Phase-Optimized K-SVD for Signal Extraction from Underdetermined Multichannel Sparse Mixtures

TL;DR

This paper introduces a phase-optimized K-SVD algorithm that learns a joint spectral-spatial dictionary for extracting signals from underdetermined multichannel mixtures, demonstrated on robot ego-noise reduction.

Contribution

It presents a novel joint spectral-spatial dictionary learning method that estimates phases and sources in underdetermined multichannel mixtures, improving signal extraction.

Findings

Outperforms conventional dictionary methods in real-room recordings

Effectively captures spatial and spectral features jointly

Enhances robot audition by reducing ego-noise

Abstract

We propose a novel sparse representation for heavily underdetermined multichannel sound mixtures, i.e., with much more sources than microphones. The proposed approach operates in the complex Fourier domain, thus preserving spatial characteristics carried by phase differences. We derive a generalization of K-SVD which jointly estimates a dictionary capturing both spectral and spatial features, a sparse activation matrix, and all instantaneous source phases from a set of signal examples. The dictionary can then be used to extract the learned signal from a new input mixture. The method is applied to the challenging problem of ego-noise reduction for robot audition. We demonstrate its superiority relative to conventional dictionary-based techniques using recordings made in a real room.