A Collelogram based Pitch and Voiced/Unvoiced Classification Method for Real-Time Speech Analysis in Noisy Environment

Abstract

Pitch estimation is frequently used in voice quality analysis. Speech in a noisy environment, the accuracy of pitch extraction is poor due to the effect of noises. This paper presents a simple technique for robust pitch estimation as well as voiced/unvoiced classification based on correlogram of noisy speech signal. Like spectrogram, the short-time autocorrelation outputs can be displayed graphically as another image called correlogram is an alternative to short time spectral analysis. This technique operates frame-by-frame basis on normalized autocorrelation function (NACF) of signal. Initially, the noisy speech signal is low pass filtered within the pitch range 50-500 Hz to obtain the pre-filtered signal. Then a threshold function is derived from the NACF. We use this threshold value for pitch position indicator and voiced/unvoiced classifier. The accurate pitch period is obtained from the weighted correlogram. The proposed pitch estimation and voiced/unvoiced classification algorithm using correlogram is very simple, fast and easily implemented in computer. The performance of the proposed algorithm is compared with recently developed EMD based method. The experimental results show that the proposed one is useful in speech analysis research.