Using TTS to Develop Phonetic Training Materials

Abstract

This study investigated whether TTS-generated speech can serve as effective training material in High Variability Phonetic Training (HVPT) and explored the learning patterns reflected in the perceptual results. A group of native Chinese undergraduates completed six sessions of self-paced, take-home HVPT for English vowels. Their vowel perception performance was measured with identification (ID) and discrimination (DISC) tasks before and after training. The results showed that accuracy on both tasks at posttest was significantly higher than at pretest, and the training did not have interactions with vowels or vowel contrasts in either task, suggesting that TTS-generated material is effective and that the training effect is not specific to certain vowels or vowel contrasts. Moreover, learners with lower initial ID performance benefited more from training, whereas the improvement in the DISC task did not depend on initial performance. These findings indicate that TTS-generated speech can serve as effective HVPT material for L2 vowel perception, with benefits that vary with learners’ initial performance and task type.