Husni, Husniza and Yusof, Yuhanis and Kamaruddin, Siti Sakira (2013) Evaluation of automated phonetic labeling and segmentation for dyslexic children’s speech. In: World Congress on Engineering 2013, 3 - 5 July 2013, London, U.K.

PDF Restricted to Registered users only Download (819kB) | Request a copy

Abstract

Phonetic labeling and segmentation have one major outback – they are time consuming, erroneous, and tedious if done manually.Although manual labeling and segmentation are always the best, automated approach is potentially promising as alternative approach for a more efficient process.In an attempt to automatically label and segment dyslexic children’s read speech, this paper investigates whether or not the automated approach can be as accurate as compared with the manual one. This is due to the highly phonetically similar reading errors produced when they read that have affected automatic speech recognition (ASR).In this work, experiments were performed using a specifically designed ASR to force-align the read speech and produce the labels and segmentations automatically.The CSLU toolkit’s force alignment algorithm has been employed to measure their performances.Selected speech data of dyslexic children’s reading in Malay were fed to the algorithm as input and the evaluation resulted in 95% agreement on phonetic labeling and only 65% on segmentation with respect to the manual ones.

Item Type:	Conference or Workshop Item (Paper)
Additional Information:	Vol II ISBN: 978-988-19252-8-2 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Uncontrolled Keywords:	automatic phonetic labeling, automatic transcription, speech recognition, dyslexic children reading.
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions:	College of Arts and Sciences
Depositing User:	Dr. Husniza Husni
Date Deposited:	02 Sep 2015 01:45
Last Modified:	02 Sep 2015 01:45
URI:	https://repo.uum.edu.my/id/eprint/15361

Actions (login required)

View Item