Warped Discrete Fourier Transform Based on Matching Pursuit

Jian Zhou
2011-01-01
Abstract:Frequency axis warping to achieve nonuniform Fast Fourier Transform resolution uses a network of cascaded first order all-pass sections for frequency warping of the signal,followed by a standard FFT.However,different frequency component in the signal will have different signal delay and the all-pass filter is a IIR filter,so some high frequency component signal will be infinitude and will be truncated before DFT.Inspired by signal representation based on matching pursuit method,we propose the WDFT based on matching pursuit algorithm.The experimental results show a better warp spectrum representation when the number of atom is more than the length of signal.It is more suitable for perceptual speech processing than ordinary WDFT.
What problem does this paper attempt to address?