| 英语发音自动评测中单词删除错误的检测方法 |
| |
| 引用本文： | 吴延年,梁维谦.英语发音自动评测中单词删除错误的检测方法[J].电声技术,2009,33(11):64-67. |
| |
| 作者姓名： | 吴延年 梁维谦 |
| |
| 作者单位： | [1]清华大学微电子所,北京100084 [2]清华大学电子工程系,北京100084 |
| |
| 基金项目： | 国家"863"高技术研究发展计划重点项目 |
| |
| 摘 要： | 基于HMM后验概率分数的自动发音评测方法中，强制匹配获得语音的音素级切分信息是重要的一步。切分结果是否准确直接影响到后验概率分数的可靠性。通过对单词发音网络增加可跨越分支，实现单删除错误检测功能，可降低说话人删除错误对前后单词切分准确性的影响。实验结果显示，单词删除错误检出率达到87．8％。
|
| 关 键 词： | 发音评测 发音错误 切分 |
A Word-deletion Error Detection Method for English Automatic Pronunciation Assessment |
| |
| WU Yan-nian,LIANG Wei-qian.A Word-deletion Error Detection Method for English Automatic Pronunciation Assessment[J].Audio Engineering,2009,33(11):64-67. |
| |
| Authors: | WU Yan-nian LIANG Wei-qian |
| |
| Affiliation: | WU Yan-nian1,LIANG Wei-qian2 (1.Institute of Microelectronics,Tsinghua University,Beijing 100084,China,2.Dept.of Electronic Engineering,China) |
| |
| Abstract: | To obtain phonetic segmentation information by force alignment is an important step in HMM-based automatic pronunciation assessment.The accuracy of segmentation has a direct impact on the result of posterior probability score.Add a parallel path span word recognition network can reduce the segmentation errors of adjacent words caused by worddeletion pronunciation errors.The experimental results indicate that the accuracy of worddeletion errors detection reaches 87.8%. |
| |
| Keywords: | pronunciation assessment pronunciation error segmentation |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
