用线性预测系数对孤立汉语语声识别的初步研究

用线性预测系数对孤立汉语语声识别的初步研究 CHINESE SPEECH RECOGNITION BY USING LINEAR PREDICTIVE COEFFICIENT METHOD 原勇 first-author 李汉 王延平 从对汉语发音构成的规律分析可知,单个字和两个字的词可以表示大多数汉语词汇。因而用音素作为汉语识别的基本单位比其它的语言可能更加合适。通过对发音的分析和实验得知,汉语发音中较多的情况是从一个音素到另一个音素的过渡。而实际上也正是这些过渡音载荷了语声的大量信息。为了有效地减少计算机的计算量,我们进行了用LPC方法对音素和过渡音识别的初步探讨。对汉语十个数字的高频预测加重语声序列的线性预测分析和识别说明利用音素和过渡音的LPC模板对汉语语声进行识别是有研究价值的。 Known from the analysis of Chinese speech, most of Chinese meaningful phrases can be represented by one or two isolated words. So maybe to choose phonemes as the foundamental elements to recognize Chinese speech is more suitable than the other choices. The transitions fron one phoneme to another are the most cases in Chinese speech articulation. In fact just these transitions do carry the majority of information of the speech. In order to reduce the computer consuming time, the LPC method has been used in the phoneme and transition recognition. The LPC analysis and recognition to the preemphasized sequences of the ten Chinese number articulations（0-9） confirmed the above conclusion. 1983-02-01 2021-04-01 2