| 本文已被：浏览 3830次 下载 5380次 |
码上扫一扫！ |
|
|
| 与文本无关的说话人自适应确认方法 |
|
张怡颖1,2, 朱小燕1,2, 张钹1,2
|
|
1.清华大学计算机科学与技术系,北京,100084;2.清华大学智能技术与系统国家重点实验室,北京,100084
|
|
| 摘要: |
| 该文提出一种与文本无关的自适应说话人确认方法.此自适应方法基于作者所提出的用全局说话人模型标准化似然得分值进行说话人确认的方法,以解决此方法应用于实际系统时存在的训练时间较长的问题,从而缩短新用户注册系统的等待时间,使新用户能够在较短的时间内开始系统的使用.实验结果充分说明了此方法的有效性;当系统有30个用户时,新用户的注册速度加快了12倍. |
| 关键词: 说话人确认,似然得分标准化,自适应方法,最大化似然概率,高斯混合模型. |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家自然科学基金（No.69823001）资助。 |
|
| An Adaptive Method for Text-Independent Speaker Verification |
|
ZHANG Yi-ying,ZHU Xiao-yan,ZHANG Bo
|
| Abstract: |
| In this paper, a novel adaptive text-independent speaker verification method is proposed. This adaptive method is based on the previous work, which uses global speaker model to normalize the likelihood score, and solves one problem of the previous method, i.e., the training time is to long. As a consequence, the waiting time for a new registration is shortened so that a new user can use the system in a short period. The experimental results fully demonstrate the effectiveness of this novel method. When the system has 30 users, the registration time for a new user is accelerated 12 times. |
| Key words: Speaker verification, likelihood score normalization, adaptive method, maximial likelihood probability, Gaussian mixture model. |
