A Survey on Context-dependent Acoustic Modeling in Automatic Speech Recognition

PhD Qualifying Examination


Title: "A Survey on Context-dependent Acoustic Modeling in Automatic
Speech Recognition"

by

Mr. Yu Ting KO


Abstract:

In 1990, it was demonstrated by KaiFu Lee that using context-dependent modeling 
units can significantly improve the recognition accuracy in automatic speech 
recognition (ASR). After his successful work, context-dependent phone models 
(also called triphones) have become the most popular modeling units in ASR 
system for more than 20 years.

However, since using context-dependent units needs much more model parameters, 
trainability becomes a challenge because of data sparsity. As a result, a great 
deal of effort has gone into balancing trainability and accuracy of the 
acoustic model. Among different proposed techniques, parameter sharing approach 
has dominated the field for more than 20 years because of their effectiveness 
in limiting the growth of model parameters without decreasing the accuracy. 
Recently, various alternative modeling techniques are proposed in order to 
further improve the accuracy.

By reviewing the techniques of context-dependent acoustic modeling in this 
survey, we aim to learn the experience and lessons from the past. More 
importantly, we hope to identify the potential directions for further research 
in context-dependent acoustic modeling.


Date:                   Wednesday, 29 June 2011

Time:                   2:00pm - 4:00pm

Venue:                  Room 4475
                         lifts 25/26

Committee Members:	Dr. Brian Mak (Supervisor)
                         Prof. Nevin Zhang (Chairperson)
 			Prof. Siu-Wing Cheng
 			Dr. Raymond Wong


**** ALL are Welcome ****