Computer Engineering and Applications ›› 2008, Vol. 44 ›› Issue (29): 239-240.DOI: 10.3778/j.issn.1002-8331.2008.29.069

• 工程与应用 • Previous Articles     Next Articles

Curve fitting of pitch contour used for tone recognition of isolated mandarin syllables

SONG Gang1,2,YAO Yan-hong2   

  1. 1.Tianjin University,Tianjin 300072,China
    2.School of Information Science and Engineering,Shandong University,Jinan 250100,China
  • Received:2007-11-22 Revised:2008-03-03 Online:2008-10-11 Published:2008-10-11
  • Contact: SONG Gang

用于汉语单音节声调识别的基频轨迹拟合方法

宋 刚1,2,姚艳红2   

  1. 1.天津大学,天津 300072
    2.山东大学 信息科学与工程学院,济南 250100
  • 通讯作者: 宋 刚

Abstract: This paper proposes a subsection curve fitting of pitch contour after summarizing the changing rules of the parameters of four tones based on the pitch target model.The eigenvector extracted by this curve fitting can avoid negative effects of the error endpoints and normalize the length of the pitch contour.An accuracy rate of 95.17% can be achieved in mandarin tone recognition using the eigenvector.

Key words: tone recognition, pitch contour, length normalization, curve fitting

摘要: 基于Target模型的思想,对汉语声调基频轨迹进行研究,总结了四种声调曲线参数的变化规律,提出了对基频轨迹进行分段曲线拟合的方法。使用该方法提取出的反映基频变化趋势的特征向量,在不受基频曲线两端不准确点影响的同时起到了长度归一化的作用。使用该向量进行声调识别,取得了95.17%的满意识别结果。

关键词: 声调识别, 基频, 长度归一化, 曲线拟合