Classificação de Variações Acústicas Emocionais com Atributos da Fonte e do Trato Vocal

This article presents a study on the classification of multiple emotional acoustic variations using the following features: vector of Hurst coefficients (pH), Mel-Frequency Cepstral Coefficients (MFCC) and Gammatone-Frequency Cepstral Coefficients (GFCC). For the analysis, two databases are used in English language, recorded in different contexts. The classification is performed by employing two classifiers: Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM). Results indicate that the excitation source feature (pH) is more efficient than the vocal tract ones (MFCC and GFCC) in characterizing the emotional variations. Regarding the classifiers, the GMM was the most efficient in modeling each emotional state.