L2 Pronunciation Intelligibility in Google Voice-to-text Applications

Abstract

There is an increase in the application of new technologies in all spheres of human 
endeavour. Industrial 4.0 is one of the recently birthed industrial revolutions through which 
machines understand human speech, thinks and comprehends human intentions. It structures 
critical components for intelligent vehicles, intelligent offices, intelligent service robots, 
intelligent industries, and so on. This furthers the structure of the intelligent ecology of the 
Internet of Things. At the centre of all these is human speech which is used to give order to 
and ask questions from Artificial Intelligence (AI) and robots. Previous studies on AI and 
linguistics have discussed the use of AI in language classroom using AI modelling 
pronunciation to enhance pronunciation performance of the second language learners. This 
study examines Automatic Speech Recognition (ASR) using Word Error Rate (WER) to 
measure the level of intelligibility of the pronunciation of man to machine. It investigates if 
there will be communication breakdown if the pronunciation is not intelligible. To achieve 
this, 30 L2 speakers of English were selected from Igbo, Hausa and Yoruba to read 135 crafted 
words into 5sentences using Google ASR application as primary data. The secondary data was 
drawn from journal articles, textbooks and the Internet. The result showed that the 
pronunciation model used to develop the application has made provision for several L2 
speakers of English in reality of the new world Englishes.