Extending Fine-Grain Scalable Audio Coding to Very Low Bitrates using Overcomplete Dictionaries

Signal representations in overcomplete dictionaries are considered here as an alternative to the traditional transform representations for fine-grain scalable audio coding. Such representations produce sparser decompositions and thus allow better coding efficiency than transform coding at very low bitrates. Moreover, the decomposition algorithms are intrinsically progressive, and flexible enough to allow an efficient transient modeling. We propose in this paper a fine-grain scalable audio coder which works on a large range of bitrates (2kbs to 128kbs). Objective measures as well as informal subjective evaluation show that this coder outperforms a comparable transform-based coder at very low bitrates.