M-band wavelet packets and filter bank trees as flexible tools in audio signal processing

This article discusses M-band wavelet packets which combine the well-known construction of 2-band wavelet packets with concepts of M-band wavelet theory. To make the resulting tilings of the time-frequency plane even more flexible, the concept of a filter bank tree (FBT) is presented. Within this framework the design of decimated filter bank cascades, realizing some arbitrary time-frequency tiling, is possible. Extensive tests on the denoising of high fidelity audio signals and comparison with several standard wavelet packet denoising techniques show the advantages of the new methods. The generality of the concept suggests its application to other problems, not necessarily restricted to the field of audio signal processing.