AI Framework Links Music Empathy Recognition to Well-Being Ratings

A new computational framework called CMET-Opt aims to recognize empathic emotions evoked by music across listener groups, and then optimize audio output toward well-being-related goals, according to a study published in Current Psychology.

The framework combines three layers of artificial intelligence. ConvNeXt V2 enhances psychoacoustic features extracted from audio, UniSpeech-SAT V2 performs cross-group empathic emotion recognition, and LLaMA-3 handles optimization decisions within a recognition–optimization loop.

In quantitative experiments, CMET-Opt reached an overall accuracy of 89.7 percent, a macro-F1 score of 87.2 percent, and a mean average precision of 90.3 percent, outperforming mainstream baseline models.

A key concern for the authors was cross-group generalization. The framework reduced the cross-group accuracy gap to 3.2 percentage points and improved an algorithmic well-being correlation score to 88.6 percent.

Because model-internal, reward-based evaluation can be misleading, the team also ran a double-blind listening experiment with 42 independent human participants. Optimized audio samples received higher human ratings for cross-group empathic consistency and well-being-related association than the original samples, with statistically significant differences at p below 0.01.

The correlation between algorithmic indicators and human subjective ratings was strong but imperfect, at r equal to 0.87, suggesting the computational measures track human judgments closely yet not completely.

The study acknowledges important limitations. Proxy group labels and conventional classification metrics cannot fully disentangle group-level response bias from genuine perceptual sensitivity, meaning the model may partly reflect labeling artifacts rather than true differences in how groups perceive music.

The authors describe CMET-Opt as a preliminary, empirically supported framework for cross-group music-evoked empathic emotion recognition and optimization, and note that larger-scale human validation remains necessary before making stronger claims about subjective well-being-related association.

Subject of Research: A machine learning framework for cross-group music empathy emotion recognition and audio optimization

Article Title: Cross-group music empathy emotion recognition and optimization model: enhanced psychoacoustic feature and decision optimization

Article References: Tao, L., & Chen, X. (2026). Cross-group music empathy emotion recognition and optimization model: enhanced psychoacoustic feature and decision optimization. Current Psychology, 45(21), Article 1649. https://doi.org/10.1007/s12144-026-10181-x

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10181-x

Keywords: music empathy, emotion recognition, machine learning, psychoacoustics, ConvNeXt V2, UniSpeech-SAT, LLaMA-3, cross-group generalization, well-being, double-blind listening experiment, Current Psychology, Cross-group