Table 9 Comparison of overall accuracies obtained on LDC corpus
From: On the use of speech parameter contours for emotion recognition
Classification test | Accuracy (%) |
|---|---|
Human listeners (using 11 listeners) [21] | 63.6 |
Automatic - using pitch contours (slope-bias) | 57.1 |
Automatic - using glottal parameter contours (midpoint) | 55.0 |
Automatic - using vocal tract parameter contours (midpoint) | 45.0 |
Automatic - HMM (three-state)-based system using all model parameters | 62.6 |
Automatic - HMM (one-state)-based system using all model parameters | 59.3 |
Automatic - GMM-based system using all parameters (frame-based) | 59.0 |
Automatic - GMM-based system using MFCC + ∆MFCC (frame-based) | 51.1 |