Efficacy of Artificial Intelligence (AI) Voice Cloning in Phonetic Self‐Imitation for L2 Pronunciation Training

ABSTRACT The idea of using a well‐matched and personalized voice (a so‐called golden speaker) in L2 acquisition enables learners to expand their pronunciation repertoire. Phonetic self‐imitation, an accent‐conversion method in which acoustic characteristics in native utterances are extracted and transferred into the learner's speech input, so that an L2 learner mirrors the one's voice synthesized with that of a native speaker, was first proposed over thirty years ago. Since then, a handful of studies have shown that self‐imitation has been shown to be effective for L2 pronunciation improvement. However, this method requires continuous enhancements with new technological capabilities related to the development of neural networks and artificial intelligence (AI). This study examines the effectiveness of selected AI tools (Revoicer and Speechify) using voice cloning in phonetic self‐imitation practice, aiming to investigate whether there is a correlation between this method and the level of L2 fluency and comprehensibility. In an eight‐week pronunciation practice, 21 Polish learners of English performed self‐imitation tasks three times a week (for a total of 45 min per week), involving imitation of AI‐modified utterances using AI tools. Progress was assessed through pre‐, post‐, and delayed post‐tests, evaluated using a seven‐point Likert scale by native English speakers and well‐experienced teachers of English. Results indicate a significant improvement in L2 fluency and comprehensibility among participants using AI‐assisted phonetic self‐imitation. The findings highlight the potential of integrating AI‐driven phonetic self‐imitation practice into L2 learning, offering new opportunities for L2 learners to improve their pronunciation skills and be able to work at their own time and pace.