Emotion Recognition from Speech Using IG-Based Feature Compensation

Computational Linguistics and Chinese Language Processing Vol. 12, No. 1, March 2007, pp. 65-78 65 © The Association for Computational Linguistics and Chinese Language Processing [Received September 8, 2006; Revised September 20, 2006; Accepted October 10, 2006] Emotion Recognition from Speech Using IG-Based Feature Compensation Chung-Hsien Wu∗, and Ze-Jing Chuang∗ Abstract This paper presents an approach to feature compensation for emotion recognition from speech signals. In this approach, the intonation groups (IGs) of the input speech signals are extracted first. The speech features in each selected intonation group are then extracted. With the assumption of linear mapping between feature spaces in different emotional states, a feature compensation approach is proposed to characterize feature space with better discriminability among emotional states. The compensation vector with respect to each emotional state is estimated using the Minimum Classification Error (MCE) algorithm. For the final emotional state decision, the compensated IG-based feature vectors are used to train the Gaussian Mixture Models (GMMs) and Continuous Support Vector Machine (CSVMs) for each emotional state. For GMMs, the emotional state with the GMM having the maximal likelihood ratio is determined as the final output. For CSVMs, the emotional state is determined according to the probability outputs from the CSVMs. The kernel function in CSVM is experimentally decided as a Radial basis function. A comparison in the experiments shows that the proposed IG-based feature compensation can obtain encouraging performance for emotion recognition. Keywords: Emotional Speech, Emotion Recognition, Intonation Group, Feature Compensation 1. Introduction Human-machine interface technology has been investigated for several decades. Recent research has put more emphasis on the recognition of nonverbal information, especially on the topic of emotion reaction. Scientists have found that emotional skills can be an important component of intelligence, especially for human-human communication. Although human-computer interaction is different from human-human communication, some theories ∗ Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan, ROC E-mail: {chwu, bala}@csie.ncku.edu.tw 66 Chung-Hsien Wu and Ze-Jing Chuang have shown that human-computer interaction essentially follows the basics of human-human interaction [Reeves et al. 1996; Picard 1997; Cowie et al. 2001]. Scientists have found that emotion technology can be an important component in artificial intelligence, especially for human-human communication [Salovey et al. 1990]. Although the human-computer interaction is different from human-human communication, some theories show that human-computer interaction basically follows the fundamental forms of human-human interaction [Reeves et al. 1996]. In this study, an emotion recognition approach from speech signals is proposed. This method consists of the definition and extraction of intonation groups (IGs), IG-based feature extraction, and feature compensation. In past years, many researchers have paid attention to emotion recognition via speech signals. Several important recognition models have been applied to the emotion recognition task, such as Neural Network (NN) [Bhatti et al. 2004], Hidden Markov Model (HMM) [Inanoglu et al. 2005], Support Vector Machine (SVM) [Kwon et al. 2003; Chuang et al. 2004], and others [Subasic et al. 2001; Wu et al. 2006; Silva et al. 2000]. Besides the generally used prosodic and …

Content truncated to 500 words