基于临界小波参数和新序列核支持向量机的说话人识别
A Sequence Kernel Support Vector Machine Based on the Critical Bandwidth Wavelet Packet Feature for
-
摘要: 在研究可再生希尔伯特空间框架的基础之上,构建出一个新的序列核来对语音序列间的相似性进行度量.特征提取部分针对传统语音短时分析技术容易出现丢失信息的现状,提出了一种基于临界带宽的小波包变换算法.用美国国家标准与技术研究所(NIST)2004年评测数据集进行实验,结果表明该方法可以大幅度提高识别率Abstract: By using the framework of Reproducing Kernel Hilbert Space,a new sequence kernel was developed to measure similarity between sequences of observations.In the feature extraction,a new wavelet packet transform algorithm was presented based on the critical bandwidth.Testing on the National Institute of Standards and Technology(NIST) 2004 evaluation database was performed and the experiment results show that this method can greatly improve the recognition rate
下载: