基于语义细胞的语音情感识别  

Speech emotion recognition based on information cell

作　　者：孙凌云[1] 何博伟 刘征[2] 杨智渊 

机构地区：[1]浙江大学现代工业设计研究所,浙江杭州310027 [2]中国美术学院设计艺术学院,浙江杭州310024

出　　处：《浙江大学学报（工学版）》2015年第6期1001-1008,共8页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61004116);浙江省自然科学基金资助项目(LY13E050005)

摘　　要：为解决语音情感识别分类器空间复杂度高的问题,将语义细胞应用于语音情感识别领域.以语义细胞混合模型为核心,提出基于单层语义细胞的语音情感识别(IC-S)算法以及基于说话人-情感识别的双层语义细胞识别(IC-D)算法.在CASIA(汉语)和SAVEE(英语)情感语料库中进行交叉验证实验,并利用F值评判识别性能.结果表明:相比常用算法(如:SVM),IC-S算法在空间和时间复杂度上具有优势;IC-D算法与SVM算法识别准确率相似,可以有效降低模型存储空间的复杂度,适用于说话人分类较少或较为固定的场景.Information cell was applied in the field of speech emotion recognition to address the problem of high space complexity of speech emotion recognition classifier. Single-layered information cell （IC-S） algorithm and speaker-emotion recognition based dual-layered information cell （IC-D） algorithm were proposed in the light of information cell mixture model. Cross-validation test on CASIA （in Chinese） and SAVEE （in English） corpus were conducted using F-score as the indicator of recognition performance. Results show that the IC-S algorithm has advantages in both time and space complexity compared to common algorithms like SVM. IC-D algorithm achieves similar recognition performance as SVM. IC-D algorithm can reduce the space complexity significantly and it is suitable for scenarios with few or fixed speakers.

关 键 词：语音情感识别 语义细胞 模糊概念 情感计算 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]