Speech Emotion Feature Analysis Based on Emotion Fingerprints

2018 2nd International Conference on Artificial Intelligence Applications and Technologies

计算机科学

Jiang, Yuantao^1 ; Deng, Kaifa^2 ; Wu, Chunxue^1

School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai, China^1

Shanghai University of Engineering Science, Shanghai, China^2

关键词: emotion fingerpringting;    Emotion recognition;    Lifting wavelet packets;    Sample entropy;    Singular values;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/435/1/012050/pdf
DOI  :  10.1088/1757-899X/435/1/012050

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

The speech recognition has caught the attention of more and more researchers as the important branch of artificial intelligence. However, speech recognition just can recognize the neutral meaning and ignore the important emotion information. So, this paper proposes an emotion fingerprint extraction algorithm. The speech signal is decomposed by lifting wavelet packets. And combining singular value entropy and sample entropy as the feature vector to characterize the speech time-frequency matrix coefficient. Then adopting the statistical values to extract emotion fingerprints. The experiments show that this algorithm can reflect the emotion information of the speech fully and distinguish several major emotions well.