| 基于置信度的无监督说话人自适应语音识别 |
| |
| 引用本文： | 丰洪才,卢正鼎. 基于置信度的无监督说话人自适应语音识别[J]. 计算机工程与科学, 2005, 27(9): 93-96 |
| |
| 作者姓名： | 丰洪才 卢正鼎 |
| |
| 作者单位： | 华中科技大学计算机科学与技术学院,湖北,武汉,430074;华中科技大学计算机科学与技术学院,湖北,武汉,430074 |
| |
| 基金项目： | 2003年国家星火计划资助项目(2003EA760004) |
| |
| 摘 要： | 本文提出一种基于词格信息的置信度计算方法，估计自适应语音识别结果的可靠性，将不可靠的语音从自适应训练集中去掉，从而减小无监督自适应与有监督自适应间的性能差异，提高无监督自适应的性能。
|
| 关 键 词： | 语音识别 说话人自适应 置信度 |
| 文章编号： | 1007-130X(2005)09-0093-04 |
| 修稿时间： | 2003-12-18 |
Speech Recognition of Confidence-Measurement-Based Unsupervised Speaker Adaptation |
| |
| FENG Hong-cai,LU Zheng-ding. Speech Recognition of Confidence-Measurement-Based Unsupervised Speaker Adaptation[J]. Computer Engineering & Science, 2005, 27(9): 93-96 |
| |
| Authors: | FENG Hong-cai LU Zheng-ding |
| |
| Abstract: | Speaker adaptation is an effective means of improving the performance of a speech recognition system. A word lattice based confidence measurement is proposed in this paper to evaluate the reliability of the recognition result and to remove the unreliable speech from the adaptation speech corpus. Then the difference between supervised and unsupervised adaptations is reduced,and the recognition rate of unsupervised recognition is increased. |
| |
| Keywords: | speech recognition speaker adaptation confidence measurement |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
| 点击此处可从《计算机工程与科学》浏览原始摘要信息 |
|
点击此处可从《计算机工程与科学》下载全文 |
