Sixth International Conference on Spoken Language Processing
In this paper, we address the problem and the use of multiple classifiers for robust recognition over the cellular network. The idea is to provide more variability to the system to be trained, and to support this variability with more number of model parameters. The main drawback is that the model size, and the computational complexity increases linearly related to different call environment. To alleviate this problem we first introduce a new measure called the average-arc-count into the decoding process. The main ad- vantage of this new measure is that many of the multiple classifiers can be shut down during the recognition stage if the average-arc-count of individual classifier exceeds a certain threshold limit for a given utterance. Secondly, we can also build individual classifiers with less number of parameters and without degrading the overall system performance. Experimental results on English connected digit recognition task show a string error rate reduction of as much as 40% by using the multiple classifiers when compared to individual CDMA systems.
Bibliographic reference. Chengalvarayan, Rathinavelu (2000): "Use of multiple classifiers for speech recognition in wireless CDMA network environments", In ICSLP-2000, vol.3, 382-385.