面向定点实现的有限词汇量汉语语音识别方法

将声道模型视为全极点模型，采用具有定点运算性质的Ｓｃｈｕｒ算法求得部分相关系数，利用ＬＰＣ对数倒谱分析求得语音识别系统中采用的待识特征量——倒谱．本系统以倒谱作为特征参数，采用加权倒谱距离求得待识量与以Ｋ－均值聚类所得的参数库之间的特征距离，利用最小距离原则进行识别．经验证，本文所采用的方法取得了较好的实验结果

A Limited Vocabulary Chinese Speech Recognition Approaches for the Fixed point Realization

Supposing voice tract model to be a full pole model and getting PARCOR coefficient by using fixed point Schur algorithm, the characteristic parameter cepstrum is obtained through LPC cepstrum analysis. This approach which adopts cepstrum as characteristic parameter uses the minimum weighted cepstral characteristic distance between the recognized parameter and the parameter library. It adopts K mean clustering algorithm to do recognizing. Experiments show that the approach has a good result.