HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Conference papers

On-line compensation for non-stationary noise

Vincent Barreaud 1 Irina Illina 1 Dominique Fohr 1
1 PAROLE - Analysis, perception and recognition of speech
INRIA Lorraine, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : An improvement of a previously proposed frame-synchronous noise compensation algorithm is presented. The previous algorithm used Stochastic Matching approach and was designed to cope with time-varying noise. The improved version of this algorithm takes into account the abrupt changes in the acoustical environment. The environment change is detected using the Shewart Control Charts detection algorithm that searches for the changes in the means of Gaussian sequence. For each noisy environment a specific mismatch function is estimated. A simple bias is used as mismatch function. After detection of environment change, the bias is reinitialized by the value corresponding to the closest noisy environment previously observed. This technique is general since no assumption is made on the nature, level and occurring time of noise. Our algorithm is evaluated on the VODIS database (speech in moving car). All the test sentences were corrupted by an additive noise occurring at the middle of each sentence. This latest version of our algorithm gives significant improvement compared to classical compensation algorithms.
Document type :
Conference papers
Complete list of metadata

Contributor : Publications Loria Connect in order to contact the contributor
Submitted on : Tuesday, September 26, 2006 - 9:41:40 AM
Last modification on : Friday, February 26, 2021 - 3:28:06 PM


  • HAL Id : inria-00099837, version 1



Vincent Barreaud, Irina Illina, Dominique Fohr. On-line compensation for non-stationary noise. Automatic Speech Recognition and Understanding Workshop - ASRU'2003, Nov 2003, St Thomas, US Virgin Islands, 4 p. ⟨inria-00099837⟩



Record views