计算机应用 ›› 2014, Vol. 34 ›› Issue (1): 265-269.DOI: 10.11772/j.issn.1001-9081.2014.01.0265

• 人工智能 • 上一篇    下一篇

鉴别性最大后验概率声学模型自适应

齐耀辉1,2,3,潘复平3,葛凤培3,颜永红2,3   

  1. 1.
    2. 北京理工大学 信息与电子学院,北京 100081;
    3. 中国科学院声学研究所 中国科学院语言声学与内容理解重点实验室,北京 100190;
  • 收稿日期:2013-07-16 修回日期:2013-09-08 出版日期:2014-01-01 发布日期:2014-02-14
  • 通讯作者: 齐耀辉
  • 作者简介:齐耀辉(1978-),女,河北石家庄人,讲师,博士研究生,主要研究方向:大词表连续语音识别;潘复平(1977-),男,安徽阜阳人,副研究员,博士,主要研究方向:大词表连续语音识别、发音质量自动评估;葛凤培(1982-),女,河北保定人,助理研究员,博士,主要研究方向:大词表连续语音识别、发音质量自动评估;颜永红(1967-),男,江苏无锡人,研究员,博士生导师,博士,主要研究方向:大词表连续语音识别。
  • 基金资助:

    国家自然科学基金资助项目

Discriminative maximum a posteriori for acoustic model adaptation

QI Yaohui1,2,3,PAN Fuping3,GE Fengpei3,YAN Yonghong2,3   

  1. 1.
    2. College of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China;
    3. Key Laboratory of Speech Acoustics and Content Understanding, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China;
  • Received:2013-07-16 Revised:2013-09-08 Online:2014-01-01 Published:2014-02-14
  • Contact: QI Yaohui
  • Supported by:

    National Natural Science Foundation

摘要: 为了更加准确地估计最小音素错误最大后验概率(MPE-MAP)自适应算法中的先验分布中心,使自适应后的声学模型参数更为准确,从而提高系统的识别性能,分别采用最大互信息最大后验概率(MMI-MAP)自适应和基于最大互信息准则与最大似然准则相结合的H-criterion最大后验概率(H-MAP)自适应估计先验分布中心,提出了基于最大互信息最大后验概率先验的最小音素错误最大后验概率(MPE-MMI-MAP)和基于H-criterion最大后验概率先验的最小音素错误最大后验概率(MPE-H-MAP)算法。任务自适应实验结果表明,MPE-MMI-MAP和MPE-H-MAP算法的自适应性能均优于MPE-MAP、MMI-MAP和最大后验概率(MAP)自适应方法,分别比MPE-MAP相对提高3.4%和2.7%。

关键词: 最大后验概率, 鉴别性最大后验概率, 最大互信息, 最小音素错误, 声学模型自适应

Abstract: For Minimum Phone Error based Maximum A Posteriori (MPE-MAP) adaptation, in order to accurately estimate the center of prior distribution and to improve the recognition performance, the Maximum Mutual Information based MAP (MMI-MAP) adaptation and H-criterion, which was the interpolation of MMI and Maximum Likelihood (ML) criterion, based on MAP (H-MAP) adaptation were used for the estimation of the center of prior distribution, which led to MMI-MAP prior based MPE-MAP (MPE-MMI-MAP) and H-MAP prior based MPE-MAP (MPE-H-MAP). The experimental results of task adaptation show that the two proposed methods both can obtain better recognition performance than MPE-MAP, MMI-MAP and MAP adaptation. MPE-MMI-MAP and MPE-H-MAP can obtain 3.4% and 2.7% relative improvement over MPE-MAP respectively.

Key words: Maximum A Posteriori (MAP), Discriminative MAP (DMAP), Maximum Mutual Information (MMI), Minimum Phone Error (MPE), acoustic model adaptation

中图分类号: