A preliminary speech learning tool for improvement of African English accents

摘要

Speech recognition systems emphasise: accent recognition, recognition system performance through calculation of word error rate (WER), pronunciation modelling, speech-based interactions (tone, pitch, volume, background noise, speaker's gender and age, speaking speed and quality of recording equipment) and speech database solutions. However, research into the use of speech recognition systems for improvement accents is scarcely available. In this paper, we focus on development of an speech recognition system for recognizing African English accents and enabling the speakers improve their English accents. This is achieved by using a dual speech recognition engine: the first, a multiple accent recogniser receives African English speech input, classifies it and sends to the second recogniser that evaluates the speech against standard English pronunciations. Speech deviations from standard English pronunciations are captured and read by the system as a way of supporting the learner to improve his/her reading proficiency. Preliminary tests indicate that terminologies that are rarely used in ordinary conversations (e.g. enthusiasm, exuberant, vague, etc) are most poorly pronounced irrespective of the educational level of the reader.