Discriminative training of GMM-HMM acoustic model by RPCL learning

This paper presents a new discriminative approach for training Gaussian mixture models(GMMs)of hidden Markov models(HMMs)based acoustic model in a large vocabulary continuous speech recognition(LVCSR)system.This approach is featured by embedding a rival penalized competitive learning(RPCL)mechanism on the level of hidden Markov states.For every input,the correct identity state,called winner and obtained by the Viterbi force alignment,is enhanced to describe this input while its most competitive rival is penalized by de-learning,which makes GMMs-based states become more discriminative.Without the extensive computing burden required by typical discriminative learning methods for one-pass recognition of the training set,the new approach saves computing costs considerably.Experiments show that the proposed method has a good convergence with better performances than the classical maximum likelihood estimation(MLE)based method.Comparing with two conventional discriminative methods,the proposed method demonstrates improved generalization ability,especially when the test set is not well matched with the training set.

Experimental Study of Discriminative Adaptive Training and MLLR for Automatic Pronunciation Evaluation

A stronger canonical model was developed to improve the performance of automatic pronunciation evaluations. Three different strategies were investigated with speaker adaptive training to normalize variations among speakers, minimum phone error training to identify easily confused phones and maximum likelihood linear regression （MLLR） adaptation to compensate for accent variations between native and non-native speakers. The three schemes were combined to improve the correlation coefficient between machine scores and human scores from 0.651 to 0.679 on the sentence level and from 0.788 to 0.822 on the speaker level.

Discriminative tonal feature extraction method in mandarin speech recognition

To utilize the supra-segmental nature of Mandarin tones, this article proposes a feature extraction method for hidden markov model （HMM） based tone modeling. The method uses linear transforms to project Fo（fundamental frequency） features of neighboring syllables as compensations, and adds them to the original Fo features of the current syUable. The transforms are discriminatively trained by using an objective function termed as ＂minimum tone error＂, which is a smooth approximation of tone recognition accuracy. Experiments show that the new tonal features achieve 3.82% tone recognition rate improvement, compared with the baseline, using maximum likelihood trained HMM on the normal F0 features. Further experiments show that discriminative HMM training on the new features is 8.78% better than the baseline.

Stream Weight Training Based on MCE for Audio-Visual LVCSR

In this paper we address the problem of audio-visual speech recognition in the framework of the multi-stream hidden Markov model. Stream weight training based on minimum classification error criterion is discussed for use in large vocabulary continuous speech recognition (LVCSR). We present the lattice re- scoring and Viterbi approaches for calculating the loss function of continuous speech. The experimental re- sults show that in the case of clean audio, the system performance can be improved by 36.1% in relative word error rate reduction when using state-based stream weights trained by a Viterbi approach, compared to an audio only speech recognition system. Further experimental results demonstrate that our audio-visual LVCSR system provides significant enhancement of robustness in noisy environments.