Computationally efficient wavelet packet coding of wide-band stereo audio signals

Abstract

This paper presents a new audio compressor based on the wavelet packet (WP) decomposition. The major drawback of the present audio compressors is the large computational effort associated with subband decomposition and psychoacoustic modeling. We integrate the psychoacoustic model with the design of the decomposition filterbank which separates the wideband signal into 28 subbands closely approximating the critical bands. The psychoacoustic model exploits noise masking and joint stereo coding to compress the subband signals. We demonstrate that the WP decomposition provides sufficient resolution to extract the time-frequency characteristics of the input signal. The WP based audio compressor provides transparent sound quality at compression rates comparable to the MPEG compressor with less than one third of the computational effort.