LINKING SENSORY PERCEPTION OF CRISPNESS TO RMS-BASED ACOUSTIC ENERGY USING ARTIFICIAL NEURAL NETWORKS

Crispness is a sensory attribute associated with brittle food materials and is closely related to acoustic emissions generated during fracture events. Its perception is influenced by the intensity and temporal evolution of sound signals produced during biting or mastication. In this study, the relationship between sensory perception of crispness and acoustic energy was evaluated, and artificial neural networks were applied to model and predict crispness from recorded fracture sounds. Acoustic signals were obtained using a manually operated lever system coupled to a dental prosthesis to simulate incisal compression, and recordings were acquired at sampling rate of 44.1 kHz. Acoustic characterization was performed using Root Mean Square (RMS) energy derived from the spectral representation of the signals, providing a frequency-resolved measure of acoustic pressure associated with fracture events. Sensory evaluation was conducted using trained panelists based on recorded audio signals played through headphones, allowing the assessment of crispness perception from auditory stimuli. Multilayer Perceptron (MLP) neural networks were developed using RMS-based features as input. The models showed stable convergence and consistent performance across different network configurations. Using RMS-based acoustic features, the model enabled reliable discrimination among samples and achieved satisfactory agreement with sensory scores (R² ≈ 0.75; MAE ≈ 0.89), indicating that acoustic energy provides relevant information for describing crispness perception. The results indicate that RMS-based acoustic energy captures meaningful characteristics of fracture events and supports the approximation of sensory perception through data-driven models. The use of a physically interpretable acoustic descriptor combined with a simple neural network architecture provides a consistent framework for relating fracture acoustics to perceived crispness and supports the development of quality control sensors and instrumental alternatives to human sensory testing. This study was supported by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Brazil, under Grant No. 400982/2025-0.