Research On The Algorithm Of Speeth Pitch Detection

Pitch period is one of the important parameters to describe the excitation source,which has been applied widely in the field of speech signal processing, such as speech synthesis, speech coding and speech recognition,etc. Detecting pitch period accurately and reliably will influence the performance of the whole speech processing system directly.Pitch detection methods,which are often used nowadays, all have good performance in original speech signal,however, speech signal is inevitably influenced by background noise,which makes those commod detected methods not get good performance in noise environment.So we focus on the study of pitch detection algorithms which have better accuracy and robustness.Firstly,this paper introduces the background and the significance of pitch detection. Secondly, the paper sums up and concludes the existing pitch detection algorithms, and introduces the fundamental theory as well as the realization of several common detection methods in details.Finally, based on the further researches on the theory of EMD,SVD and the existing detection algorithms, several algorithms are proposed as following.(1) According to the problem that the performance of common pitch detection methods is influenced by background noise strongly,by reference to the wavelet de-noising method,a self-adaptive de-noising method for speech signal is proposed. At the same time a new voiced/unvoiced decision method based on EMD is proposed.(2) Based on the EMD self-adaptive de-noising algorithm and the median smoothing method,a improved pitch detection method which combines with CAMDF and ACF is proposed. (3) A new pitch detection algorithm is proposed based on the theory of detecting periodic signals by SVR spectrum.The methods extract voice sound by the voiced/unvoiced decision methods based on EMD.It calculates the average amplitude difference of the unvoice sound and analysis the SVR spectrum of the results of the calculation. Finally,the pitch cycle is extracted by peak detection.This paper also finishes the program's designing, carries on the simulation experiments for above algorithms in Matlab7.0 simulation environment, and makes contrast test under different SNR.Experiment's results show that the new algorithms have high accuracy in pitch period estimate,and have better character in resisting the noise.