of being consider as constant through the analysis periods as in ordinary methods. The results are accurate for speech of signal to noise ratio equals to 2 dB.