Structured Recurrent Neural Network And Its Applications In Automatic Speech Recognition

Recent years have witnessed massive developments in various deep learning models,they can be roughly divided into three categories:fully connected deep neural networks,recurrent neural networks,and convolutional neural networks.These new models have brought significant performance improvement to applications like machine translation,automatic speech recogni-tion and vision.In this thesis,we will focus on recurrent neural networks and its applications in automatic speech recognition.From a language modeling perspective,we will give a detailed review about the recent developments on recurrent neural network.And then,we will talk about our works about re-structuring recurrent neural network for better automatic speech recognition performance.