High Quality Wavelet-Packet Based Audio Coder with Adaptive Quantization

In this paper, we propose an audio coder for high quality low bit-rate applications based on the wavelet-packet transform. The primary goal of the proposed coder will be to combine high quality reconstructed audio with shorter delay than existing methods while keeping the bit rate low. Present audio coders, such as MPEG, are capable of transparent or near-transparent monophonic CD quality at 128kbps or sometimes 64kbps. Some recent wavelet packet methods can further reduce bit-rate, but at the expense of higher computational complexity or lower quality. Reductions in computational effort are of considerable value in real-time applications. In our proposed coder, the wavelet packet decomposition tree structure will be chosen to match the critical bands determined by a psychoacoustic model. For each 2-band cell of the wavelet packet tree, a different length wavelet filter will be used. A simplified Fourier domain method will be used to determine the masking threshold for each critical band. To take advantage of the masking effect in human hearing, a scalar quantization table is determined adaptively by comparing the threshold against the quantization error obtained from the inverse wavelet packet transform.