Accelerated Convolutive Transfer Function-Based Multichannel NMF Using Iterative Source Steering

2025-09-30

AI Summary

CTF-MNMF achieves excellent performance in highly reverberant blind source separation, but its iterative projection (IP) update requires matrix inversion for each source, resulting in high computational complexity and limiting practical deployment. To address this, we propose Iterative Source-Oriented Fast CTF-MNMF (ISO-CTF-MNMF), which eliminates matrix inversions by introducing a source-oriented update rule that operates entirely without inversion operations. The method preserves the convolutional transfer function modeling and multi-channel nonnegative matrix factorization framework while enabling efficient filter coefficient updates. Experiments on speech separation tasks demonstrate that ISO-CTF-MNMF matches or surpasses the original IP-based method in separation quality—e.g., achieving 0.3–0.8 dB improvement in SI-SNR—while reducing per-iteration computational complexity by approximately 60%. This substantial efficiency gain significantly enhances real-time capability and practical deployability.

Technology Category

• Machine Learning: Matrix & Tensor Methods
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Search and Optimization: Non-convex Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Among numerous blind source separation (BSS) methods, convolutive transfer function-based multichannel non-negative matrix factorization (CTF-MNMF) has demonstrated strong performance in highly reverberant environments by modeling multi-frame correlations of delayed source signals. However, its practical deployment is hindered by the high computational cost associated with the iterative projection (IP) update rule, which requires matrix inversion for each source. To address this issue, we propose an efficient variant of CTF-MNMF that integrates iterative source steering (ISS), a matrix inversion-free update rule for separation filters. Experimental results show that the proposed method achieves comparable or superior separation performance to the original CTF-MNMF, while significantly reducing the computational complexity.

Problem

• Reduces computational cost of blind source separation
• Eliminates matrix inversion in filter updates
• Maintains performance in reverberant environments

Innovation

• Uses iterative source steering for efficient separation
• Eliminates matrix inversion to reduce computational cost
• Maintains performance in highly reverberant environments