Acoustic Analysis of Nigerian English Vowels Based on Accents

Abstract

Accent has been widely acclaimed to be a major source of automatic speech recognition (ASR) performance degradation. Most ASR applications were developed with native English speaker speech samples not minding the fact that the majority of its potential users speaks English as a second language with a marked accent. Nigeria like most nations colonized by Britain, speaks English as official language despite being a multi-ethnic nation. This work explores the acoustic features of energy, fundamental frequency and the first three formats of the three major ethnic groups of Nigerian based on features extracted from five pure vowels of English obtained from subjects who are Nigerians. This research aimed at determining the differences or otherwise between the pronunciations of the three major ethnic nationalities in Nigeria to aid the development of ASR that is robust to NE accent. The results show that there exist significant differences between the mean values of the pure English vowels based on the pronunciation of the three major ethnics: Hausa, Ibo, and Yoruba. The differences can be explored to enhance the performance of ASR in recognition of NE.