Audio-based emotion recognition using GMM supervector an SVM linear kernel

In this paper, we present an audio-based emotion recognition model by using OpenSmile, Gaussian mixture models (GMMs) Supervector and Support vector machines (SVM) with Linear kernel. Features are extracted from audio characteristics of emotional video through OpenSmile into Mel-frequency Cepstral Coefficient (MFCC) of 39 dimensions for each video. Furthermore, these features are normalized to the same size using GMM Supervector with 32 mixture components. Finally, data is classified using SVM with Linear Kernel. To evaluate the model, this paper using the AFEW2017 dataset and SAVEE dataset and show comparable the results on the state-of-the-art network. The experimental results perform with 37% on AFEW and 73.5% on SAVEE dataset. Our proposed achieves improved emotion recognition from audio as compared to several other models.