Large vocabulary continuous speech recognition of Uyghur: Basic research of acoustic model

Abstract

In this paper, we analysis the characteristics of Uyghur languge, introduce components of speech recognition system, design a program to calculate trigram, bigram and unigram language models based on the corpus of the entire text of the novel Qum basqan sheher, then estimating acoustic models uses unigram and bigram language models calculated above, particularly with some probability theory for acoustic modeling. And discuss some potential problems of the probabilistic speech recognition approaches. © 2012 Springer-Verlag GmbH.