Speech Emotion Recognition in Persian using Spectro-Temporal Features

These days, speech emotion recognition has considered in the cases where there is a relationship between man and machine. Despite many efforts in this field, there is still a great gap between the natural feelings of humans and the computer's perception of it. The main reason for this is the inability of the computer to understand the user's feelings. The purpose of this paper is to design a system to recognize Persian emotional speech database, which includes five emotions of happiness, exhausting, fear, anger and sadness. In this paper, after extraction of four-dimensional features of scale, rate (speed), time and speech frequency with the help of the human auditory model system, two-dimensional features of the scale and frequency was obtained that the maximum amount of these features was used as a feature vector. Finally, the extracted features were classified using support vector machine. The results of the experiments show that the proposed algorithm provides acceptable performance compared to automatic speech emotion recognition in Persian.