Complex Wavelet Modulation Sub-Bands and Speech
Jean-Marc Luneau, Lebrun, Jerome, Søren Holdt Jensen · 2008
A new class of signal transforms called Modulation Transforms has recently been introduced. They add a new dimension to the classical time/frequency representations, namely the modu-lation spectrum. Although very efficient to deal with different applications like feature extraction, speech recognition and also analysis for audio coding, these transforms show their limits e.g. when used to remove non-trivial noise from speech signals. Modulation sub-band decompositions based on the computation of the Hilbert envelope have been proved to create disturbing ar-tifacts. We detail a new method to deal properly with the phase and the magnitude of the modulation spectrum in a linear and analytic framework based on a complex wavelet transform. This Complex Wavelet Modulation Sub-Band transform gives some interesting results in speech denoising and proposes a new ap-proach for analytic signal processing in general. Index Terms: speech analysis, complex wavelets, modulation spectrum, denoising, phase signal.