基于稀疏特征迁移的语音情感识别  

Speech Emotion Recognition Using Sparse Feature Transfer

作　　者：宋鹏[1] 金赟[2] 查诚[3] 赵力[2,3] 

机构地区：[1]烟台大学计算机与控制工程学院,烟台264005 [2]东南大学儿童发展与学习科学教育部重点实验室,南京210096 [3]东南大学信息科学与工程学院,南京210096

出　　处：《数据采集与处理》2016年第2期325-330,共6页Journal of Data Acquisition and Processing

基　　金：山东省自然科学基金(ZR2014FQ016;ZR2015PF010)资助项目;国家自然科学基金(61273266;61403328;61403329)资助项目;东南大学基本科研业务费(CDLS-2015-04)资助项目

摘　　要：为了解决语音情感识别系统中训练数据和测试数据来自不同数据库所引起的识别率降低的问题,提出了一种基于稀疏特征迁移的语音情感识别方法。通过引入稀疏编码获取情感特征在不同数据库条件下的共同稀疏表示;同时引入最大区分差异(Maximum mean discrepancy,MMD)来衡量不同数据库条件下稀疏表示分布之间的距离,并将其作为稀疏编码目标函数的约束条件,从而获得较为鲁棒的稀疏特征。实验结果表明,相比传统语音情感识别方法,基于稀疏特征迁移的语音情感识别方法显著提高了跨库条件下的情感识别率。In speech emotion recognition system ,recognition rates will drop drastically when the training and the testing utterances are from different corpora .To solve this problem ,a novel sparse feature trans‐fer approach is proposed .By employing sparse coding algorithm ,the common sparse feature representa‐tion of emotion features from different corpora is obtained .Meanwhile ,the maximum mean discrepancy （MMD） algorithm is introduced to measure the distance between different distributions ,and is used as the regularization term for the objective function of sparse coding .Finally ,the robust sparse features are achieved for recognition .Experimental results show that ,compared to traditional methods ,the proposed approach can significantly improve the recognition rates for cross databases .

关 键 词：语音情感识别 特征迁移 稀疏编码 

分 类 号：TN912.3[电子电信—通信与信息系统]