基于稀疏表示分类的说话人识别算法及其在智能考勤系统中的应用 |
| |
引用本文: | 邢玉娟,谭萍.基于稀疏表示分类的说话人识别算法及其在智能考勤系统中的应用[J].工业仪表与自动化装置,2016(2):84-87. |
| |
作者姓名: | 邢玉娟 谭萍 |
| |
作者单位: | 兰州文理学院 计算机系,兰州,730000 |
| |
基金项目: | 甘肃省教育厅科研基金项目(2014A-125) |
| |
摘 要: | 智能考勤系统在企业对员工的出勤考查中具有十分重要的作用。为了提高语音考勤系统的准确性,提出基于稀疏表示的说话人识别算法。该算法在通用背景模型的基础上提取说话人语音的高斯混合模型超向量,采用线性判别分析技术对超向量进行信道补偿和降维,再由低维超向量形成稀疏表示的过完备字典。根据话者测试语音在过完备字典上的重构误差,对话者的身份进行验证。实验结果表明,基于稀疏表示分类的语音考勤系统具有良好的性能。
|
关 键 词: | 语音考勤 说话人识别 稀疏表示 高斯混合模型超向量 线性判别分析 |
The speaker recognition algorithm based on sparse representation and its application in the intelligent attendance system |
| |
Abstract: | Intelligent attendance system plays important role in the business-to-employee attend-ance check. In order to improve the recognition accuracy of speech attendance system, a novel speaker recognition algorithm based on sparse representation was proposed in this paper. This algorithm extracted GMM-super vectors of speaker’ s speech according to the GMM-UBM of registered speakers, and a-dopted linear discriminant analysis as method of channel compensation and dimensionality reduction. By doing so, the over-completed dictionary constituted of these higher discriminant and lower dimensional vectors. The paper reconstruct testing vectors on over-completed dictionary, and selected the category of minimal reconstruction error as target speaker. The experimental results show that the method has better performance. |
| |
Keywords: | speech attendance speaker recognition sparse representation GMM-super vector linear discriminant analysis |
本文献已被 CNKI 万方数据 等数据库收录! |
|