An improved warped DFT algorithm based on signal sparse representation

Warped Discrete Fourier Transform(WDFT) offers an alternative to increase frequency resolution at any selected parts of spectral axis without extending the number of sampling points. In order to have a stable and robust spectrum, the warp coefficient must be selected carefully, and this is not allowd in many applications, such as speech processing, since the warp coefficient is computed by human auditory principle. Inspired by sparse signal representation theory, we propose a WDFT algorithm based on matching pursuit methodology. The experimental results show that a better warp spectrum representation and a better signal reconstruction performance are got when the number of basic signal is more than the length of signal. It is more suitable for perceptual speech processing than ordinary WDFT.