计算机工程与应用 ›› 2010, Vol. 46 ›› Issue (25): 146-148.DOI: 10.3778/j.issn.1002-8331.2010.25.043

• 数据库、信号与信息处理 • 上一篇    下一篇

语种辨识的多特征信息应用

曾秀花1,2,杨 鉴2,徐永华2   

  1. 1.曲靖师范学院 物理与电子工程学院,云南 曲靖 655011
    2.云南大学 信息学院,昆明 650091
  • 收稿日期:2009-12-09 修回日期:2010-04-13 出版日期:2010-09-01 发布日期:2010-09-01
  • 通讯作者: 曾秀花

Application of multiple speech features to language identification

ZENG Xiu-hua1,2,YANG Jian2,XU Yong-hua2   

  1. 1.School of Physics & Electronic Engineering,Qujing Normal University,Qujing,Yunnan 655011,China
    2.School of Information Science and Engineering,Yunnan University,Kunming 650091,China
  • Received:2009-12-09 Revised:2010-04-13 Online:2010-09-01 Published:2010-09-01
  • Contact: ZENG Xiu-hua

摘要: 提取动态的高层语言学特征建立了改进的语种相关的、联合的GMM-LM语种辨识方案。该方案减小了不同语种的高斯混合模型和语言模型之间的相关性,也降低了训练的复杂度。还提出了基于特征提取层和判决层融合技术的语种辨识系统。该系统利用了不同类型的特征对区分不同语种的贡献来增加不同语种语料之间的差异,并使相同语种的语料之间的差异减小。实验表明,设计的语种辨识系统具有较好的扩展性;基于特征提取层和判决层的融合系统能够有效地提高系统识别率。

关键词: 语种识别, 高斯混合模型, 特征融合

Abstract: Dynamic high-level language cues are extracted to build the language-dependent and combined GMM-LM language identification system which decreases the correlation between Gaussian Mixture Model and Language Model.It also reduces the training complexity.The feature-level and decision-level multiple features fusion is proposed to the GMM-LM system which distinguishes multiple languages using different feature cues.This system can maximize the speech diversity of different languages and minimize the speech diversity of same languages.It concludes that this modified Gaussian Mixture Model recognizer followed by Language Model has better system expansibi1ity;meanwhile the fusion systems based on feature-level and decision-level can get better performance.

Key words: language recognition, Gaussian mixture model, feature fusion

中图分类号: