Fine-grain scalable audio coding based on envelope restoration and the SPIHT algorithm

We present strategies for perceptual improvements of embedded audio coding based on psychoacoustic weighting and spectral envelope restoration. The encoding schemes exhibit fine-grain bitrate scalability via the set partitioning in hierarchical trees (SPIHT) algorithm. Weighting factors and envelope parameters are transmitted under careful consideration of the amount of side information. For low bitrates, where the number of actually transmitted waveform coefficients is low, missing coefficients are shaped w.r.t. the spectral envelope. In our approach, the envelope information is transmitted in form of band-wise values of the l₁-norm. Sets of standardized audio files as well as various audio data of contemporary music are encoded and the results are analyzed with objective measures of perceptual quality. The proposed coding scheme competes in perceptual quality with existing state-of-the-art fixed bitrate coders such as MPEG-2/4 AAC. For low bitrates, the proposed embedded coding envelope restoration (ECER) improves the perceptual audio quality notably.