摘要
为能够准确有效地对含噪声语音信号进行起止位置的端点检测,该文提出了一种基于广义维数距离的端点检测方法。首先利用覆盖法求取广义维数得到该语音信号的三维特征向量,包括容量维数、信息维数、关联维数;然后计算信号的维数特征距离;最后根据特征距离对语音信号类别进行决策分类。实验结果表明,与仅使用单一维数特征检测语音起止端点相比,该文所提出的方法具有较好的鲁棒性,对混杂有不同噪声、不同信噪比的语音信号都能有较好的检测结果,尤其适用于低信噪比的语音端点检测。
Based on the feature distance of generalized dimension, a speech endpoint detection method is proposed in order to detect the noisy-corrupted speech efficiently. Through calculating the generalized dimension by covering the signal with n-dimension boxes, three dimension feature vectors including the box dimension, the information dimension and the correlation dimension are got. Then dimension feature distance could be calculated and used to make a classification for the speech signal. Experimental results show that compared with the detection using one dimension feature only, the proposed method is more robust to the endpoint detection of speech signal containing different noise and SNR, especially for the lower SNR signal.
出处
《电子与信息学报》
EI
CSCD
北大核心
2007年第2期465-468,共4页
Journal of Electronics & Information Technology
基金
国家自然科学基金(60302027)
浙江省教育厅科研基金(20030620)资助课题
关键词
语音端点检测
广义维数
特征距离
Speech endpoint detection
Generalized dimension
Feature distance
