Abstract:

The extracting pitch frequency based on the harmonic autocorrelation algorithm is discussed. First, a harmonic autocorrelation model is derived. By this model a leastsquares fitting optimization technique is developed to extract a pitch harmonic from the noise speech. Then, the harmonic number is determined by an objective function formulated as the shorttime autocorrelation function of the noisy speech weighted an impulse train. The pitch contour is obtained by pitch harmonic and harmonic number. The experiments show that in low signal-to-noise ratio, the error rate of harmonic autocorrelation method is 5.0% and 13.6% lower than that of autocorrelation method.