Influence Linear Discriminant Analysis on isolated digit recognition

The paper deals with Automatic Speech Recognition system (ASR) with focus on isolated digits recognition in Slovak language. The paper discusses reduction of dimension of feature space. There are applied two ways dimension reductions. The first way feature subspace reduction is bases on manually selection some coefficients from feature matrix. The second way feature subspace reduction is automatic and use Linear Discriminant Analysis. In the first part of the article, there is briefly defined ASR system and introduce 3TDCM method (tree Two-Dimensional Cepstral Matrix). The next part of the article focuses on exploitation of LDA end use in ASR system. In the third part of the paper, there are shown experiments with isolated digit recognition. Moreover, there are compare systems ASR with automatic and manual feature subspace selection and influence feature subspace reduction on performances of ASR system.