| 语音分段在基于GMM-SVM说话人确认中的应用 |
| |
| 引用本文： | 饶为,王典洪,麦文伟. 语音分段在基于GMM-SVM说话人确认中的应用[J]. 电子技术, 2010, 47(3): 18-19 |
| |
| 作者姓名： | 饶为 王典洪 麦文伟 |
| |
| 作者单位： | 中国地质大学(武汉)机械与电子信息学院;香港理工大学电子及资讯工程学系 |
| |
| 摘 要： | 在说话人确认系统的实际应用中,让用户提供大量的训练语音是不现实的,所以在GMM-SVM系统中,正样本点数通常只有一个,而负样本点数远远多于正样本点数,造成SVM分类超平面严重偏向负样本,这种情况对于支持向量机的性能影响很大。针对此问题,提出了基于时间间隔对语音数据进行分段的方法,来增多正样本点数,得到更好的分类超平面。美国国家标准与技术研究所(NIST)2002年说话人识别数据库上的实验证明,语音分段的方法能在一定程度上提升整个说话者确认系统的识别精度和鲁棒性。
|
| 关 键 词： | 语音分段 GMM超向量 支持向量机 通用背景模型 说话人确认 |
Application of Utterance Partition Based on Time Interval in GMM-SVM Speaker Verification |
| |
| Rao Wei,Wang Dianhong,Man-Wai Mak. Application of Utterance Partition Based on Time Interval in GMM-SVM Speaker Verification[J]. Electronic Technology, 2010, 47(3): 18-19 |
| |
| Authors: | Rao Wei Wang Dianhong Man-Wai Mak |
| |
| Affiliation: | 1. Faculty of Mechanical and Electronic Information;China University of Geosciences 2. Department of Electronic and Information Engineering;Hong Kong Polytechnic University |
| |
| Abstract: | In a practical speaker verification system,it is impractical for users to provide large amount of training data for enrollment. Hence there is only one sample point for positive class,while the number of sample point for negative class is much more than that of the sample point for positive class,which causes the SVM hyperplane to tend to the negative class seriously thereby significantly affecting the performance of SVM. In regard to this problem,the method of utterance partition based on time interval is ... |
| |
| Keywords: | utterance partition based on time interval GMM supervector support vector machine(SVM) universal background model(UBM) speaker verification |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
