Romanian language robust continuous speech recognition [articol]

Abstract

In this paper we present some principles for continuous speech recognition, insisting on the influence of several environmental factors that could affect recognition performances. A Romanian language recognizer has been trained using both context dependent and context independent models. Multistyle training strategy was used to train the recognizer with various levels of artificial noise added on the clean speech. Experimental results prove that this scheme strongly increase the system robustness to additive noise.