Perceptual dissimilarities, requiring high-cost user ratings, have contributed to designing well-distinguishable vibrations for associated meaning delivery. Appropriate metrics can reduce the cost, but known metrics in vibration similarity/dissimilarity could not predict them robustly. We propose a physiology-based model (PM) that predicts the perceptual dissimilarities of a given vibration set via two parallel processes: Neural Coding (NC), mimicking the neural signal transfer, and One-dimensional Convolution (OC), capturing rhythmic features. Eight parameters were trained using six datasets published in the literature to maximize Spearman's Rank Correlation. We validated PM and six metrics of RMSE, DTW, Spectral/Temporal Matchings, ST-SIM, and SPQI in twelve datasets: six trained and six untrained datasets including measured accelerations. In all validations, PM's predictions showed robust correlations with user data and similar structures in perceptual spaces. Other baseline metrics showed better fit in specific datasets, but none of them robustly showed correlations and similar perceptual spaces over twelve datasets.
https://doi.org/10.1145/3544548.3580686
The ACM CHI Conference on Human Factors in Computing Systems (https://chi2023.acm.org/)