Cantonese Tone Recognition Using the Hilbert-Huang Transform

MPhil Thesis Defence


Title: "Cantonese Tone Recognition Using the Hilbert-Huang Transform"

By

Mr. Ying Fung LAM


Abstract

Cantonese is a very popular spoken language/dialect. When compared to 
other tonal languages, Cantonese is well known for its rich set of 9 tones 
and similar tone contours between tones. Automated tone recognition of 
Cantonese is very challenging. Hilbert-Huang Transform (HHT) is a newly 
developed empirical algorithm that works on non-stationary and nonlinear 
signals. In this study, we examine the HHT algorithm for its performance 
on Cantonese tone recognition for isolated syllables. Firstly, HHT is used 
as a frequency detection tool applied to syllables from the CUSYL corpus. 
Experimental results show a 25% improvement in the accuracy of the 
fundamental frequency detection compared to peak picking Fast Fourier 
transform. Secondly, we improve both the performance and the accuracy of 
the HHT on the CUSYL corpus by experimenting with various parameters used 
by the core component of HHT, the Windowed Average-based Empirical Mode 
Decomposition (WA-BASED EMD). Finally, Support Vector Machines (SVM) are 
used as binary classification tools. Pitch track information obtained by 
HHT together with tone information from the CUSYL corpus is used to train 
a set of 6 SVMs with more than 1000 syllables. Experimental results show a 
78.5% speaker-independent tone recognition rate for Cantonese isolated 
syllables. The result is favorable compared to the results of other 
studies.


Date:			Friday, 10 January 2014

Time:			10:00am – 12:00noon

Venue:			Room 3501
 			Lifts 25/26

Committee Members:	Dr. David Rossiter (Supervisor)
 			Prof. Andrew Horner (Chairperson)
 			Dr. Brian Mak


**** ALL are Welcome ****