The Research On Pitch Detection Algorithm Of Speech With Noisy

Extracting the characteristic parameters of the speech signal accurately is the premise of its analysis and application. The pitch period of speech signal equals to the cycle of the voiced speech signal, which carries the most useful information of the speech signal. It can be effectively used to speech synthesis, speech recognition, speech coding and speaker identification techniques, etc.Speech signals are often lost in the noise when in the noisy environment. Base on the in-depth study for the characteristics of the noisy speech signal and the existing methods, this paper proposes new algorithms with higher accuracy and less complexity.Analyses the original pitch detection methods, do experimental simulation and analysis, summarize their respective advantages and disadvantages.For the spatial autocorrelation filtering method in the low SNR can achieve good noise reduction effect, this paper puts forward a new pitch detection algorithm, which combine the spatial autocorrelation filtering method and the time-domain waveform. The algorithm applies the spatial autocorrelation and center clipping to do pretreatment, and then extract the desired speech signal waveform information, according to the text matching rules required for signal matching, and then calculated signal pitch cycle.According to the multi-resolution characteristics of empirical mode decomposition, the paper combines the EMD soft threshold de-noising method and Hilbert-Huang transform method to extract the pitch period. The algorithm firstly use the EMD to decompose the signal, and secondly set the threshold on each IMFs to remove noise, thirdly calculate instantaneous frequency and instantaneous amplitude on each branch of the IMF component to generate the Hilbert spectrum; finally, deal the Hilbert spectrum with the corresponding rule, resulting in a final pitch period locus.