Towards modelling the acoustic differences between the diphthongs of Afrikaans and English L1 and L2 speech

This paper presents aspects of a study which aims to model the acoustic differences between first language (L1) and second language (L2) speech. Although humans can distinguish between accents fairly easily, they are difficult to determine computationally and quantifiably. The ultimate goal of this study is to enable improvements in automatic speech recognition (ASR) systems by applying the models to assist in speaker adaptation. The so called “true” diphthongs found in Afrikaans are addressed in the study