Emotion Classification of Voice Recordings Using Deep Learning

Abstract

In this work, we present methods for voice emotion classification using deep learning techniques. To processing audio signals, our method leverages spectral features of voice recordings, which are known to serve as powerful representations of temporal signals. To tackling the classification task, we consider two approaches to processing spectral features: as temporal signals and as spatial/2D signals. For each processing method, we use different neural network architectures that fit the approach. Classification results are analyzed and insights are presented.

Keywords

• voice sentiment detection
• mood recognition
• speech emotion recognition
• cepstral features