A Model for Pitch Estimation Using Wavelet Packet Transform Based CEPSTRUM Method

Abstract

A computationally efficient model for pitch estimation of mixed audio signals is presented. Pitch estimation plays a significant role in music audition like music information retrieval, automatic music transcription, melody extraction etc. The proposed system consists of channel separation and periodicity detection. The input signal is created by mixing two sound signals. First removes the short time correlations of the mixed signal. The model divides the signal into number of channels using wavelet packet transform. Computes the cepstrum of each channels and sums the cepstrum functions. The summary cepstrum function is further processed to extract the pitch frequency of two input signal separately. The model performance is demonstrated to be comparable to those of recent multichannel models. The proposed system can be verified by simulating the system in MATLAB.