A study of switching state segmentation in segmental switching linear Gaussian hidden Markov models for robust speech recognition

摘要

In our previous works, a switching linear Gaussian hidden Markov model (SLGHMM) and its segmental derivative, SSLGHMM, were proposed to cast the problem of modeling a noisy speech utterance in robust automatic speech recognition by a well-designed dynamic Bayesian network. An important issue of SSLGHMM is how to specify a switching state value for each frame of the feature vector in a given speech utterance. In this paper, we propose several approaches for addressing this issue and compare their performance on Aurora3 connected digit recognition tasks.