Learning New Words for Pronunciation Lexicon from Audio Data

doi:10.7652/xjtuxb201606012

您当前的位置：

首页 >

文章列表页 >

Learning New Words for Pronunciation Lexicon from Audio Data

更新时间：2025-07-09

- Learning New Words for Pronunciation Lexicon from Audio Data
- Vol. 50, Issue 6, Pages: 75-82(2016)
- 作者机构：
  
  解放军信息工程大学信息系统工程学院,郑州,450002
- 作者简介：
- 基金信息：
- DOI：10.7652/xjtuxb201606012
  CLC： TN912.3
- Online First：10 June 2016，
  
  Published：2016
- 稿件说明：
移动端阅览
Learning New Words for Pronunciation Lexicon from Audio Data[J]. 2016, 50(6): 75-82.
DOI：

Learning New Words for Pronunciation Lexicon from Audio Data[J]. 2016, 50(6): 75-82. DOI： 10.7652/xjtuxb201606012.

摘要

针对已有的发音字典扩展方法只能从文本数据中学习新词而无法学习到音频数据中新词的问题

提出了一种基于混合语音识别系统的发音字典新词学习方法。该方法首先分别采用音节和字母音素对混合识别系统对音频数据进行集外词识别

利用系统间的互补性得到尽可能多的新词及其发音候选

然后借助感知器与最大熵模型对得到的新词及发音进行优化

降低错误率

最后实现发音字典的扩展

并利用语法语义信息完成对语言模型参数更新。基于华尔街日报(WSJ)语料库的连续语音识别实验表明:该方法可以有效学习到音频数据中的未知新词

采取的数据优化策略极大地提高了所得新词及发音的精度; 在词错误率指标下

字典扩展后系统的识别性能相对基线系统提高约13.4%。

Abstract

A self-learning method of new pronunciation lexicons based on a hybrid speech recognition system is proposed to solve the problem that the existing self-expanding methods of pronunciation lexicons can only learn new words from text data but cannot learn from audio data. The method utilizes both the syllables and the graphones hybrid systems to recognize the out-of-vocabulary words in the audio data and then obtains as many new words with their pronunciations as possible by using the complementary information of the two systems. Then the new word and its pronunciation candidates are optimized using a perceptron model and a maximum entropy model to reduce the error rate. Finally

the lexicon is expanded and the language model parameters are updated by using syntactic and semantic information. Experimental results of continuous speech recognition on Wall Street Journal speech database show that the proposed method learns new words from audio data effectively

and the accuracy is greatly improved by using the data optimization strategies. The extended lexicon system yields a relative gain of 13.4% over the base line system in terms of word error rates.

关键词

Keywords

references

DAVEL M, MARTIROSIAN O. Pronunciation diction-nary development in resource-scarce environments [C]∥Proceedings of International Speech Communication Association. Grenoble, France: ISCA, 2009: 2851-2854.

BISANI M, NEY H. Joint-sequence models for grapheme-to-phoneme conversion [J]. Speech Communication, 2008, 50(5): 434-451.

RAO K, PENG F, SAK H, et al. Grapheme-to-phoneme conversion using long short-term memory recurrent neural networks [C]∥Proceedings of International Conference on Acoustics, Speech, and Signal Processing. Piscataway, NJ, USA: IEEE, 2015: 4225-4229.

TIM S, OCHS S, TANJA S. Web-based tools and methods for rapid pronunciation dictionary creation [J]. Speech Communication, 2014, 56(1): 101-118.

BERT R, KRIS D, MARTENS J. An improved two-stage mixed language model approach for handling out-of-vocabulary words in large vocabulary continuous speech recognition [J]. Computer Speech and Language, 2014, 28(1): 141-162.

郑铁然, 韩纪庆, 李海洋. 基于词片的语言模型及在汉语语音检索中的应用 [J]. 通信学报, 2009, 30(3): 84-88.

ZHENG Tieran, HAN Jiqing, LI Haiyang. Study on performance optimization for Chinese speech retrieval [J]. Journal on Communications, 2009, 30(3): 84-88.

HE Y Z, BRIAN H, PRTER B. Subword-based modeling for handling OOV words in keyword spotting [C]∥Proceedings of International Conference on Acoustics, Speech, and Signal Processing. Piscataway, NJ, USA: IEEE, 2014: 7914-7918.

QIN L, RUDNICKY A I. OOV word detection using hybrid models with mixed types of fragments [C]∥Proceedings of International Speech Communication Association. Grenoble, France: ISCA, 2012: 2450-2453.

BASHA S, AMR M, HAHN S. Improved strategies for a zero OOV rate LVCSR system [C]∥Proceedings of International Conference on Acoustics, Speech, and Signal Processing. Piscataway, NJ, USA: IEEE, 2015: 5048-5052.

BLACK A W, TAYLOR P, CALEY R. The festival speech synthesis system [EB/OL].(2002-12-27)[2016-01-04]. http: ∥www.festvox.org/docs/manual-1.4.3/.

韩冰, 刘一佳, 车万翔. 基于感知器的中文分词增量训练方法研究 [J]. 中文信息学报, 2015, 29(5): 49-54.

HAN Bing, LIU Yijia, CHE Wanxiang. An incremental learning scheme for perceptron based Chinese segmentation [J]. Journal of Chinese Information, 2015, 29(5): 49-54.

李素建, 王厚峰, 俞士汶. 关键词自动标引的最大熵模型应用研究 [J]. 计算机学报, 2004, 27(9): 1192-1197.LI Sujian, WANG Houfeng, YU Shiwen. Research on maximum entropy model for keyword indexing [J]. Chinese Journal of Computers, 2004, 27(9): 1192-1197.

KLEIN D, MANNING C. Feature-rich part-of-speech tagging with a cyclic dependency network [C]∥Proceedings of Human Language Technology and North American Chapter of the Association for Computational Linguistics. Cambridge, MA, USA: ACL, 2003: 252-259.

MILLER G. WordNet: a lexical database for English [J]. Communications of the ACM, 1995, 38(11): 39-41.

Carnegie Mellon University. The CMU pronunciation dictionary [EB/OL].(2007-03-19)[2016-01-04]. http: ∥www.speech.cs.cmu.edu/cgi-bin/cmudict.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

A Continuous Speech Recognition Method Using Dependent Feature Transformation and Combination of Subspace Region

Mechanism and Optimal Control Strategy for Hysteresis in Rapid Dehumidification Response of Indirect Thermal Systems in Electric Vehicles

Numerical Simulation of the Energy Conversion Process in Thermoacoustic Engines with Different Stack Materials

Integrated and Joint Design Optimization of Centrifugal Compressor Impeller and Casing Treatment at Multiple Working Points

Windage Loss and Flow Characteristics in Impeller Back Gaps of sCO2 Radial Inflow Turbines

Related Author

陈斌 1

胡平舸 2

屈丹 1

Related Institution

Institute of Information System Engineering, Information Engineering University

⁰