Real-Time Speech Emotion Recognition

Abstract

This paper presents an analysis of three distinct speech emotion recognition methods, aiming to implement a real-time solution specifically for customer service applications in Morocco. The goal is to enhance the quality of customer relationship management in call centers by providing more convenient and usable emotion recognition tools. We used the CaFE speech dataset and we experimented with three methods. Namely, a Random Forest classifier with 100 decision trees, a three-layer Long Short-Term Memory (LSTM) model, and Multi-Layer Perceptron classifier combined with the state of the art pretrained speech model Wav2Vec for feature extraction. The results show that the best performing model is Random Forests with an accuracy of 85%. We integrate the model in a real-time scenario, showcasing its potential for practical applications in improving customer interactions in Moroccan call centers.