Télécom Paris

Hybrid Dereverberation Using Blind Deconvolution and Spectral Subtraction to Compensate for Motion of Source
  • Ken'ichi Furuya, (NTT Cyver Space Laboratories, NTT Corporation)
  • Akitoshi Kataoka, (NTT Cyver Space Laboratories, NTT Corporation)
  • Microphone arrays and array signal processing
  • Sound enhancement and sound separation
Get the paper in PDF format
Acrobat Reader (version 5 minimum) is necessary to read this document.

A hybrid dereverberation method for speech enhancement in a situation requiring adaptation where a speaker shifts his head and impulse responses are frequently changed under reverberant conditions is presented. We combine MINT-based blind deconvolution with modified spectral subtraction of the estimation error of inverse filters obtained in practice. Our method computes inverse filters by estimating the correlation matrix between input signals that can be observed, without measuring room impulse responses. The transient performance of the proposed method in the adaptation is demonstrated with experiments using measured room impulse responses.

©2006 Télécom Paris/TSI
Edition : Télécom Paris -- 2006