Robust speech recognition method based on discriminative learning of environmental features
Jiqing Han, Munsung Han, Gyu-Bong Park, Jeongue Park, Chengfa Wang · 2003
Learning the influence of additive noise and channel distortions from training data is an effective approach for robust speech recognition. We have proposed a novel method of discriminative learning of environmental features according to minimum classification error (MCE) criterion in previous work, in which additive noise is expressed by the weighted combination of multiple types of noises, and the channel distortions are assumed to consist of the channel distortions of the whole training data and the current utterance. In this paper, we use a Gaussian distribution to stand for the distribution of additive noise, and adaptively learn the combination factors of the channel distortions. The current method is proven better than the former one by experiments.