作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

›› 2003, Vol. 29 ›› Issue (12): 97-99.

• 论文 • 上一篇    下一篇

汉语连续数字串识别的改进

程雪林,吴开政,李宗葛   

  1. 复旦大学计算机科学系,复旦大学计算机科学系,复旦大学计算机科学系 上海 200433
  • 发布日期:2003-06-25

Improvement of Chinese Connected Digits Recognition

CHENG Xuelin,WU Kaizheng,LI Zongge   

  1. (Computer Science Department of Fudan University, Shanghai 200433)
  • Published:2003-06-25

摘要: 介绍了利用三音子模型和基频信息提高汉语连续数字串的识别率。在汉语连续数字串识别中“8”和“2” 是容易混淆的数字,而“9”和“6”在识别时会在末尾插入一个“5”而变成“95”和“65”。三音子模型将不同上下文的同一个数字区分开来,明显提高了识别率。基频反映了声调的变化,将它们作为后处理进一步降低了错误率。

关键词: 汉语连续数字串, 三音子, 基频, NBest

Abstract: In this paper,triphone models and pitch information are incorporated in the recognition of Chinese connected digits. In Chinese connected digits recognition,"8"and"2"are one of the confusable pairs of digits.Moreover,"9" and "6" are easily recognized as "95" and "65". Triphone models can distinguish the same digit under different contexts, so they can markedly improve the recognition rate. The pitches interpret the variance of tones. When they are used as the post processing the error rate is further decreased.

中图分类号: