Emotion Recognition plays an important role in robust speech recognition. In this paper, for emotion recognition the Wavelet decomposed coefficients of speech are used. The Wavelet approximate and detail coefficients are improved using Teager Energy Operator. This improved coefficient's entropy is calculated in feature extraction stage and used for emotion classification. The analysis is carried out on Polish Emotional Database. The four emotions namely anger, joy, neutral and sad are considered which creates a four class problem. The Euclidean distance is applied as a feature classifier and giving the nearest emotion that matches to test input speech. The performance is evaluated based on the ability of system to recognize emotion independent of speaker. Teager Energy Operator which reflects the nonlinear vortex flow interaction of speech and entropy as a feature vector truly minimizes the calculations for emotion recognition. The entropy as a feature outperforms the other statistical features extracted from coefficients.
DWT, Entropy, Euclidean distance, Teager Energy Operator(TEO)
