Blue Archive TTS Model Gets Natural Voice Upgrade With Style-Bert-VITS2
The shift to a model that treats punctuation as phonemes and uses style vectors marks a concrete technical step toward more emotionally responsive character voices in a live mobile game.
Reporting from 1 sources: 4Gamer.net.
At NEXON Korea's NDC26 conference, NEXON Games engineer Kim Myeong-ji detailed the development of a new TTS model for Blue Archive. The team adopted Style-Bert-VITS2, an open-source Japanese-focused model, to improve emotional expression and naturalness in character voices. The upgrade targets better reproduction of nuances like distress and joy, addressing specific feedback on long vowels and pitch in Japanese lines.
NEXON Games engineer Kim Myeong-ji presented the development of a new TTS model for Blue Archive at NDC26, the company's annual developer conference. The team selected Style-Bert-VITS2, an open-source model built for Japanese, after testing multiple candidates against criteria for character personality, emotional expression, and naturalness. Kim noted that Japanese lines required special attention to long vowels and pitch accents, which affect meaning in the language. The model treats punctuation marks as phonemes and uses a style vector to vary tone, moving beyond the single-tone generation of standard TTS. At the session, attendees heard a side-by-side comparison of Arona's voice from the early model and the current one, with Kim reporting improved reproduction of emotions such as distress.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.