Pitch determination of noisy speech using wavelet transform in time and frequency domains

Accurate determination of the pitch period or fundamental frequency of speech signals is crucial for certain types of speech coding algorithms. A comparatively new pitch extraction technique involves the use of the wavelet transforms (WT) in the time domain. The authors consider the modification of the existing time domain pitch determination algorithm, based on the WT to obtain improved performance, and present a new frequency domain fundamental frequency detector for noisy speech signals. Computer simulations show that the WT-based time and frequency domain pitch period determination algorithms can achieve accurate detection of the pitch period or fundamental frequency for noisy speech.<>