Robust Emotion Recognition using Combination of Excitation Source, Spectral and Prosodic Features - Robust Emotion Recognition Using Spectral and Prosodic Features - page 75

Table 4.3 Emotion classification performance using the combination of prosodic and system
features
Emotions Emotion recognition performance (%)

System features Prosodic features Syst. + Pros. features

(0 . 7 + 0 . 3)

Anger 57 67 70

Disgust 67 73 70

Fear 63 70 77

Happiness 73 77 83

Neutral 83 60 80

Sadness 73 60 83

Sarcasm 77 57 77

Surprise 63 53 70

Average 69.5 64.63 76.25

90

System
Prosody
System+Prosody

80

70

60

50

40

30

20

10

0

Anger Disgust Fear Happy Neutral Sad Sarcastic Surprise

Emotons

Fig. 4.5 Comparison of emotion recognition performance with respect to each emotion using
spectral, prosodic, and spectral + prosodic features derived on IITKGP-SESC

In general, spectral features outperformed prosodic features while classifying the
emotions. However, in the case of active emotions like anger, disgust, fear, and hap-
piness, spectral features performed poorly compared to their prosodic counterparts.
It indicates that the prosodic features dominate the spectral features during classifica-
tion of the above mentioned emotions. For slow emotions, especially like sadness and
sarcasm, spectral features performed quite well. In the case of neutral and sarcasm
considerable improvement in the recognition performance is not observed even after
the combination of features. These observations are shown in the form of a bar graph
in Fig. 4.5 . After combining spectral and prosodic features, there is a considerable
improvement of more than 6% compared to the performance of spectral features.
This indicates that prosodic features provide considerable supplementary emotion
specific measure to the spectral features.