Laughter is an important social signal that conveys different emotions like happiness, sadness, anger, fear, surprise, and disgust. Therefore, detecting emotions in the laughter is useful for estimating the emotional state of the user. This paper presents work that detects the emotions in Iranian laughter by using audio features and running four machine learning algorithms, namely, Sequential Minimal Optimization (SMO), Multilayer Perception (MLP), Logistic, and Radial Basis Function Network (RBFNetwork). We extracted features such as intensity (minimum, maximum, mean, and standard deviation), energy, power, first 3 formants, and the first thirteen Mel Frequency Cepstral Coefficients. Two datasets are used: one that contains segments of full laughter episodes and one that contains only laughter onsets. Results indicate that MLP algorithm produce the highest rate of accuracy which is 86.1372% for first dataset and 85.0123% for second dataset. Besides, using the combination of MFCC and prosodic features led to better results. This means that recognition of emotions is possible at the start of laughter, which is useful for real-time applications.
展开▼
