Independent Component Analysis Using Spherical Microphone Arrays

Abstract

Spherical microphone arrays provide a new and promising tool for the spatial analysis of complex sound fields. Considerable previous work has investigated the use of spherical microphone arrays for phase-mode beamforming, which relies on the Bessel-weighted spherical harmonic transform of the sound field. In this paper, we investigate the advantages that spherical microphone arrays provide for blind separation of convolutive mixtures using independent component analysis. We demonstrate that applying a standard, linear independent component analysis model in the phase-mode domain enables one to both localize the sources and resolve the permutation problem that plagues most implementations of independent component analysis. As well, we show that the standard linear independent component analysis model can be incorporated into beamforming approaches for source localization and source separation. Simulation results indicate that this approach works in realistic scenarios that include room reverberation.

A dereverberation algorithm for spherical microphone arrays using compressed sensing techniques

“…According to the results, we see that the proposed algorithm outperforms the other techniques in all three simulated shoebox rooms. This is expected since (1) the ICA-based approach is unable to enhance the target signal efficiently at low and high frequencies due to the HOA encoding error [2] and (2) the MUSIC technique can only effectively detect the position of the target signal which makes this approach similar to beamformingbased approaches.…”

“…Spherical harmonic analysis, as used in higher order ambisonics (HOA) [4], is a powerful tool for describing the spatial properties of sound fields. The spherical harmonic expansion of a sound field can be obtained from a spherical microphone array [2,5]. The spherical harmonic expansion, B, of a sound field corresponding to a set of plane waves can be expressed as a simple matrix product:…”

“…There are 12 microphones located on the surface of a rigid sphere with a radius of 3 cm; the other 12 microphones are located on the surface of a open sphere with a radius of 15 cm. Please refer to [2] for a detailed description of the processing.…”

“…These signals were approximately 4 seconds in duration with all of the audio processed at a sampling rate of 16 kHz. We compared the proposed algorithm with the ICA-based approach [2] and the MUSIC (MUltiple SIgnal Classification) algorithm [10] combined with the MINT technique. The performance of these algorithms were evaluated using the Perceptual Evaluation of Speech Quality (PESQ) measure as described in the ITU-T recommendation P.862 [11] and the Segmental Signal-to-Reverberation Ratio (SegSRR) [12].…”

“…As well, working in the spherical harmonic domain has several advantages including scalability and the ability to rotate the sound scene by a simple matrix operation. It is also shown in [2] that a convolutive mixture in time domain can be transformed into a instantaneous mixture in the spherical harmonic domain, which provides significant advantages for source localization and separation [3]. We present a simulation experiment in order to demonstrate the effectiveness of the proposed algorithm and we also compare the performance of the proposed algorithm with other algorithms.…”