A preliminary speech analysis for recognizing emotion

Abstract

Some speech analysis to extract emotion from voice is discussed. An emotional Malay and English voice database has been developed, consisting six basic emotions namely happiness, sadness, disgust, fear, anger and surprise. As the target is content independent emotion recognition, 4 short sentences that have the most natural meaning is adopted for the illustration and analysis. A study on speech prosody is done to identify the emotional features of voice. Variation on the sample's energy, duration, and pitch for different emotions is compared. Spectrogram analysis is done on some samples to observe the effect of formant. It is found that duration, average energy and pitch can provide some indication of emotional content of a speech, but it is not enough to correctly represent the emotions. Even though there are slightly different pattern for English and Malay samples, it is still reasonable to assume that there are standard acoustic configurations in expressing particular emotions.