基于小波变换和线性预测的基音提取.pdfVIP

36 10 2010 5 Vol.36 No.10 Computer Engineering May 2010 ·· 2010 A N945 1 1,2 (1. 1000842. () 100084) 97.8%0 dB 91.1% Pitch Extraction Based on Wavelet Transformation and Linear Prediction 1 1,2 LI Kun , LIU Jia (1. Department of Electronic Engineering, Tsinghua University, Beijing 100084; 2. Tsinghua National Laboratory for Information Science and Technology, Beijing 100084) AbstractTo effectively extract the pitch period, an improved wavelet transformation method is presented. On basis of wavelet filter, the frequency multiplication suppressed coefficients are added to remove most of the frequency multiplication errors. According to the voice, the system combines the linear prediction, which has a suitable order adjusting to the speech, to remove the format interference, while avoiding excessive damage to the pitch. Experimental results show the accuracy rate of the method is 97.8% at quite environment, and when the SNR is 0 dB, the one of it is 91.1%. The system has a good accuracy, robustness, and noise immunity. Key wordspitch extraction; wavelet transformation; linear prediction 1 2 60 Hz~500 Hz ()