基于模糊聚类神经网络的语音识别方法

Speech Recognition Based on Fuzzy Clustering Neural Network

提出了一种基于模糊神经网络的语音识别方法.该方法以模糊系统模型为基础,利用改进的模糊聚类辨识算法,构成一种新型的模糊聚类神经网络(FCNN),并将其作为概率密度函数的估计器,对每个状态的输出进行预测.它不仅能有效地在语音识别中引入帧间相关信息,而且能克服状态输出概率密度函数为混合高斯分布的束缚.通过对非特定人汉语孤立词和连续音节的语音识别实验,证实了该方法的有效性.

In this paper, a method of speech recognition based on fuzzy clustering neural network is presented. Based on the fuzzy system model, a novel fuzzy clustering neural network （FCNN） is built with improving fuzzy clustering identification algorithm in this method, and uses it to act as estimator of probability density function, which could forecast output of the each state. The method not only would be effective to describe the inter-frame correlation information, but overcome tie of the state probability density function being mixture Gauss distribution. Through the experiments of speaker-independent mandarin isolated word and continuous syllable speech recognition, the validity of the proposed method could be verified.