Exploiting the Sparsity of the Sinusoidal Model Using Compressed Sensing for Audio Coding

Anthony Griffin (1) , Christos Tzagkarakis (1) , Toni Hirvonen (1) , Athanasios Mouchtaris (1) , Panagiotis Tsakalides (1)

1 ICS-FORTH - Institute of Computer Science [FORTH, Heraklion]

Audio signals are represented via the sinusoidal model as a summation of a small number of sinusoids. This approach introduces sparsity to the audio signals in the frequency domain, which is exploited in this paper by applying Compressed Sensing (CS) to this sparse representation. CS allows sampling of signals at a much lower rate than the Nyquist rate if they are sparse in some basis. In this manner, a novel sinusoidal audio coding approach is proposed, which differs in philosophy from current state-of-the-art methods which encode the sinusoidal parameters (amplitude, frequency, phase) directly. It is shown here that encouraging results can be obtained by this approach, although inferior at this point compared to state-of-the-art. Several practical implementation issues are discussed, such as quantization of the CS samples, frequency resolution vs. coding gain, error checking, etc., and directions for future research in this framework are proposed.

Domains: Signal and Image Processing; Signal and Image processing

Anthony Griffin, Christos Tzagkarakis, Toni Hirvonen, Athanasios Mouchtaris, Panagiotis Tsakalides. Exploiting the Sparsity of the Sinusoidal Model Using Compressed Sensing for Audio Coding. SPARS'09 - Signal Processing with Adaptive Sparse Structured Representations, Inria Rennes - Bretagne Atlantique, Apr 2009, Saint Malo, France. ⟨inria-00369613⟩