Computing of masking thresholds for audio coders based on a quaternionic 4-band wavelet packet transform

Abstract

A wavelet packet based on 4-band building blocks was used to implement an auditory model for 44.1kHz and 16kHz sampling frequency. The underlying paraunitary filter bank is implemented using a quaternionic lattice structurally insensitive to the quantization of its coefficients. Both the linear phase and orthogonality are possible for 4-band wavelets, so a better perceptual quality can be expected and an increased compression ratio for the coders based on the proposed solution in comparison to standard 2-band wavelet packets or a warped DFT transform. These features and a low computational complexity predestinate this approach to be a tempting alternative to widely known solutions.