Speaker-dependent low-level acoustic feature extraction for emotion recognition

Abstract

In this paper, accuracy for emotion recognition using low- level acoustic features is investigated. The aim of any speech emotion recognition system is to extract acoustic features that are representative of the emotional state of the speaker. Frequency formants, intensity, and pitch are the low-level features proposed for characterizing four different emotions, anger, happy, sadness, and neutral, using acoustic data. Low- level features describe the acoustic, prosodic, and spectral properties of the speech signal and limit the complexity of emotion recognition systems. An algorithm is designed for characterizing each emotion using the acoustic features. It has been proven that various aspects of a speaker’s physical and emotional state can be identified by speech alone. However, the accuracy of such analyses has not been optimized due to acoustic variabilities such as length and complexity of human speech utterance, gender, speaking styles and speech rate. It has also been found that speaker-dependent systems are more accurate in emotion recognition than that of speaker-independent systems. Since speech emotion recognition is relatively a newer field, the set of most powerful features which can distinguish different emotions is not defined; hence, examining the accuracy of emotion recognition using selected acoustical features is an important task.