| 本文已被：浏览 3862次 下载 5444次 |
码上扫一扫！ |
|
|
| 一种新的说话人确认方法 |
|
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: |
| 文章在对说话人确认和说话人辨认进行比较研究的基础上，提出一种新的说话人确认方法．同传统方法相比，该方法通过建立非特定说话人模型综合多个说话人的语音特性，使其能够对于不同的待确认语音给出不同的判决阈值，从而解决了说话人确认在判决阈值设置上存在的困难．实验结果表明，该方法能够显著降低说话人确认系统的错误接受率和错误拒绝率，为说话人确认应用于保密性要求较高的环境提供了一条有效的途径． |
| 关键词: 说话人确认 说话人辨认 判决阈值 错误拒绝率 错误接受率 |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家自然科学基金资助. |
|
| A Novel Speaker Verification Method |
|
ZHANG Yiying,ZHU Xiaoyan,ZHANG Bo
|
| Abstract: |
| In this paper, on the basis of comparison of speaker verification and speaker identification, a novel speaker verification method is proposed. Compared to the conventional method, the speakerindependent speaker model is established to represent the universal feature of speech, and is used during the verification phase so that the decision threshold is adapted to different input speech and decided during verification phase. The difficulties of setting the decision threshold for the conventional method are al... |
| Key words: Speaker verification speaker identification decision threshold false rejection rate false acceptance rate. |
