Publication:
Analysis of accent-sensitive words in multi-resolution mel-frequency cepstral coefficients for classification of accents in Malaysian English
Analysis of accent-sensitive words in multi-resolution mel-frequency cepstral coefficients for classification of accents in Malaysian English
Date
2013-06
Authors
M.A. Yusnita
M.P. Paulraj
Sazali Yaacob
R. Yusuf
Shahriman Abu Bakar
Journal Title
Journal ISSN
Volume Title
Publisher
Research Projects
Organizational Units
Journal Issue
Abstract
This paper investigates the most accent-sensitive words for Malaysian English (MalE) speakers in multi-resolution 13 Mel-frequency cepstral coefficients. A text-independent accent system was implemented using different numbers of Mel-filters to determine the optimal settings for this database. Then, text-dependent accent systems were developed to rank the most accent-sensitive words for MalE speakers according to the classification rates. Prior work has also been conducted to test the significance of the wordlist for both gender and accent factors, and to investigate any interaction between these two factors. Experimental results show that male speakers have a higher intensity of accent effects compared with female speakers by 3.91% on text-independent and 3.47% on text-dependent tasks. Another finding has proven that by selecting appropriate words that carry severe accent markers could improve the task of speaker accent classification. An improvement of at most 8.45% and 8.91% was achieved on the male and female datasets, respectively, following vocabulary selection.
Description
Keywords
Accent classification,
Analysis of variance,
Factorial design,
K-nearest neighbors,
Malaysian english,
Mel-frequency cepstral coefficients