Sinusoidal wavelet transforms for pitch estimation of music signals

Abstract

Pitch is related to fundamental frequency in an audio signal. It is a perceptual measure and depends on the auditory system. Discrete Fourier transforms which are most commonly used provide same resolution through the entire range of frequencies and give the output as an average over the entire bin. Resolution of human auditory system varies logarithmically which maps just as the multiresolution analysis based on wavelet transforms. Sinusoidal wavelet transforms are proposed here to detect two continuously occurring signals to a good resolution both in frequency and time. The accuracy can be increased to a desired value by increasing the number of scaling factors used. The technique can be used for note identification and musical instrument recognition through pattern classification.