摘要
本文提出了一种汉语清辅音的分类方法和数字信号处理算法.这种方法从语流中捕捉塞音的重要特征──冲直条以及清浊音的界限,在此基础上将汉语含清辅音的单音节划分为塞音和非塞音,塞音又进一步划分为非进气塞音和送气塞音或塞擦音.方法的数字信号处理算法在进行有效性模拟实验时,对汉语1267个全音节中的910个清辅音作这种分类体系的预分类的平均正确率达92.4%.表明这种声学预处理方法和数字信号处理算法对汉语语音识别系统原理和机制的改进有实际应用价值.这个数字信号处理方法应用了短时傅立叶变换.
A pre-classification method and its digital signal processing algorithm for Chinese voiceless consonants are proposed. The important features ot stop, spike fill, and voieeless/unvoiceless boundary are detected and marked. Then the Chinese voieeless consonants are divided according to these features into stop and non-stop. The stop is further divided into unaspirated stop, aspirated stop and fricative stop. Test on a data set of 910 Chinese voiceless consonants from an all-syllable databe of 1267 syllable tokens spoken by a male speker shows that the algorithm performs well, with a 92. 4% correct rate. The proposed method and algorithm are of value to improve the mechanism and performance of Chinase spech recognition system.
出处
《电子器件》
CAS
1997年第1期333-337,共5页
Chinese Journal of Electron Devices
关键词
语音识别
特征提取
分类
冲直条
spech recognition, feature extraction, classification, spike fill