Efficient Feature Extraction for Emotion Recognition System

Emotion can be detected by human easily but computer cannot detect that easily. Therefore, audio feature extraction plays an important role in analyzing and characterizing audio content. Moreover, the extracted feature should be robust to various distortions and have a good reliability property for a good classification system. This paper focuses on feature selection method, which aims to extract effective acoustics, features to improve the performance of emotion recognition. To obtain a useful feature extraction in this system, the combination of MFCC (Mels Frequency Cepstral Coefficients) with Berouti Spectral Subtraction is used. Then, the system is built with the Random Forest classifier for the efficient classification. The modelling technique is speaker-independent and text-independent. This system is carried out with Matlab 2016a and the overall system accuracy is about 68.3% corresponding to the created dataset.