Please use this identifier to cite or link to this item:
Title: An Empirical Study of Tone Models for Robust Speech Recognition of Isolated Thai Words
Authors: Prukkanon, Nutthacha 
Miyanaga, Yoshikazu 
Issue Date: 2007
Publisher: University of the Thai Chamber of Commerce
University of the Thai Chamber of Commerce
Source: Nutthacha Prukkanon, Yoshikazu Miyanaga (2007) An Empirical Study of Tone Models for Robust Speech Recognition of Isolated Thai Words. UTCC Engineering Research Papers.
Journal: UTCC Engineering Research Papers
Abstract: Robust speech recognition system is necessary inreal environment, whereas tone in Thai speech recognitionsystem is an importance as well. Therefore, we attempt to findtone model for enhancing Thai speech recognition system innoisy environment. In this paper we present an empirical studyby configuration fundamental frequency contour, which one ofprosodic information for tone language. Fundamental frequency(F0) and its first order time derivatives (dF0) are composed tothree tone models: F0, dF0, and F0_dF0. They are appended withnon-tonal speech recognition system that used RSF/DRAalgorithm to improve noise robust recognition. We evaluatedtone models using seven words in five different tones with threetypes of noise from NOISEX-92; babble, car, and white noises,resulting in a corpus of 6,370 utterances. Experimental resultsshow that the F0_dF0 model gives significantly improve phonetic,tonal, and word recognition.
ISSN: 1906-1625
Rights: This work is protected by copyright. Reproduction or distribution of the work in any format is prohibited without written permission of the copyright owner.
Appears in Collections:EN: Journal Articles

Files in This Item:
File Description SizeFormat 
986fulltext.pdf521.97 kBAdobe PDFThumbnail
Show full item record Recommend this item

Page view(s)

checked on Jul 11, 2019

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.