[Colloquium] Xie/M.S. Presentation/May 12, 2005

[Margaret Jaffey]

This is an announcement of Zhimin Xie's Master's Presentation.

------------------------------------------------------------------------ 
-----
Date:  Thursday, May 12, 2005

Time:  2:00 p.m.

Place:  Ryerson 255

M.S. Candidate:  Zhimin Xie

M.S. Paper Title:  An Acoustic Feature Based Vowel Detector

Abstract:
In the last two decades, statistically-based recognition methods have
dominated the speech recognition field. However, further improvements on
these methods require successful incorporation of speech knowledge.
Recently, a variety of efforts have been made to combine speech  
knowledge
into probabilistic frameworks.

A compact vowel detector, based only on two acoustic features,  
periodicity
and energy, is implemented to show the significance of speech knowledge.
It performs with 86.8\% accuracy and 22.4\% total error rate on TIMIT
data, comparable to a few HMM-based Sphinx models. The robustness of  
this
detector, however, outperforms them. Even under adverse environments,  
like
telephone channel (NTIMIT data) and noisy situations (TIMIT data with
additive white noise), it still works stably with accuracy over 86.0\%  
and
total error rate less than 23.0\%.

Advisor:  Prof. Partha Niyogi

A draft copy of Zhimin Xie's MS Paper is available in Ry 161A.

--  
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Margaret P. Jaffey				margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 161A)		(773) 702-6011
The University of Chicago		http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-