摘要
为了进一步提高针对汉语语音的唇形特征识别效果,分析实际汉语语音发音过程中声母韵母之间音素的变换规律,以及连读等发音习惯而造成的口形变化,利用唇形特征所对应的音素帧间的相关性,采用二阶隐马尔可夫模型来对唇形特征参数序列进行学习和识别,从而分析汉语唇形识别效果.基于独立汉字发音的实验表明,在针对特定人的识别条件下,在最优的加权因子(m∶n=1.5∶1)特征组合条件下,针对同一组融合得到的特征向量,考虑了音素帧间的相关性后,识别率提高了1.2%.可见汉语音节中音素帧间的相关性与唇形特征的变化规律相对应,有利于提高唇形识别的效果.
In order to improve the recognition rate of lipreading for Chinese phoneme. The context information of Chinese phoneme is considered. Second-order Hidden Markov Model is implemented to train and test the lip' s feature sequences to capture the changing discipline between consonant and vowel in Chinese phoneme. The accuracy of recognition rates are tested with the same lip feature vectors. The experimental results based on isolated Chinese words show that the context information of Chinese phoneme can produce better recognition result when applied to lipreading. A maximum recognition rate was improved by 1.2% under the best weighted coefficients (m : n = 1.5 : 1). It can see that the changing discipline of lip feature vectors fits for the context information of Chinese phoneme, which can produce better recognition result of lipreading.
出处
《河北工业大学学报》
CAS
北大核心
2010年第3期37-41,共5页
Journal of Hebei University of Technology
基金
国家自然科学基金(60674111)
天津大学985工程资助项目