基于模糊神经网络的声母识别  

CONSONANT RECOGNITION BASED ON FUZZY NEURAL NETWORK

作　　者：梅勇[1] 王群生[1] 徐秉铮[1] 

机构地区：[1]华南理工大学无线电系,广州510641

出　　处：《电子科学学刊》1998年第2期188-193,共6页

基　　金：国家自然科学基金;省自然科学基金

摘　　要：模板匹配法技术是汉语声母识别中较为成功的算法,但它的缺陷影响了其恢复错误、改善识别性能。神经网络(NN)和模糊系统的结合,保留了双方的优点,充分利用了模糊神经网良好的容错性能、计算性能、分类性能和决策性能。本文重点研究了两种基于模糊神经网的声母识别方案,通过对其结构、识别率和特点的分析,可看出模糊神经网的声母识别性能明显优于模板匹配法,是更适于语音识别的网络。Conventional template matching technique is a successfully used algorithm in Chinese consonant recognition, yet its disadvantage limits its recovery of error,improvement of performance. The hybrid system based on the combination of neural network(NN) and fuzzy system maintains their advantages and makes full use of error tolerence performance, calculation performance, classification performance and decision performance for fuzzy neural network. In this paper, two kinds of cosonant recognition schemes based on fuzzy neural network are studied in detail. From the discussion of their structure,recognition rate and characteristics,it can be seen that recognition performance of fuzzy neural network is superior to template matching scheme and thus is more suitable for speech recognition.

关 键 词：声母识别 模糊系统 神经网络 语音识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TN912.34[自动化与计算机技术—控制科学与工程]