Skip to main navigation Skip to search Skip to main content

Kronecker product overdetermined independent vector analysis for online blind source separation

  • Kang Chen
  • , Xianrui Wang
  • , Yichen Yang
  • , Andreas Brendel
  • , Gongping Huang
  • , Zbyněk Koldovský
  • , Jingdong Chen
  • , Jacob Benesty
  • , Shoji Makino
  • Wuhan University
  • Northwestern Polytechnical University Xian
  • Fraunhofer Institute for Integrated Circuits
  • Technical University of Liberec
  • Institut national de la recherche scientifique
  • Waseda University

Research output: Contribution to journalArticlepeer-review

Abstract

Online blind source separation is essential for real-time speech communication and human-machine interaction. Among existing solutions, overdetermined independent vector analysis (OverIVA) has demonstrated superior separation performance by jointly exploiting source independence and the orthogonal constraint between the source and noise subspaces. However, when employed with large microphone arrays, the dimensionality of the separation filters grows rapidly, which poses significant challenges for reliable parameter estimation in online implementations. To address this limitation, this paper proposes a generalized Kronecker product decomposition for online OverIVA, in which each source extraction filter is represented as a sum of multiple Kronecker products of shorter sub-filters. The proposed formulation introduces an explicit Kronecker order that flexibly controls the model rank, enabling a smooth trade-off between representation capability and parameter efficiency. An efficient alternating optimization algorithm based on iterative projection is derived to estimate the structured extraction filters in an online manner. Simulation results demonstrate that the proposed method achieves improved separation performance and enhanced robustness.

Original languageEnglish
Article number110880
JournalSignal Processing
Volume251
DOIs
StatePublished - Feb 2027
Externally publishedYes

Keywords

  • Kronecker product decomposition
  • Online blind source separation
  • Overdetermined independent vector analysis

Fingerprint

Dive into the research topics of 'Kronecker product overdetermined independent vector analysis for online blind source separation'. Together they form a unique fingerprint.

Cite this