MODELO DE PROCESAMIENTO DE VOZ PARA LA CLASIFICACIÓN DE ESTADOS

Abstract

In this work, in general terms, we report the progress made in the recognition and classi cation of emotions from voice signals, the objectives proposed for the conclusion and a description of relevant work found so far. Emotion recognition is a problem that has been treated in di erent manners, taking into account various ways of obtaining data, such as reading biometric signals (blood pressure, pulse, among the most important), detection of sequences of movements of the human body walking, monitoring of facial expression, voice signal, among others. There are works dealing with the issue by analyzing di erent types of signals and others who deal with one type of signal, for the purposes of this project, working with the voice signal. Actually, the techniques used for the recognition and classi cation of emotions are neural networks, Gaussian mixtures, Hidden Markov Models, mainly. In this thesis makes use of support vector techniques and associative memories, the combination of both result in an overall bene t of 99% accuracy in the classi - cation of emotions. It uses the database Berlin [2], which is free and with which many researchers have carried out their work and where the reported results have not exceeded 82 %.