Auditory-inspired sparse representation of audio signals

This article deals with the generation of auditory-inspired spectro-temporal features aimed at audio coding. To do so, we first generate sparse audio representations we call spikegrams, using projections on gammatone/gammachirp kernels that generate neural spikes. Unlike Fourier-based representations, these representations are powerful at identifying auditory events, such as onsets, offsets, transients, and harmonic structures. We show that the introduction of adaptiveness in the selection of gammachirp kernels enhances the compression rate compared to the case where the kernels are non-adaptive. We also integrate a masking model that helps reduce bitrate without loss of perceptible audio quality. We finally propose a method to extract frequent audio objects (patterns) in the aforementioned sparse representations. The extracted frequency-domain patterns (audio objects) help us address spikes (audio events) collectively rather than individually. When audio compression is needed, the different patterns are stored in a small codebook that can be used to efficiently encode audio materials in a lossless way. The approach is applied to different audio signals and results are discussed and compared. This work is a first step towards the design of a high-quality auditory-inspired “object-based” audio coder.

■■■Figure optionsDownload as PowerPoint slideResearch highlights
► Adaptive selection of kernels in sparse representations improves their efficiency.
► Design of perceptual-relevant sparse representations enhances audio quality.
► Extraction of audio objects in sparse representations reduces bitrate.