Development of High-Precision Emotion Estimation Method using Speech Sound Information with Environmental Noise Reduction and Low Sampling Rate

Abstract

Current research on emotion estimation demonstrates its feasibility at a reduced sampling rate of 6 kHz, thus moving away from traditional methods that depend on higher sampling rates; however, low sampling rates have not been adequately investigated. In addition, noise factors have been limited to electronic sounds rather than environmental. Therefore, this study explores the development of a high-precision emotion estimation method using spoken speech data, focusing on scenarios with environmental noise and low sampling rates. To suppress noise, the proposed method extracts feature quantities for emotion classification using band-pass filters and stacked autoencoders. However, the construction of a high-precision emotion estimation model with these feature quantities required further investigation. Thus, emotion estimation was investigated using a one-dimensional convolutional neural network. The results showed an emotion estimation accuracy of 94.7%, indicating successful noise control. Future work will build on this research to develop emotion estimation methods using spoken speech data that can be employed even in noisy environments.

Keywords: Human Sensing, Voice Analysis, Emotional Analysis, Noise Reduction

DOI: 10.54941/ahfe1004677