Noise PSD Insensitive RTF Estimation in a Reverberant and Noisy Environment

Research output: Chapter in Book/Conference proceedings/Edited volumeConference contributionScientificpeer-review

35 Downloads (Pure)

Abstract

Spatial filtering techniques typically rely on estimates of the target relative transfer function (RTF). However, the target speech signal is typically corrupted by late reverberation and ambient noise, which complicates RTF estimation. Existing methods subtract the noise covariance matrix to obtain the target plus late reverberation covariance matrix, from where the RTF is estimated. However, the noise covariance matrix is typically unknown. More specifically, the noise power spectral density (PSD) is typically unknown, while the spatial coherence matrix can be assumed known as it might remain time-invariant for a longer time. Using the spatial coherence matrices we simplify the signal model such that the off-diagonal elements are not affected by the PSDs of the late reverberation and the ambient noise. Then we use these elements to estimate the target covariance matrix, from where the RTF can be obtained. Hence, the resulting estimate of the RTF is insensitive to the noise PSD. Experiments demonstrate the estimation performance of our proposed method.
Original languageEnglish
Title of host publicationProceedings of the ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Place of PublicationPiscataway
PublisherIEEE
Number of pages5
ISBN (Electronic)978-1-7281-6327-7
ISBN (Print)978-1-7281-6328-4
DOIs
Publication statusPublished - 2023
Event48th IEEE International Conference on Acoustics, Speech and Signal Processing 2023 - Rhodes Island, Greece
Duration: 4 Jun 202310 Jun 2023

Conference

Conference48th IEEE International Conference on Acoustics, Speech and Signal Processing 2023
Abbreviated titleICASSP 2023
Country/TerritoryGreece
CityRhodes Island
Period4/06/2310/06/23

Bibliographical note

Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

Keywords

  • RTF estimation
  • spatial filter
  • Eigenvalue Decomposition

Fingerprint

Dive into the research topics of 'Noise PSD Insensitive RTF Estimation in a Reverberant and Noisy Environment'. Together they form a unique fingerprint.

Cite this