| 一种新的鲁棒声纹特征提取与融合方法 |
| |
| 引用本文： | 罗元,孙龙.一种新的鲁棒声纹特征提取与融合方法[J].计算机科学,2016,43(8):297-299, 317. |
| |
| 作者姓名： | 罗元 孙龙 |
| |
| 作者单位： | 重庆邮电大学光电工程学院 重庆400065,重庆邮电大学光电工程学院 重庆400065 |
| |
| 基金项目： | 本文受重庆市教委科学技术研究项目基金(KJ130512)资助 |
| |
| 摘 要： | 为提高说话人确认系统在噪声环境下的鲁棒性,在利用听觉外周模型改进Mel频率倒谱系数(Mel FrequencyCepstral Coefficient,MFCC)的基础上,结合感知线性预测系数(Perceptual Linear Predictive Coefficient,PLPC),以类间区分度为依据,在特征域对两种声纹特征进行融合,提出一种新的声纹特征提取方法,并对基于该特征的说话人确认系统的噪声鲁棒性进行研究。针对不同信噪比的语音信号进行了融合特征与原始特征的对比实验,结果表明,融合特征在模拟餐厅噪声环境中的错误率更低,较MFCC与PLPC分别降低了2.2%和3.1%,说话人确认系统在噪声中的鲁棒性得到提升。
|
| 关 键 词： | Gammatone特征参数 感知线性预测 类间区分度 特征融合 鲁棒性 说话人确认 |
| 收稿时间： | 2015/5/13 0:00:00 |
| 修稿时间： | 2015/7/26 0:00:00 |
New Method of Robust Voiceprint Feature Extraction and Fusion |
| |
| LUO Yuan and SUN Long.New Method of Robust Voiceprint Feature Extraction and Fusion[J].Computer Science,2016,43(8):297-299, 317. |
| |
| Authors: | LUO Yuan and SUN Long |
| |
| Affiliation: | College of Optoelectronic and Engineering,Chongqing University of Posts and Telecommunication,Chongqing 400065,China and College of Optoelectronic and Engineering,Chongqing University of Posts and Telecommunication,Chongqing 400065,China |
| |
| Abstract: | In order to promote the robustness of speaker verification system in noise circumstance,this paper improved MFCC based on auditory periphery model,finished the fusion of improved MFCC and PLPC according to the inter-cluster distinctness,obtained a new voiceprint feature and tested its robustness in noise circumstance.An experiment based on different SNR between fused feature and original feature was finished.The experimental results show that the fused feature can effectively increase the voiceprint recognition compared to MFCC and PLPC by 2.2% and 3.1% in simulated restaurant noise. |
| |
| Keywords: | GFCC PLP Inter-cluster distinctness Feature fusion Robustness Speaker verification |
|
| 点击此处可从《计算机科学》浏览原始摘要信息 |
|
点击此处可从《计算机科学》下载全文 |
|
