The emotion recognition systems have become important for the diversity of its applications. Several methodologies have been proposed based on how emotions are reflected in biological systems, such as facial expressions, the activity of the nervous system or the prosody of voice. The detection of emotions by voice processing is an approach that involves a noninvasive procedure that produces results with an acceptable rate of detection. In this work an algorithm for features extraction was developed, that efficiently classify different emotional states. Thus, emotions that have not been trained can be associated with a trained emotion both belonging to the same region of the valence-arousal plane.