摘要
借助汉语发音口形的生理特点，在音素识别这一水平上进行汉语语音的辅助识别，具体给出了一种口形形状识别和灰度的统计方法及其具体实现．实验结果基本与理论估算相吻合，对５个元音的口形区别正确率在８０％以上。
This paper presents in detail a statistical method and implementation of mouth shape's shape recognition and gray level based on perfect image features of mouth shapes at the level of vowels recognition. The results of the recognition experiments approximately accord with the reasoning evaluations. The recognition rate is over 80% for five vowels' mouth shapes. It proposes a beneficial associate approach for language's voice wave recognition.
出处
《软件学报》
EI
CSCD
北大核心
1999年第2期205-209,共5页
Journal of Software
基金
浙江省自然科学基金
北京大学视觉与听觉国家实验室基金
关键词
汉语语音识别
口语看话
模式识别
口形形状识别
Chinese speech recognition, mouth shape features abstracting, lip reading, statistical pattern recognition.
