基于LVQ混合网络的非特定语音识别

NONSPECIFIC SPEECH RECOGNITION BASED ON LVQ HYBRID NETWORK

介绍一种新的等距离采样参数归一化方法。针对美尔倒谱系数(MFCC)和一阶、二阶美尔倒谱系数,提出了一种新的学习矢量量化(LVQ1)和改进学习矢量量化(LVQ2)结合的识别算法。仿真结果表明所提出的算法相对于只用LVQ1网络识别,可以有效改善学习效率。

A novel method of parameters normalisation,in which the signal is sampled equidistantly, is proposed in this paper. For the cepstral coefficients of MFCC, MFCC and MFCC, a new speech recognition approach based on combining the learning vector quantisation （LVQ1） with improved learning vector quantisation （LVQ2） is presented. Simulation result shows that the proposed algorithm effectively improves the learning efficiency in contrast to only using LVQ1 network recognition.