SER: Speech Emotion Recognition Application Based on Extreme Learning Machine

Abstract

Nowadays, device control is commonly using the human body feature or voice recognition technology. To expand the functionality of voice recognition, plenty of researchers have developed speech emotion recognition. By recognizing sound emotions, a system can provide better and beneficial decision-making output. This paper describes the development of an application that is able to recognize speech emotions using Extreme Learning Machine (ELM). We use the dataset from Toronto Emotional Speech Set (TESS). The dataset contains 2800 data points (audio files) in total and has high quality audio that focused on female voices to ensure the reliability of the data. The Speech Emotion Recognition application was design as web-based application that used Golang and Python which built with Extreme Learning Machine and Random Forest to recognize speech emotions. As a result, the functionality test shows that the application was able to satisfy 6 out of 6 requirements, and the accuracy test shows an accuracy value of 100% by identifying 70 out of 70 test data.