Contribution of Pitch Measures over Time to Emotion Classification Accuracy
Pol van Rijn, David Poeppel, Pauline Larrouy-Maestri
Abstract
Speech carries information about the emotional state of a speaker. Empirical evidence shows that acoustic cues in emotional prosody change over time. Here, we compile and implement a list of proposed features to capture the change of pitch over time. Our results show that these measures significantly improve the classification of emotions when applied in addition to a standard feature set for emotional prosody. We show that the improvement over baseline is present when applying the same procedure to different corpora. This study demonstrates that changes of F0 contain useful information for the classification of emotions and pave the way for a better understanding of the role of acoustic features in emotion recognition.