Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Online Overdetermined Independent Vector Analysis Based on Bilinear Decomposition

Kang Chen, Xianrui Wang, Yichen Yang, Andreas Brendel, Gongping Huang, Zbyněk Koldovský, Jingdong Chen, Jacob Benesty, Shoji Makino

TL;DR

This work tackles online blind source separation in overdetermined microphone arrays by addressing parameter explosion in OverIVA through a bilinear (Kronecker-based) decomposition of each demixing filter. An alternating iterative projection framework updates the two coupled sub-filters, enabling robust online estimation with far fewer parameters. Empirical results on large arrays show approximately 10 dB improvements in SIR and SDR over conventional methods, validating the approach for real-time speech separation in noisy, reverberant environments. The BiIVA method thus offers a practical path toward scalable, robust online BSS in high-channel-count scenarios.

Abstract

Online blind source separation is essential for both speech communication and human-machine interaction. Among existing approaches, overdetermined independent vector analysis (OverIVA) delivers strong performance by exploiting the statistical independence of source signals and the orthogonality between source and noise subspaces. However, when applied to large microphone arrays, the number of parameters grows rapidly, which can degrade online estimation accuracy. To overcome this challenge, we propose decomposing each long separation filter into a bilinear form of two shorter filters, thereby reducing the number of parameters. Because the two filters are closely coupled, we design an alternating iterative projection algorithm to update them in turn. Simulation results show that, with far fewer parameters, the proposed method achieves improved performance and robustness.

Robust Online Overdetermined Independent Vector Analysis Based on Bilinear Decomposition

TL;DR

This work tackles online blind source separation in overdetermined microphone arrays by addressing parameter explosion in OverIVA through a bilinear (Kronecker-based) decomposition of each demixing filter. An alternating iterative projection framework updates the two coupled sub-filters, enabling robust online estimation with far fewer parameters. Empirical results on large arrays show approximately 10 dB improvements in SIR and SDR over conventional methods, validating the approach for real-time speech separation in noisy, reverberant environments. The BiIVA method thus offers a practical path toward scalable, robust online BSS in high-channel-count scenarios.

Abstract

Online blind source separation is essential for both speech communication and human-machine interaction. Among existing approaches, overdetermined independent vector analysis (OverIVA) delivers strong performance by exploiting the statistical independence of source signals and the orthogonality between source and noise subspaces. However, when applied to large microphone arrays, the number of parameters grows rapidly, which can degrade online estimation accuracy. To overcome this challenge, we propose decomposing each long separation filter into a bilinear form of two shorter filters, thereby reducing the number of parameters. Because the two filters are closely coupled, we design an alternating iterative projection algorithm to update them in turn. Simulation results show that, with far fewer parameters, the proposed method achieves improved performance and robustness.
Paper Structure (9 sections, 23 equations, 2 figures, 1 algorithm)

This paper contains 9 sections, 23 equations, 2 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Signal model
  3. Problem Formulation
  4. Online OverIVA
  5. Proposed Method: Online OverIVA with Bilinear Decomposition
  6. Simulations
  7. Simulation Setup
  8. Results and Analysis
  9. Conclusions

Figures (2)

  • Figure 1: Convergence performance of the compared source separation algorithms: (a) SIR improvement and (b) SDR improvement. Scores are computed over $2$-second segments and averaged across $50$ samples and two speakers.
  • Figure 2: Spectrograms of the observed signal, the target signal, and the separated signals obtained using AuxIVA, OverIVA, and the proposed BiIVA method.