A diagonal plus low-rank covariance model for computationally efficient source separation - Archive ouverte HAL Access content directly
Conference Papers Year : 2017

A diagonal plus low-rank covariance model for computationally efficient source separation

(1) , (2)
1
2

Abstract

This paper presents an accelerated version of positive semidef-inite tensor factorization (PSDTF) for blind source separation. PSDTF works better than nonnegative matrix factoriza-tion (NMF) by dropping the arguable assumption that audio signals can be whitened in the frequency domain by using short-term Fourier transform (STFT). Indeed, this assumption only holds true in an ideal situation where each frame is infinitely long and the target signal is completely stationary in each frame. PSDTF thus deals with full covariance matrices over frequency bins instead of forcing them to be diagonal as in NMF. Although PSDTF significantly outperforms NMF in terms of separation performance, it suffers from a heavy computational cost due to the repeated inversion of big covariance matrices. To solve this problem, we propose an intermediate model based on diagonal plus low-rank covariance matrices and derive the expectation-maximization (EM) algorithm for efficiently updating the parameters of PSDTF. Experimental results showed that our method can dramatically reduce the complexity of PSDTF by several orders of magnitude without a significant decrease in separation performance. Index Terms— Blind source separation, nonnegative matrix factorization, positive semidefinite tensor factorization, low-rank approximation.
Fichier principal
Vignette du fichier
mlsp-2017-liutkus.pdf (941.8 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01580733 , version 1 (01-09-2017)

Identifiers

  • HAL Id : hal-01580733 , version 1

Cite

Antoine Liutkus, Kazuyoshi Yoshii. A diagonal plus low-rank covariance model for computationally efficient source separation. IEEE international workshop on machine learning for signal processing (MLSP), Sep 2017, Tokyo, Japan. ⟨hal-01580733⟩
231 View
1017 Download

Share

Gmail Facebook Twitter LinkedIn More