用于汉语全音节识别的新特征参数——频变参数

New Speech Parameter for Recognition of Chinese Syllable-Parameter of Spectrum Variance

本文探讨了一种特定人的汉语全音节语音识别方案,介绍了一种基于人耳听觉特性的语音参数的提取方法,对以1/3倍频程分布的16个通道滤波器组的对数能量输出用非线性时域归正方法归正到定长,然后求出相邻通道间频谱的变化量,即得到一组新的特征参数——频变参数.这组参数能够较好地反映语音中与感知有关的特性,如高音、音强、音调等.音节被选用来作为识别的基本单位,以400个汉语无调音节作为字表.最后给出了识别结果.

A new speaker-dependent speech recognition strategy based on Chinese syllable is discussed, and a feature abstraction approach is introduced, in which a new speech parameter based on auditory characteristic can be abstracted. This parameter of spectrum variance can reflect the character of Chinese speech concerned with speech perception. The syllable is selected to be used as the basic unit for recognition. The word vocabulary consists of 400 basic Chinese syllable. The result of experiment using this recognition strategy is presented.