Télécom Paris


GET
EURASIP
HINDAWI
Dual-Microphone Robust Front-End for Arm's-Length Speech Recognition
Author(s)
  • Holly Francois, (Motorola)
  • David Pearce, (Motorola)
  • James Rex, (Motorola)
Topics
  • Noise reduction techniques
  • 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
This paper describes a novel method of improving speech recognition performance in non-stationary background noise. A two-microphone array both enhances speech and provides a continuous estimate of background noise. This processing is integrated with the ETSI standard Advanced Front-End for Distributed Speech Recognition, with the continuous noise estimate being used to design the first-stage Wiener filter. This increases noise suppression and hence reduces recognition error rate by up to 50% compared with single-microphone input.

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