An effective pitch detection method for speech signals with low signal-to-noise ratio

Pitch detection in noisy environment plays an important role in speech analyzing and recognition. In this paper, an effective pitch detection method is proposed. Noised speech is denoised by an improved form of spectral subtraction method. A linear predictive coding analysis is performed on the segmented speech, and the segmented speech is filtered by the inverse filter to give the linear prediction error. The cepstrum of the linear prediction error and the autocorrelation function of the cepstrum are calculated. The result of the simulation shows that compared with the autocorrelation function pitch detection method, a distinct improvement in effect can be seen by using this improved method in pitch detection.