Advanced Acoustic Monitoring Using Psychoacoustic Heatmap Machine Learning Models for Noise Impact Prediction in Air-Conditioned Building Environments

Air-conditioning systems are vital for indoor environmental quality. However, noise can offset its benefits, making acoustic monitoring important. Recent research revealed that sound quality perceptions can be described by three psychological dimensions: <i>Evaluation</i>, <i>Potency</i>, and <i>Activity</i> (EPA). This is the first study to develop psychoacoustic heatmap machine learning models (PHMLM) for predicting sound quality and the negative noise impacts (<i>O1</i>: <i>Discomfortable</i>, <i>O2</i>: <i>Annoying</i>, <i>O3</i>: <i>Stressful</i>, and <i>O4</i>: <i>Unacceptable</i>) of air conditioning sounds using a 227 × 227-pixel psychoacoustic heatmap as input for machine learning. A total of 1208 jury listening tests were conducted with 101 participants on 30 s soundtracks from air-conditioned environments. Psychoacoustic heatmaps were generated by converting time-varying psychoacoustic metrics (<i>N</i>, <i>S</i>, <i>R</i>, and <i>FS</i>) into intensity maps containing 51,529 pixels of multidimensional acoustic information. The PHMLMs achieved predictive performance with correlation coefficients of 0.79, 0.80, and 0.62 for <i>E</i>-, <i>P</i>-, and <i>A</i>-scores, respectively. Compared to traditional regression models (TRM), PHMLM-EPA demonstrated significantly better performance with 31% lower mean absolute error (4.4 vs. 6.4) and higher regression slope (0.798 vs. 0.587). Moreover, PHMLM-EPA demonstrated a higher goodness-of-fit than TRM (+55% to +95%) and traditional acoustic metric <i>L</i>_Aeq (+87% to +95%). The approach offers an advanced acoustic monitoring method for sustainable building designs.