Análise do sinal de fala para reconhecimento de emoções utilizando representação semântica

Abstract

The interest in automatically analyzing human emotions has grown in recent years, mainly due to its possible applications in society. Over the last 20 years, several works were proposed in order to extract features in speech signals capable of representing emotions. Most works use features associated to prosody and the spectrum of the speech signal. In this work, speech analysis methodologies are presented, exploring mainly pre-processing, feature extraction and classification techniques. Furthermore, a methodology based on semantic representation is used in order to obtain a non-linear mapping of the feature space. Then, it is possible to use the semantic representation to add complementary information and, consequently, increase the discrimination between the classes. The results obtained show that semantic representation is able to significantly increase the accuracy in the task of speech emotion recognition.