A Survey of Deep Learning Techniques in Speech Recognition

The last decade has seen a rapid progress of deep learning techniques in various fields since the proposal of an efficient algorithm for training deep belief networks. Amongst these, speech recognition has seen significant breakthroughs with the application and integration of deep learning architectures. Deep Belief Networks have successfully outperformed GMM based acoustic models within the GMM-HMM hybrid that had dominated the field of speech recognition until now. Recent research with Convolutional Neural Networks and Recurrent Neural Networks for acoustic modeling has shown promising results. Both of these architectures are now being used to solve the problem end-to-end. With this article, a survey is provided on the application of three deep learning architectures in the field of speech recognition, namely, Deep Belief Networks, Convolutional Neural Networks and Recurrent Neural Networks.