Emotion Analysis Using Auditory ASMR via Physiological Signals and Federated Learning

Human emotions are the interplay of thought processes and perceptions. Autonomous Sensory Meridian Response (ASMR) offers a unique window into this dynamic interplay by eliciting transient emotional responses. These responses, detectable through wearable biosensors, can provide valuable insights into automated human emotions. This study delves into the physiological effects of auditory ASMR stimuli to detect emotion arousal levels. It collects Electrodermal Activity (EDA) and Heart Rate (HR) physiological data from 23 participants exposed to happiness and calmness-inducing ASMR audios. The study employs statistical data analysis to differentiate between the emotional arousals induced by the auditory stimuli. It proposes a novel AI-driven human emotion cognition system named FED-ASMR utilizing Federated Learning (FL) to preserve users' privacy and sensitive physiological data. It uses a Multi-Layer Perceptron (MLP) neural network to classify arousal levels and distinguish between happy and calm emotions, achieving an accuracy of 88.80 %. This research addresses privacy concerns and lays the groundwork for utilizing ASMR audios for human emotion cognition systems.