Télécom Paris


GET
EURASIP
HINDAWI
Inverse Filtering for Speech Dereverberation Less Sensitive to Noise
Author(s)
  • Takafumi Hikichi, (NTT Communication Science Laboratories)
  • Marc Delcroix, (NTT Communication Science Laboratories)
  • Masato Miyoshi, (NTT Communication Science Laboratories)
Topics
  • 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.

Abstract
Inverse filtering of room transfer functions (RTFs) is an attractive approach for speech dereverberation because high performance could be achieved. However, speech signals received at the microphones may suffer from disturbances such as RTF fluctuations and interfering noise. In such cases, the dereverberation performance may be severely degraded. In our previous report, we proposed reducing the energy of the inverse filter to make the filter less sensitive to RTF fluctuations. This paper evaluates this design method for the inverse filter in terms of the filter's sensitivity to additive noise. The experimental results show that the proposed method is effective in the presence of additive noise, as well as RTF fluctuations.

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