Reconfiguring local characteristics in contemporary Muju: an AI-assisted study of vocal style in Chinese local opera.

Background: Muju () is a local opera rooted in Chun'an County, Zhejiang Province, China. Since the re-establishment of its professional troupe in 2015, its vocal style has changed substantially, driven by the recruitment of young performers trained in other opera genres (Huangmeixi, Yueju, Wuju). Previous research has described this change through historical narratives and insider accounts, but the extent to which cross-genre training may shape vocal production has remained insufficiently examined with empirical methods. Objective: This study examines how performers' prior vocal training in other operatic genres may affect vocal production when performing Muju, and how this process may have contributed to changes in the genre's local vocal characteristics over time. Methods: A total of 691 vocal excerpts were analyzed, including 196 Muju excerpts across three historical stages and 495 reference excerpts from Yueju, Huangmeixi, and Wuju. A three-layer Transformer-decoder model was used as part of an AI-assisted, score-based acoustic-proximity analysis. The model-derived scores were treated as exploratory indices of relative acoustic proximity rather than as validated measures of stylistic similarity. In addition, a case-based acoustic comparison of the same aria was conducted among three performances: a historical male Muju/Sanjiaoxi reference, a contemporary Huangmeixi-trained female performer, and a contemporary Chun'an local female reference. Results: Model-derived scores suggest that Contemporary Muju may show lower relative proximity to Traditional Sanjiaoxi than Old Muju does (Cohen's Conclusion: Taken together, the findings are consistent with the possibility that the vocal profile of contemporary Muju has been reconfigured rather than replaced. They are also compatible with the possibility that cross-genre training contributed to this process through the transfer of habitual vocal-production patterns. The acoustic case study provides a focused performance-level illustration, whereas the AI-assisted score-based analysis offers exploratory corpus-level context requiring cautious interpretation.