面向情感语音识别的非线性几何特征提取算法

发布时间：2018-03-29 14:39

本文选题：相空间重构　切入点：情感语音识别　出处：《计算机工程与应用》2017年20期

【摘要】：针对现有时域、频域属性特征在区分情感状态上存在的局限性,提出一种基于相空间重构理论的非线性几何特征提取方法。首先,通过分析情感语音信号的最小延迟时间和嵌入维数来实现相空间重构;其次,在重构相空间下分析并提取基于轨迹描述轮廓的五种非线性几何特征;最后,结合韵律特征、MFCC特征和混沌特征,设计实验方案验证所提特征区分情感状态的能力并通过特征选择获得情感信息完整的最优特征集合。选用德语柏林语音库中的五种情感(高兴、悲伤、中性、愤怒、害怕)作为实验数据来源,支持向量机作为识别网络。实验结果表明:与韵律特征、MFCC特征和混沌特征相比,所提特征不仅可以有效地表征语音信号中的情感差异性,也能够弥补现有特征在刻画情感状态上的不足。
[Abstract]:In view of the limitations of the existing temporal and frequency-domain attribute features in distinguishing emotional states, a nonlinear geometric feature extraction method based on phase space reconstruction theory is proposed. The phase space reconstruction is realized by analyzing the minimum delay time and embedding dimension of emotional speech signal. Secondly, five nonlinear geometric features based on trajectory description contour are analyzed and extracted in the reconstructed phase space. Finally, Combining prosodic features with MFCC features and chaotic features, An experimental scheme was designed to verify the ability of the proposed features to distinguish emotional state and to obtain the optimal set of features with complete emotional information through feature selection. Five emotions (happy, sad, neutral, angry) were selected from the German Berlinic language corpus. The experimental results show that compared with the prosodic feature MFCC feature and chaotic feature, the proposed feature can not only effectively represent the emotional difference in speech signal, but also be used as a recognition network. Can also make up for the existing characteristics in the depiction of emotional state deficiencies.
【作者单位】：太原理工大学信息工程学院;
【基金】：国家自然科学基金(No.61371193)
【分类号】：TN912.34

【相似文献】