Reconstruction of room impulse responses over an extended spatial domain using block-sparse and kernel regression methods

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

164 Downloads (Pure)

Abstract

The acquisition of the spatio-temporal properties of a sound field over large spatial domains is a challenging task that requires a large number of transducers. Alternatively to these classical approaches, sound field reconstruction methods utilise fewer measurements and enable the interpolation and extrapolation of the measured data. This study exploits the spatio-temporal characteristics of the room impulse response (RIR) to reconstruct the sound field over large spatial apertures. The direct sound field and the early reflections are reconstructed using block-sparsity techniques, and therefore no assumption is made upon the wavefronts. Finally, the late reverberation of the RIR is evaluated using kernel ridge regression, taking advantage of the generalisable stochastic properties of random wave fields. The proposed model reconstructs accurately the RIRs and outperforms classical approaches that are based on the use of plane waves, which are used here as a benchmark for comparison.
Original languageEnglish
Title of host publicationProceedings of the 24th International Congress on Acoustics
Number of pages8
Publication date2022
Publication statusPublished - 2022
Event24th International Congress on Acoustics - Hwabaek International Convention Center, Gyeongju, Korea, Republic of
Duration: 24 Oct 202228 Oct 2022
Conference number: 24
https://ica2022korea.org

Conference

Conference24th International Congress on Acoustics
Number24
LocationHwabaek International Convention Center
Country/TerritoryKorea, Republic of
CityGyeongju
Period24/10/202228/10/2022
Internet address

Keywords

  • Sound field reconstruction
  • Room impulse response
  • Kernel ridge regression

Fingerprint

Dive into the research topics of 'Reconstruction of room impulse responses over an extended spatial domain using block-sparse and kernel regression methods'. Together they form a unique fingerprint.

Cite this