Comparison of perceptual features efficiency for automatic identification of emotional states from speech.