Improve English Pronunciation at Sentence Level for Thai EFL Learners With Thai Automatic Speech Recognition Model

Abstract

ASR (Automatic Speech Recognition) is favorably chosen as a learning technology, which is used for English pronunciation practice. However, the base ASR model usually does not recognize the accent of the EFL (English as a Foreign Language) learner, especially in the sentence level. This research aims to extend the platform to improve English pronunciation from word level to sentence level for Thai EFL learners using a fine-tuned ASR with Thai datasets to detect Thai accent mispronounced sounds. The sentences are classified with CEFR level to provide learning steps for learners. The twelve of Grade 12 Thai native students were selected as sampling process. The results show that 75% of the samples have improved their pronunciation according to the CEFR level after using our system. Furthermore, the verb's present and past forms are most problematic.