An adaptive stereo basis method for convolutive blind audio source separation

Abstract

We consider the problem of convolutive blind source separation of stereo mixtures, where a pair of microphones records mixtures of sound sources that are convolved with the impulse response between each source and sensor. We propose an Adaptive Stereo Basis (ASB) source separation method for such convolutive mixtures, using an adaptive transform basis which is learned from the stereo mixture pair.The stereo basis vector pairs of the transform are grouped according to the estimated relative delay between the left and right channels for each basis, and the sources are then extracted by projecting the transformed signal onto the subspace corresponding to each group of basis vector pairs. The performance of the proposed algorithm is compared with FD-ICA and DUET under different reverberation and noise conditions, using both objective distortion measures and formal listening tests. Preprint submitted to Elsevier Science 10 August 2007NOTICE: this is the author's version of a work that was accepted for publication in Neurocomputing. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in : Neurocomputing 71(10-12), 2087-2097, June 2008. doi: 10.1016/j.neucom.2007 The results indicate that the proposed stereo coding method is competitive with both these algorithms at short and intermediate reverberation times, and offers significantly improved performance at low noise and short reverberation times.

Subjective and Objective Quality Assessment of Audio Source Separation

“…A number of studies have been performed to assess the subjective quality of certain source separation schemes [10], [11], [12], [13], [14], [15], [16], [17]. Most studies consider either a single criterion, such as overall quality [10], [14], [15], preference [13, p. 138] or musical noise salience [12], [18], or a set of criteria restricted to speech [11], [16].…”

“…Independent Component Analysis (ICA) in [13], time-frequency masking in [10] or simulated separation in [16], and a narrow range of sound material, e.g. male speech in [14], [17] or isolated notes from a single musical instrument in [15]. The resulting scores can hence not be compared due to the lack of a common absolute reference.…”

“…In parallel to the subjective studies [10], [11], [12], [13], [14], [15], [16], [17], the objective evaluation of source separation algorithms has also received some attention. A common approach to evaluating the quality of an estimated source signal is to compute the Signal to Distortion Ratio (SDR) between the energy of the reference, i.e.…”

Fast Dictionary Learning for Sparse Representations of Speech Signals

“…Double sparsity refers to seeking a sparse decomposition and a dictionary D = AB such that the atoms in A are sparse over the fixed dictionary B, such as Wavelets or the discrete cosine transform (DCT). Also, in previous results in [17], it was found that dictionary atoms learned from speech signals with a sparse coding method based on ICA (SC-ICA) [18], are localized in time and frequency. This appears to suggest that for certain types of signals (e.g.…”

A theoretical discussion on the foundation of Stone’s blind source separation

“…Each row of W is a 1 × M un-mixing vector w i related to one of extracted signals. Many strategies have been used for BSS [6,[8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25]. One of them is projection pursuit [6] that is based on the central limit theorem (CLT) [7].…”