Classification of Negative Emotion Speech Intensity Based on Similarity Algorithm

The emotional speech recognition is mainly aimed at the division of the basic discrete emotions, and the researches concerning the strength boundaries of each emotion speech are relatively rare. Given this situation and based on the emotional intensity theory proposed by Plutchik, the paper tries to adopt the similarity algorithm to classify the negative emotion. By adoption of the neutral emotions and three negative emotions (anger, fear and sadness) from EMO-DB and CASIA emotional speech database and by using the distance similarity algorithm and K-means clustering algorithm, the emotional speech samples are calculated. On this basis, the grades of the three negative emotions can be divided. At the same time, the divided results above verified by the support vector machine in the experiment. The experimental results show that the basic negative emotions divided by the similarity algorithm obtain a better recognition results. Compared with the distance similarity algorithm, the K-means clustering similarity algorithm is smoother for all kinds of emotion recognition results, and the result is better. In addition, the classification of emotion intensity based on clustering similarity algorithm has broad application prospects in diagnosing the grades of disease of mental disorders such as depression.