Figure 4From: Perceptual audio features for emotion detectionArchitecture of the perceptual feature extraction and the classification system. The reference set is shared by both the training and the test datasets. For each training/test audio utterance, N perceptual vectors are extracted and fed into a classifier. Classifier design has been performed based on the perceptual features extracted from the training data. Similarly, the test features are derived from the test data. S-MV has been used for the category labeling.Back to article page