Perceptual zerotrees for scalable wavelet coding of wideband audio

Abstract

This paper introduces a new algorithm for scalable coding of wideband audio signals. The technique is based on quantization of bi-orthogonal wavelet transformed coefficients using a perceptual zerotree method. An initial zerotree estimate of the wavelet coefficients is computed, followed by scalar quantization of the coefficients according to perceptual thresholds. The choice of wavelet decomposition and encoding parameters for each frame is adapted to the source characteristics employing a rate distortion criterion. The scalability of the coder is due to the tree structure, which enables graceful degradation with decrease in bit rate. Preliminary subjective tests indicate near-transparent quality for average bit rates in the range of 1.5 to 2.5 bits per sample.