论文部分内容阅读
指出汉语单字语音存在一种可截尾特性，并且得到与之相关的三个结论（称为尾音可切除原则）：（１）如果适当切除单字部分青尾特征，识别率不会明显下降，甚至有所提高．（２）切尾后识别时间明显缩短，分析和实验结果表明：若采用动态时间规整算法（ＤＴＷ），识别时间与特征矢量长度的平方成正比关系．（３）实验指出，音尾特征的截除极限为特征矢量总长度的１／３．根据上述原则，从假设一检验的认知理论出发，提出一种汉语连接词的识别算法，并在ＤＴＷ模型上得以实现．实验测试集包括２００个特定人发音样本，其中２字词１６２个，３字词２２个，４字词１６个，正识率为９１％．该算法对待识词的字数没有限制，井且随待识字数的增加，识别时间只作线性增长．
It is pointed out that there exists a truncated characteristic in Chinese monophonic speech, and three conclusions related to it (called tail exclamability) are obtained: (1) If the green-tail character of one-character part is properly cut off, the recognition rate will not be obviously decreased; has seen an increase. (2) The recognition time is significantly shortened after tailing, and the analysis and experimental results show that the recognition time is proportional to the square of the length of eigenvector if dynamic time warping algorithm (DTW) is used. (3) Experiments show that the cut-off limit of the tail feature is 1/3 of the total length of the feature vector. According to the above principle, this paper proposes a recognition algorithm for Chinese connectives from the cognitive theory of hypotheses-test, which is realized on the DTW model. The experimental test set includes 200 specific human voice samples, of which 162 words are 2 words, 22 words are 3 words and 16 words are 4 words, and the positive recognition rate is 91%. There is no limit to the number of words that the algorithm treats, and the recognition time increases only linearly with the increase of the number of words to be recognized.