A new perceptual entropy-based method to achieve a signal adapted wavelet tree in a low bit rate perceptual audio coder

This paper outlines a wavelet based audio coding scheme that uses a perceptual entropy criterion to achieve a signal adapted wavelet tree. We are interested in decompositions adapted to the nature of audio signals, but attending to the characteristics of the human hearing system. Our audio coder is compared with two other wavelet based coding schemes: the first one implements a fixed wavelet decomposition which closely mimics the critical band division at the inner ear, while the second one uses an adaptive wavelet tree controlled by a classical entropy criterion, such as Shannon entropy. The results confirm that perceptual entropy-based decomposition is the way you must take if you are looking for maximum compression rates and transparent coding. To achieve a further bit rate reduction ensuring transparent coding, our audio coder performs a filter order optimization, which is possible by using a novel algorithm to translate the psycho-acoustic information from the frequency to the wavelet domain. Experimental results indicate that the proposed coder ensures transparent coding of monophonie CD quality audio signals at bit rates lower than 64 kb/s (1.45 bit/sample) for most audio signal. Subjective listening tests tell us that our coding method behaves as well as ISO/MPEG 1 — layer III and much better than MPEG 2-layer II at that bit rate.