基于N元文法的领域语法语料扩展算法
详细信息 查看官网全文
- 英文论文题名：Ngram based corpus extension method with domain-specific grammar
- 论文作者：梁宏 ; 黎塔
- 英文论文作者：LIANG Hong ; LI Ta ; Key Laboratory of Speech Acoustics and Content Understanding ; Institute of Acoustics ; Chinese Academy of Sciences
- 年：2016
- 作者机构：中国科学院声学研究所语言声学与内容理解重点实验室;
- 论文关键词：N元文法 ; 语言模型 ; 语音识别 ; 领域语法
- 英文论文关键词：Ngram ; Language Model ; Automatic Speech Recognition ; Domain-Specific Grammar
- 会议召开时间：2016-10-28
- 会议录名称：2016年全国声学学术会议论文集
- 语种：中文
- 分类号：TN912.34
- 学会代码：OGSM
- 会议名称：2016年全国声学学术会议
- 会议地点：中国湖北武汉
- 主办单位：中国声学学会
- 学会名称：中国声学学会
- 页数：4
- 文件大小：276k
- 原文格式：D
- 会议级别：全国
摘要
语言模型训练语料的数据稀疏性问题,一直是语音识别任务所面临的一个严重问题。基于领域语法扩展训练语料的方法,能够有效地缓解特定域语言模型训练数据稀疏性问题。本文在领域语法的基础上,提出了一种基于语言模型N元文法(Ngram)的语料扩展算法。该算法通过统计领域语法的Ngram,对Ngram进行扩展,并训练语言模型。实验表明,将该方法应用于特定域语音识别系统中,不仅能提高语音识别系统的识别性能,还能加快语言模型的训练速度,降低语言模型训练对服务器硬盘空间的需求。
The training corpus sparseness of Language Model(LM) is one of the biggest problems in Automatic Speech Recognition(ASR). Training corpus extension method based on domain-specific grammar can solve data scarcity of relative domain Language Model. In this paper, with domain-specific grammar rules, the corpus extension algorithm based on Ngram of Language Model is proposed. This approach first counts Ngram counts of domain-specific grammar. Then it expands Ngrams with Ngram counts file. Finally, the Language Model is trained with the extension Ngram counts file. The result shows that, with this method, it not only can improve the performance of speech recognition system, but also can speed up the training process, lower the requirement of hard disk.
The training corpus sparseness of Language Model(LM) is one of the biggest problems in Automatic Speech Recognition(ASR). Training corpus extension method based on domain-specific grammar can solve data scarcity of relative domain Language Model. In this paper, with domain-specific grammar rules, the corpus extension algorithm based on Ngram of Language Model is proposed. This approach first counts Ngram counts of domain-specific grammar. Then it expands Ngrams with Ngram counts file. Finally, the Language Model is trained with the extension Ngram counts file. The result shows that, with this method, it not only can improve the performance of speech recognition system, but also can speed up the training process, lower the requirement of hard disk.
引文
[1]S.F.Chen,J.Goodman.An Empirical Study of Smoothing Techniques for Language Modeling.Proceedings of the 34th Meeting of the Association for Computational Linguistics(ACL'96).
[2]Y.Si,M.Chen,Q.Zhang,J.Pan,Y.Yan.Block Based Language Model for Target Domain Adaptation towards Web Corpus.Journal of Computational Information Systems 9:22(2013)9139-9146.
[3]H.Schwenk.Continuous space language models.Computer Speech&Language,21(3):492–518,2007.
[4]Tomas Mikolov,Martin Karafi′at,Lukas Burget,Jan Cernock`y,and Sanjeev Khudanpur.Recurrent neural network based language model.In INTERSPEECH,pages 1045–1048,2010
[5]Martin Sundermeyer,Ralf Schlüter,and Hermann Ney:"LSTM Neural Networks for Language Modeling",Proc.of Interspeech2012
[6]司玉景,肖业鸣,徐及,潘接林,颜永红,面向口语统计语言模型建模的自动语料生成算法.自动化学报,2014,40(12):2808-2814Si Yu-Jing,Xiao Ye-Ming,Xu Ji,Pan Jie-Lin,Yan Yong-Hong.Automatic text corpus generation algorithm towards oral statistical language modeling.Acta Automatica Sinica,2014,40(12):2808—2814.
[7]邢永康,马少平.统计语言模型综述[J].计算机科学,2003,30(9):22-26.Xing Yong-Kang,Ma Shao-Ping.A Survey on Statistical Language Models[J].Computer Science,2003,30(9):22-26.
[8]Zhang X,Trmal J,Povey D,et al.Improving deep neural network acoustic models using generalized maxout networks[C]//Acoustics,Speech and Signal Processing(ICASSP),2014 IEEE International Conference on.IEEE,2014:215-219.
[2]Y.Si,M.Chen,Q.Zhang,J.Pan,Y.Yan.Block Based Language Model for Target Domain Adaptation towards Web Corpus.Journal of Computational Information Systems 9:22(2013)9139-9146.
[3]H.Schwenk.Continuous space language models.Computer Speech&Language,21(3):492–518,2007.
[4]Tomas Mikolov,Martin Karafi′at,Lukas Burget,Jan Cernock`y,and Sanjeev Khudanpur.Recurrent neural network based language model.In INTERSPEECH,pages 1045–1048,2010
[5]Martin Sundermeyer,Ralf Schlüter,and Hermann Ney:"LSTM Neural Networks for Language Modeling",Proc.of Interspeech2012
[6]司玉景,肖业鸣,徐及,潘接林,颜永红,面向口语统计语言模型建模的自动语料生成算法.自动化学报,2014,40(12):2808-2814Si Yu-Jing,Xiao Ye-Ming,Xu Ji,Pan Jie-Lin,Yan Yong-Hong.Automatic text corpus generation algorithm towards oral statistical language modeling.Acta Automatica Sinica,2014,40(12):2808—2814.
[7]邢永康,马少平.统计语言模型综述[J].计算机科学,2003,30(9):22-26.Xing Yong-Kang,Ma Shao-Ping.A Survey on Statistical Language Models[J].Computer Science,2003,30(9):22-26.
[8]Zhang X,Trmal J,Povey D,et al.Improving deep neural network acoustic models using generalized maxout networks[C]//Acoustics,Speech and Signal Processing(ICASSP),2014 IEEE International Conference on.IEEE,2014:215-219.