舞蹈即兴创作与音乐实时互动匹配算法研究

打开文本图片集

文章编号：1003-6180（2026）02-0064-07

[中图分类号]TP391.41

[文献标志码]A

Research on the Algorithm for Real-Time Interactive Matching of Dance Improvisation and Music

ZHANG Zhaoxing

（Anqing Normal University School of Music， Anqing 246000， China）

Abstract： A real-time interactive matching algorithm that integrates a Generative Adversarial Network （GAN） is proposed. This algorithm employs a Spatio-Temporal Graph Convolutional Network （ST-GCN） to extract multimodal features encompassing rhythmic aesthetics and physical motion laws. It models the human skeleton as a non-Euclidean graph structure and constructs feature parameters based on the division of centripetality and centrifugality. A dual-tower Transformer architecture grounded in contrastive learning is adopted to establish cross-modal alignment mapping， optimizing the shared latent space of music and motion features through a temporal warping alignment layer and rhythmic consistency constraints. Additionally， a Conditional Variational Autoencoder （CVAE） and probabilistic motion primitives are introduced， combined with Kalman filtering， to generate adaptive motion trajectories. These trajectories are then used to drive virtual characters through forward kinematics. Experimental results demon-strate that the algorithm exhibits no model-penetration phenomena across various dance motions. The generated motions are highly consistent with the music rhythm， unified in style， and effectively follow the musical states， showcasing superior matching generalization capabilities.

Keywords： dance improvisation creation; real-time interaction with music; multi-modal; cross-domain features; adaptive trajectory

舞蹈即兴创作指一种无预设编排、以即时灵感与身体反应为核心的特殊舞蹈创作方式，舞者通过环境、情绪等直接反应生成舞蹈结构 [1] ，打破原本的舞蹈表演固定模式，注重身体与外部刺激的实时互动，具有即时性、不可预测性、开放性，创作基于音乐驱动，在艺术表达与创新教育等场景应用广泛.

研究舞蹈即兴创作与音乐实时互动匹配算法是目前的重点。（剩余7487字）