Introducing difficulty-levels in pronunciation learning

This paper presents a method to introduce the notion of a student selected difficulty level for the task of pronunciation learning of a second language. Three difficulty levels, novice, acceptable and native, are constructed using a tri-/biand unigram language model each using a specific set of broad phonetic groups determined using underspecification. The evaluation of the experiment is based on a target language of American English where Nigerian, Hungarian, Indonesian, Syrian, Pakistani and Portuguese students participate. The results show that the addition of difficulty levels returns specific mis-pronunciation information for a student indicating what pronunciation practice is required before they go to the next difficulty level.