• Acoustic comfort is a fundamental perception of the sound environment. • Facial expressions represent authentic emotional responses occurring in a natural state. • Facial expression recognition can serve as a valuable tool for soundscape assessment. • Visual information enhanced the valence positive change and reduced the arousal level. • There is a significant positive correlation between acoustic and valence chances. Acoustic comfort is a key perceptual dimension of soundscape quality and plays an important role in human health and well-being. Conventional acoustic comfort assessment methods mainly rely on subjective questionnaires or contact-based physiological measurements, which are limited in efficiency, intrusiveness, and real-time applicability. From a soundscape perspective, this study investigates how different sound sources influence acoustic comfort perception and explores the feasibility of using facial expressions as a non-invasive indicator of acoustic comfort. Participants' facial expressions, emotional responses, and subjective acoustic comfort ratings were obtained under audio-only and audio-visual conditions, with stimuli drawn from common geophysical, biological, human, and mechanical sound sources. The results reveal distinct patterns of emotional valence, arousal, and acoustic comfort across sound source categories, highlighting the close relationship between emotional responses and perceived acoustic comfort. To support the analysis of facial expression dynamics related to acoustic comfort, a two-branch differential network is employed to mitigate individual differences in facial expressions and enhance prediction robustness. Experimental results indicate that the acoustic comfort of subjects can be reliably reflected by facial expressions, with the prediction accuracy of 91.25% for subject-dependent and 83.59% for subject-independent. These findings demonstrate the potential of facial expressions as a human-centered, non-invasive tool for acoustic comfort assessment and contribute to the development of data-driven soundscape evaluation methods.
Wu et al. (Mon,) studied this question.