Paralinguistic model for emotions recognition with deep neural networks
Eugene Yu. Shchetinin, Leonid Sevastianov, Dmitry Kulyabov, Edik Ayrjan, Anastasia Demidova
In this paper the computer paralinguistic model for emotions recognition based on deep neural networks is proposed. The main stages of its construction were studied and relevant models of the deep convolutional networks and recurrent networks with LSTM memory cell were used. Intensive computer experiments on the emotions recognition from human speech with proposed model were conducted. As the data for research and testing of our model RAVDESS dataset of audio recordings was selected. The results showed a high efficiency of the explored model, and the accuracy estimates for some classes of emotions were reached 90%.