Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multichannel blind speech source separation with a disjoint constraint source model

Jianyu Wang, Shanzheng Guan

TL;DR

This work addresses MBSS by exploiting speech sparsity more explicitly. It introduces a disjoint constraint source model using a Bingham prior on basis vectors and a Laplace prior on activations, integrated into MNMF and ILRMA to form $s$-MNMF and $s$-ILRMA, with optimization via majorization-minimization and auxiliary functions. Empirical results on two-channel simulations with varying reverberation show notable SDR/SIR gains over strong baselines, demonstrating improved robustness to reverberation and better exploitation of sparse spectral structure. The approach provides a principled way to augment existing MBSS frameworks with sparsity priors, potentially benefiting real-world multichannel speech separation tasks.

Abstract

Multichannel convolutive blind speech source separation refers to the problem of separating different speech sources from the observed multichannel mixtures without much a priori information about the mixing system. Multichannel nonnegative matrix factorization (MNMF) has been proven to be one of the most powerful separation frameworks and the representative algorithms such as MNMF and the independent low-rank matrix analysis (ILRMA) have demonstrated great performance. However, the sparseness properties of speech source signals are not fully taken into account in such a framework. It is well known that speech signals are sparse in nature, which is considered in this work to improve the separation performance. Specifically, we utilize the Bingham and Laplace distributions to formulate a disjoint constraint regularizer, which is subsequently incorporated into both MNMF and ILRMA. We then derive majorization-minimization rules for updating parameters related to the source model, resulting in the development of two enhanced algorithms: s-MNMF and s-ILRMA. Comprehensive simulations are conducted, and the results unequivocally demonstrate the efficacy of our proposed methodologies.

Multichannel blind speech source separation with a disjoint constraint source model

TL;DR

This work addresses MBSS by exploiting speech sparsity more explicitly. It introduces a disjoint constraint source model using a Bingham prior on basis vectors and a Laplace prior on activations, integrated into MNMF and ILRMA to form -MNMF and -ILRMA, with optimization via majorization-minimization and auxiliary functions. Empirical results on two-channel simulations with varying reverberation show notable SDR/SIR gains over strong baselines, demonstrating improved robustness to reverberation and better exploitation of sparse spectral structure. The approach provides a principled way to augment existing MBSS frameworks with sparsity priors, potentially benefiting real-world multichannel speech separation tasks.

Abstract

Multichannel convolutive blind speech source separation refers to the problem of separating different speech sources from the observed multichannel mixtures without much a priori information about the mixing system. Multichannel nonnegative matrix factorization (MNMF) has been proven to be one of the most powerful separation frameworks and the representative algorithms such as MNMF and the independent low-rank matrix analysis (ILRMA) have demonstrated great performance. However, the sparseness properties of speech source signals are not fully taken into account in such a framework. It is well known that speech signals are sparse in nature, which is considered in this work to improve the separation performance. Specifically, we utilize the Bingham and Laplace distributions to formulate a disjoint constraint regularizer, which is subsequently incorporated into both MNMF and ILRMA. We then derive majorization-minimization rules for updating parameters related to the source model, resulting in the development of two enhanced algorithms: s-MNMF and s-ILRMA. Comprehensive simulations are conducted, and the results unequivocally demonstrate the efficacy of our proposed methodologies.
Paper Structure (9 sections, 18 equations, 1 figure)

This paper contains 9 sections, 18 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Signal model and problem formulation
  3. Proposed Sparseness Model and Source Separation Algorithm
  4. Cost Function
  5. Source Separation Algorithm
  6. Simulations
  7. Simulation Setup
  8. Simulation Results
  9. Conclusions

Figures (1)

  • Figure 1: SDR and SIR improvements of the studied methods.