| 段长在汉语语音识别系统后处理阶段的应用 |
| |
| 引用本文： | 李曜,刘加. 段长在汉语语音识别系统后处理阶段的应用[J]. 清华大学学报(自然科学版), 2009, 0(Z1) |
| |
| 作者姓名： | 李曜 刘加 |
| |
| 作者单位： | 清华大学电子工程系; |
| |
| 基金项目： | 国家自然科学基金委员会与微软亚洲研究院联合资助项目(60776800);;国家“八六三”高技术项目(2006AA010101,2007AA04Z223) |
| |
| 摘 要： | 针对传统的隐含马尔可夫模型(hidden Markov model,HMM)存在的缺陷,该文提出了一种在识别的后处理阶段使用段长模型的方法,并应用在基于HMM的汉语识别系统上。该方法利用归一化的段长模型对识别系统的解码结果重新打分,比较前后两次算出的分数从而选出更可靠的识别结果。实验表明,通过该方法将段长模型应用在识别过程中,可以显著提高识别系统的性能,大量减少识别结果中的插入错误。数据显示,该方法使识别系统的音节错误率下降了大约10%,识别系统最终的插入错误和删除错误都低于1%。
|
| 关 键 词： | 语音识别 后处理段长模型 段长归一化 |
Duration model for post-processing in a Mandarin speech recognition system |
| |
| LI Yao,LIU Jia. Duration model for post-processing in a Mandarin speech recognition system[J]. Journal of Tsinghua University(Science and Technology), 2009, 0(Z1) |
| |
| Authors: | LI Yao LIU Jia |
| |
| Affiliation: | Department of Electronics;Tsinghua University;Beijing 100084;China |
| |
| Abstract: | The traditional Hidden Markov Model(HMM) has some limitations for Mandarin speech recognition.This paper presants a duration Model for the post-processing stage for Mandarin speech recognition system based on HMM.The best candidate from the decoder is rescored by the normalized duration model with more credible results obtained by comparing the two scores.The syllable error rate is reduced by 10% and the improvement of the insertion error rate is especially significant,which indicates that the method works ... |
| |
| Keywords: | speech recognition post-processor duration model duration normalization |
| 本文献已被 CNKI 等数据库收录！ |
|
