Figure 6

Distribution of the feature AHSM versus time for two audio utterances sampled, respectively, from q2 and q3 modes of VAM [16]. q2 and q3 locate at the positive and the negative sides of the valence dimension, respectively. It can be observable that the feature AHSM provides a clear distinction between q2-q3 on the valence dimension. q3 is more likely to have tonation fluctuations in negative valence mode (i.e., sadness) which avoids q3 mode to have a harmonic structure as smooth as the q2 signal (i.e., neutral). The periodicity in q2 plainly appears from AHSM values in contrary to the fluctuating harmonics of q3. If the speech utterance has a periodicity, it can be evaluated based on the AHSM.