期刊名称：International Journal of Computer Technology and Applications
电子版ISSN：2229-6093
出版年度：2011
卷号：2
期号：3
页码：463-470
出版社：Technopark Publications
摘要：The Modular approach and Neural Network approach are well known concepts in the research and engineering community. By combining these two together, the Modular Neural Network approach is very effective in searching for solutions to complex problems of various fields. The aim of this study is the distribution of the complexity for the ambiguous words classification task on a set of modules. Each of these modules is a single Neural Network which is characterized by its high degree of specialization. The number of interfaces, and there with possibilities for filtering external acoustic – phonetic knowledge, increases a modular architecture. Modular Neural Network (MNN) for speech recognition is presented with speaker dependent single word recognition in this paper. Using this approach by taking computational effort into account, the system performance can be accessed. The active performance is found maximum for MFCC while training with Modular Neural Network classifiers as 99.88%. The active performance is found maximum for LPCC while training with Modular Neural Network classifier as 99.77%. It is found that MFCC performance is superior to LPCC performance while training the speech data with Modular Neural Network classifier
关键词：Phonetic knowledge; Modular architecture; Mel-frequency cepstral coefficient; Linear predictive coefficients; Classifier; Training
