A Non-linguistic Speech Emotion Recognition

Abstract

In computer-human interaction, understanding human behavior is one of the main tasks for a machine to make the whole process as natural as possible. Conversation is one of the quickest and most natural methods of communication between humans. By understanding the emotion behind the speech, machines can better assist humans. The purpose of this project is to build a machine learning model that will recognize emotion in the audio containing speech. The modeling of the SER can be done respective and irrespective of the semantic contents of the speech itself. This research will aim to focus on tone and pitch recognition of the voice. Which will help in recognizing the emotional state of the human using the non-linguistic aspect of the speech. As human emotions can get complicated to the point that even humans cannot always precisely understand the emotion of the speaker, we can summarize them into four main categories neutral, positive, angry, and sad emotions. These will be the focus of the predictions for the developed model. Comparison will be done between various implementations to find out the best fitting model for the non-linguistic approach.