A LOW-COMPLEXITY SPEECH DEREVERBERATION METHOD BASED ON KRONECKER PRODUCT DECOMPOSITION

Xiaojin Zeng, Hongsen He, Jingdong Chen

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The weighted-prediction-error (WPE) method, which uses variance-normalized multichannel delayed linear prediction to suppress the late reverberation component, is one of the most effective algorithms for speech dereverberation. The prediction-error filter in WPE, however, has to be long enough to estimate accurately the late reverberation component, which results in a high computational complexity. To deal with this issue, this paper proposes to use Kronecker product to decompose the high-dimensional multichannel linear prediction-error filter into three groups of low-dimensional filters. This scheme not only leads to effective dereverberation performance, but also reduces the computational load, which is very conducive to real-time speech systems.

Original languageEnglish
Title of host publicationProceedings of the 28th International Congress on Sound and Vibration, ICSV 2022
PublisherSociety of Acoustics
ISBN (Electronic)9789811850707
StatePublished - 2022
Event28th International Congress on Sound and Vibration, ICSV 2022 - Singapore, Singapore
Duration: 24 Jul 202228 Jul 2022

Publication series

NameProceedings of the International Congress on Sound and Vibration
ISSN (Electronic)2329-3675

Conference

Conference28th International Congress on Sound and Vibration, ICSV 2022
Country/TerritorySingapore
CitySingapore
Period24/07/2228/07/22

Keywords

  • computational complexity
  • Kronecker product decomposition
  • multichannel linear prediction
  • speech dereverberation
  • weighted-prediction-error

Fingerprint

Dive into the research topics of 'A LOW-COMPLEXITY SPEECH DEREVERBERATION METHOD BASED ON KRONECKER PRODUCT DECOMPOSITION'. Together they form a unique fingerprint.

Cite this