Research on a new method of preprocessing and speech synthesis pitch detection

Abstract

In order to solve pre-existing audio voice signal edge extraction is not accurate, easily lead to useful signal missing and the extraction of speech signals pitch detection easily misjudge the problems, proposed a threshold preprocessing voice method and an improved method of extracting the pitch period, we used moving window and polynomial smoothing method (Savitzky-Golay) smoothing the original speech signal, then preprocessing sound files by setting the amplitude threshold method to treat speech signal, and get ideal speech signals. Do processing by clipping voice signal center, Then use the modify autocorrelation function (MACF) and modify of the average magnitude difference function (W-AMDF) separately dealt with voice waveform, according to the characteristics of the waveform to processing signals, in order that troughs of the waveform position become the location of pitch period, thus the signal pitch period extracted. Experimental results show that the using the method can effectively improve the sound files of effective pre-signal extraction accuracy, improve the efficiency of signal processing and reduced the pitch detection misjudge significantly.