New Paradigm in Speech Recognition: Deep Neural Networks

This paper addresses the topic of deep neural networks (DNN). Recently, DNN has become a flagship in the fields of artificial intelligence. Deep learning has surpassed state-of-the-art results in many domains: image recognition, speech recognition, language modelling, parsing, information retrieval, speech synthesis, translation, autonomous cars, gaming, etc. DNN have the ability to discover and learn complex structure of very large data sets. Moreover, DNN have a great capability of generalization. More specifically, speech recognition with DNN is the topic of our work in this paper. We present an overview of different architectures and training procedures for DNN-based models. In the framework of transcription of broadcast news, our DNN-based system decreases the word error rate dramatically compared to a classical system.