Process Modeling and Emotional Response Simulation of Musical Tension Evolution

Modeling the dynamic relationship between musical tension and emotional response remains challenging because existing approaches typically treat tension prediction and emotion recognition as independent tasks. This study proposes a process-oriented computational framework that integrates cognitive expectation with continuous emotional response simulation through hierarchical signal modeling and dynamic coupling mechanisms. A twelve-dimensional tension descriptor is extracted from symbolic scores and audio signals, while a Transformer-XL-enhanced IDyOM model estimates note-level surprise to characterize expectation violations. The resulting features are recursively encoded by a gated recurrent unit with expectation-aware modulation to generate continuous tension trajectories. Furthermore, a conditional variational autoencoder models listener-specific cognitive sensitivity, and a differential-equation-based coupling mechanism links tension evolution with arousal and valence dynamics to establish a closed-loop perception–expectation–response simulator. Experimental results demonstrate a peak localization error of 0.42 s, an arousal tracking RMSE of 0.152, a valence tracking RMSE of 0.176, and statistically significant causal effects of expectation modulation on simulation accuracy. The proposed framework provides an effective strategy for temporal signal fusion, dynamic state estimation, and adaptive response modeling, offering valuable references for intelligent audio signal processing, multimodal information perception, and computational sensing systems in electromagnetic signal analysis and next-generation interactive media applications.