Innovative Study of Japanese Speech Teaching Models in ChatGPT Application Scenarios
Gangjie Song, Wanzhi Ma · 2024
We investigated a Japanese speech dialogue practice system based on ChatGPT to provide personalized language learning. We simulated Japanese speech, including standard pronunciation, different accents, and speech rate variations for training speech dialogue generation models. Using speech synthesis techniques, we implemented diverse virtual dialogues to mimic the speech characteristics of different speakers. The system was integrated with a pronunciation assessment system to measure students’ pronunciation accuracy and provide them with instant feedback through speech recognition technology. We studied how to make ChatGPT better understand and generate speech dialogues, considering contextual, affective, and daily conversation scenarios. Through real-time interactions and feedback, students can have a personalized learning experience in the speech dialogues. Students had a positive attitude towards this voice dialog practice system and thought that it provided a fun and effective way to learn.