Variants of deep artificial neural networks for speech recognition systems

This paper presents a survey of basic methods for acoustic and language model development based on artificial neural networks for automatic speech recognition systems. The hybrid and tandem approaches for combination of Hidden Markov Models and artificial neural networks for acoustic modelling are given. The creation of language models using feedforward and recurrent neural networks is described. The survey of researches, conducted in this field, shows that application of artificial neural networks at the stages of both acoustic and language modeling allows decreasing word error rate.