Title: Bayesian Adaptation and Combination of Deep Models for Automatic Speech Recognition
Dr. Chin-Hui Lee, ECE, Chair , Advisor
Dr. Biing Hwang Juang, ECE
Dr. Mark Clements, ECE
Dr. Geoffrey Li, ECE
Dr. Sabato Marco Siniscalchi, Enna Kore
The objective of the proposed research is to deploy a Bayesian adaptation framework for deep model based ASR systems to combat the degradation of the recognition accuracy, which is typically observed under potential mismatched conditions between training and testing. This dissertation addresses the problem in three directions. The first direction is to perform Bayesian adaptation directly on the discriminative DNN models. Maximum a posteriori estimation and multi-task learning techniques are employed in the manner of regularization in the DNN updating formula. In the second direction, we try to cast the DNN into a generative framework to better leverage Bayesian techniques. Classic structured MAP adaption is adopted by using bottleneck features derived from deep neural networks. In the third direction, we employ a hierarchical Bayesian system combination technique to further enhance the adaptation performance by leveraging the complementarity of the discriminative and generative adaptive models.