The automatic speech recognition problem consists of transcribing voice to text. It is one of the most important areas in machine learning, due to the vast quantity of applications which may have in every day life.Recently, hybrid acoustic models have been proposed for automatic speech recognition tasks: systems based on combining hidden Markov models and neural networks. In this work, hybrid models of such characteristics are studied and also other systems which combine automata inferred by grammatical inference methodologies and neural networks are proposed. Finally, a novel approach to language modeling is proposed by using neural networks. |