MLLR based speaker adaptation for indian accents

Title

MLLR based speaker adaptation for indian accents

Creator

Balaji V.; Sadashivappa G.

Description

Speech Recognition has become an inherent and important feature of today's mobile based apps. Speech input is a very popular option for people with limitations of using the keyboard / mouse in a computer system. Nowadays, more voice messages are used than written text as they also convey the emotions of the speakers. As solutions are developed with native speakers of a language, many of the English input systems have higher accuracy for native speakers than for people with English as their second language (L2), especially for Asian population. The complexity increases since the accent and intonation of Indian speakers are varied from region to region and state to state. This paper analyses an effective speaker adaptation mechanism implemented with Indian speaker profiles and with a very small amount of adaptation data. This research is to facilitate a speaker adaptive system for the speech disabled users with limited disabilities like stuttering and/or unintelligible speech due to illness like cerebral palsy. Experimental results show improvements in the recognition accuracy for speakers speaking small sentences. 2017 University of Bahrain. All rights reserved.

Source

International Journal of Computing and Digital Systems, Vol-6, No. 5, pp. 293-301.

Date

2017-01-01

Publisher

University of Bahrain

Subject

Automatic speech recognition (ASR); HMM (hidden markov model); Linear transformations; Maximum likelihood linear regression (MLLR); Speaker adaptation

Coverage

Balaji V., Department of Computer Science, Christ University, Bangalore, India; Sadashivappa G., Department of Telecommunication Engineering, R V College of Engineering, Bangalore, India

Rights

All Open Access; Gold Open Access; Green Open Access

Relation

ISSN: 2210142X

Format

Online

Language

English

Type

Article

Identifier

https://doi.org/10.12785/IJCDS/060508

https://www.scopus.com/inward/record.uri?eid=2-s2.0-85104911803&doi=10.12785%2fIJCDS%2f060508&partnerID=40&md5=c9e9cbf48c5864d71fd361c3855030b5