A Fuzzy Based Approach for Emotion Recognition in Speech

摘要: This paper uses LPC analysis to extract emotion features from speech. From this analysis, 18 features namely pitch. Jitter, energy, duration and 14 LPC coefficients are extracted from each voice samples to represent the emotion features of six basic emotions; happiness, sadness, fear. anger, surprise and disgust. These 18 features extracted from different samples give rise to 18 fuzzy sets. However, when we have limited number of samples and the variance range between fuzzy set is large, the choice of a proper fuzzification function is crucial. In this paper, we have devised a fuzzification function, which depends on the variance of the fuzzy set. by introducing structural parameters in the membership function. This structural parameter, s and t in the membership function helped to model the emotion parameter variation for individual emotion and thus improve the recognition rate.