人体语音特征提取身份优化验证仿真研究

发布时间：2018-05-05 08:16

本文选题：人体发声过程 + 音段韵律　；参考：《计算机仿真》2017年02期

【摘要】：对人体语音征提取身份优化验证,可为说话人识别奠定基础。进行人体语音征提取身份验证时,应分析人体语音音段韵律特征矢量序列,提取最优音段韵律的高维特征值和特征向量,但是传统方法通过对标注音节的持续采样点数进行分析完成检测,但是不能精确分析人体语音音段韵律特征矢量序列,无法准确提取最优音段韵律的高维特征值和特征向量,存在人体语音征提取身份验证误差大的问题。提出一种改进混沌的人体语音征提取身份优化验证方法。上述方法先融合于混沌理论采集人体发声过程中音段韵律原始信号,将原始韵律信号映射到高维空间实现音段韵律相空间重构,映射相空间中音段韵律间相邻轨道发散的平均变化率,然后利用K-均值聚类的方法对音段韵律的语音帧进行聚类,获取规范化的音段韵律特征矢量序列,将规范化的音段韵律特征矢量序列投影到音段韵律高维核空间中,提取最优音段韵律的高维特征值和特征向量,依据人体语音征提取身份优化验证,仿真结果证明,所提方法特征提取精确度高,能够有效地提升人体语音征提取身份验证的辨识率。
[Abstract]:The identification optimization of human speech feature extraction can lay a foundation for speaker recognition. In the process of human speech feature extraction, we should analyze the sequence of prosodic vector of human speech segment, and extract the high dimensional characteristic value and feature vector of the optimal segment prosody. However, the traditional methods can not accurately analyze the sequence of prosodic feature vectors of human speech segments, and can not accurately extract the high dimensional characteristic values and feature vectors of the optimal segment prosody by analyzing the number of continuous sampling points of the tagged syllables, but the traditional methods can not accurately analyze the sequence of prosodic feature vectors of human speech segments. There is a problem of large error in the identification of human speech sign extraction. An improved chaotic identification method for human speech feature extraction is proposed. Firstly, the method is integrated into chaos theory to collect the prosody signal of segment in the process of human voice, and the original prosodic signal is mapped to the high-dimensional space to reconstruct the phase space of segment prosody. In the mapping phase space, the average variation rate of the divergence of adjacent tracks between segments prosody is obtained, and then the speech frames of segment prosody are clustered by K-means clustering method, and the normalized segment prosodic characteristic vector sequences are obtained. The normalized segment prosodic feature vector sequence is projected into the segment prosodic high-dimensional kernel space, the high-dimensional characteristic value and eigenvector of the optimal segment prosody are extracted, and the identity optimization verification is extracted according to the human speech sign. The simulation results prove that, The proposed method has high accuracy of feature extraction and can effectively improve the identification rate of human speech feature extraction.
【作者单位】：商丘学院计算机工程学院;兰州理工大学理学院;
【分类号】：TN912.3

【相似文献】