Real-time speech emotion recognition using acoustic features and a probabilistic heuristic classifier

Abstract

This paper presents a real-time speech emotion recognition system based on acoustic feature extraction and a probabilistic-heuristic classifier, intended for intelligent monitoring and the management of exceptional situations. The proposed model formalizes the process of voice signal analysis by segmenting the signal into short frames, extracting a vector of relevant acoustic features, and combining probabilistic inference with heuristic rules to determine the emotional class. The system was implemented on a resource-constrained embedded platform using the ESP32 microcontroller and the INMP441 digital microphone, enabling near-real-time voice signal acquisition and processing, as well as communication via Wi-Fi and Bluetooth. Experimental validation included 50 tests based on short utterances spoken under different emotional states. The system achieved recognition rates of 87% for anger, 92% for fear, 48% for sadness, 94% for the neutral emotional state, and 75% for panic. These results confirm the functionality and feasibility of the proposed solution, as well as its potential for integration into intelligent systems for the early detection of relevant affective states in critical contexts.