Reconhecimento de Emoções na Fala em Português usando Redes Neurais

Abstract

Speech emotion recognition has become an important area within affective com puting due to its application in systems aiming for more natural human-machine interac tions. This monograph presents the development of a system for recognizing emotions in Portuguese audio using deep neural networks, specifically Convolutional Neural Networks (CNN) and Long Short-Term Memory Networks (LSTM). The main challenge in automatic emotion recognition lies in the variability of emo tional expressions among individuals and cultures, in addition to the inherent difficulties in extracting and analyzing prosodic and acoustic features. The proposed methodology aims to overcome these limitations by utilizing the emoUERJ dataset, which contains Portuguese recordings expressing different emotions. Through the analysis of spectro grams and the use of signal processing techniques, the developed model was tested in both noise-free and noisy scenarios, achieving significant results. The experiments indicate that the combination of CNNs and LSTMs offers robust performance, allowing for the automatic extraction of relevant features directly from raw data. The proposed model demonstrates effectiveness in the task of emotion recognition in Portuguese speech.