ARMA倒谱用于语音识别的特征参数

本文介绍了一种以ARMA倒谱系数为语音识别特证参数的快速递推算法,并给出了实验结果,证明它优于LPC倒谱,是语音识别中一种新的有效方法。

Speech recognition by ARMA cepstrum characteristics

Not all ARMA models are reversible. But rt finds that the ir-reversibilty would be untenable while only the output can be measured and excitating source can not for speech production. And proved that the jdentific model of speech must be minium-phased by reason of convergence and stable vocal-tract p-ropagation. From the view of that, the recurrence relations between the model cepstrum cofficients (?)ould be listed. The equalities for findind ARMA coffi-cients obtained with the relations between the ARMA model and AR model. And ARMA cepstrum fast recurrence formula based on minium-phased obtained The effectiveness is proved by speech recognition experiments. It is found that it is a new effective characteristics for speech recognition.