Hybrid Wavelet Packet and Discrete Cosine Transform with Optimum Bit Allocation Applied to
High-Quality Audio Coding

 Abstract

The wavelet filter bank analysis-synthesis technique has been widely applied to many areas of digital signal processing, especially in image and video coding. In this thesis, we propose a hybrid Wavelet Packet and DCT audio compression system, which divides the audio signal into 26 subbands via Wavelet Packet analysis and selectively performs DCT in each subband according to the flatness measure of time and frequency of this subband. The proposed coder adopts optimum bit allocation with nonideal reconstruction filters to transform the minimum masking threshold in frequency domain obtained from psychoacoustic model into the masking threshold in Wavelet domain. The WP or DCT coefficients are then quantized with uniform quantizers according to masking threshold, so that we can reduce the data rate but still have high quality. Finally, the quantized coefficients are encoded with arithmetic coding and encapsulated with other side information. The experiments show that, only 52 kbps is needed for proposed audio coder to achieve MP3 64-kbps quality. At the same bit rate of 64 kbps, the proposed audio coding system can provide not only better quality than MP3 and AAC LC profile but also superior to AAC HE profile!