Title :
Method for adaptive on-line data fusion in multi-channel automatic speech recognition systems
Author_Institution :
Dept. of Comput. Syst. & Technol., Tech. Univ. of Gabrovo, Bulgaria
Abstract :
An this paper describes a method for adaptive, parameter-independent, online channels´ weight estimation, based on entropy in the output probabilities from ANN classifiers, rather than the noise level or SNR estimation. A recursive formula for channel combination calculation, based on approximation of all possible channels combination, is deduced. The proposed method has been used to develop multi-channel distributed speech recognition (DSR) system. From experiments a conclusion can be drawn that the use of proposed method result in absolute system accuracy improvement of 12.4% in comparison with the base one-channel system from ETSI Aurora Project.
Keywords :
neural nets; sensor fusion; speech recognition; ETSI Aurora Project; adaptive on-line data fusion; adaptive parameter-independent on-line channels weight estimation; base one-channel system; multi-channel automatic speech recognition systems; multi-channel distributed speech recognition system; output probabilities; Acoustic noise; Automatic speech recognition; Entropy; Fusion power generation; Noise generators; Noise level; Noise reduction; Signal to noise ratio; Speech recognition; Telecommunication standards;
Conference_Titel :
Intelligent Systems, 2002. Proceedings. 2002 First International IEEE Symposium
Print_ISBN :
0-7803-7134-8
DOI :
10.1109/IS.2002.1044280
