This study presents a methodology for developing an automated support system in the classification of phonographic signals (PCG). First, the PCG signals were preprocessed. You then decomposed by the decomposition technique empirically (EMD) with some of its variants and vibration analysis by decomposition of Hilbert (HVD) independently, where the computational cost and the error was compared in the reconstruction of the original signal generating constructs from IMFs. Then the characteristics of the statistical moments data generated by the Hilbert-Huang Transform (HHT), plus cepstral coeffcients at frequencies of Mel (MFCC) and four of its variants were extracted. Finally, a subset of features was selected using sets of fuzzy approximation (FRS), principal component analysis (PCA) and floating sequential forward selection (SFFS) simultaneously to be used as inputs to the hidden Markov model (HMM) ergodic adjusted particle swarm optimization (PSO), in order to provide an objective and accurate to improve reliability in detecting heart murmurs mechanism, obtaining results in the classification of about 96% with sensitivity values higher 0.8 and higher specificity to 0.9, using cross-validation (70/30 split with 30 fold)