Compression of high-quality audio signals using adaptive filterbanks and a zero-tree coder

In this work we present a subband coding system for compressing high-quality audio signals. Two alternative subband decomposition schemes are considered: One consists of time-varying analysis and synthesis filterbanks that adapt their structure to the time-varying properties of the input signal; the other is based on decomposition of each block of samples separately. Both schemes are based on the wavelet packets theory and the best-basis selection algorithm. The coder and decoder used in the system are based on the zero-tree algorithm which was originally developed for coding still images and is adapted here for coding audio signals. The performance of the proposed system is shown to be superior to the MPEG-Audio layer I standard, achieving an improvement of up to 10.6 dB in segmental SNR at output bit rate of 64 kbit/sec and up to 6.5 dB at output bit rate of 128 kbit/sec, for a variety of music signals. However, it turns out that most of the improvement is due to the zero-tree coding algorithm.