A Time Domain Estimation Algorithm for Speech Signal Pitch Period

Abstract

Abstract In speech recognition and speech synthesis, accurate estimation of the pitch period is an important part of speech processing. The traditional direct peak estimation method and the autocorrelation function method are both effective time domain estimation algorithms. The autocorrelation method is a pitch period estimation algorithm suitable for low SNR. Both algorithms need to get accurate peak position estimation. In this paper, a multi-line cut method which is a method for judging the position of the peak point is proposed. The multi-line cut method is used to intercept the sampled data of the waveform by using multiple cut lines. The median value is calculated by the starting and ending points of the cut line position, and the peak position is indirectly evaluated. By minimizing the impact of interference on the peak estimate, the likelihood of falling into local extreme points is reduced, therefore a more accurate peak point estimate than the direct search for peak points can be obtained. The simulation results show that compared with the traditional direct peak estimation method, the performance of peak estimation by the multi-line cut method can be greatly improved, and the multi-line cut method can be used to estimate the peak value in the autocorrelation method, and also achieve a certain performance improvement. In addition, the number of cut lines is directly related to performance, and the more the number is, the better the performance is. The complexity of this method is not high and easy to implement.