Researchers from the University of Aizu received the Best Application Paper Award at IEEE iCAST 2026.

When accented speech beats native one?

Most TTS systems are built to sound as native as possible. Together with Li Tang(Doctoral program), and Prof. Julian Villegas trained a system to sound like a Japanese person speaking English.
We did this by taking an English speech synthesizer and fine-tuning it using recordings from Japanese learners of English. Then, we tested it on 14 native Japanese listeners. The results were encouraging: Listeners understood the Japanese-accented sentences better than the native-accented ones, and they consistently recognized the accent as Japanese.
This work is a step toward real-time accent conversion. Take air traffic control, for example: English is the mandated language for communication. Adapting the accent of an instruction to match what a controller or pilot is used to hearing could ease communication, without the extra time and complexity a translation step would add.

This work received the Best Application Paper Award on September 24, 2026, at the 2026 International Conference on Awareness Science and Technology (iCAST).

URL conference: https://icast2026.org.cn/
URL lab: https://onkyo.u-aizu.ac.jp/