Novel model for pitch estimation using hybrid DWT-DCT HPS

Abstract

Pitch is an important feature of speech. Therefore, extraction of pitch becomes a vital task for processes like speaker coding, speaker recognition, speech synthesis, speech recognition and many such applications. The few available algorithms such as Discrete Cosine Transform (DCT) based pitch extraction, harmonic product spectrum (HPS) which is obtained from DCT are useful for extraction of pitch. In this paper, we propose a hybrid Discrete Wavelet Transform-Discrete Cosine Transform (DWT- DCT HPS) based pitch extraction. A voice sample is taken and de-segmented into 36 bands in the frequency domain. Then on those bands spatial domain transformation is performed to get the most prominent features. The Gross pitch error (GPE) and Fine pitch error (FPE) criteria is used as a measure to find the accuracy of the novel method. The result depicts that the novel proposed Hybrid method is better as compared to DCT-HPS in terms of Pitch error.