Automatic Speech Recognition Using Deep Learning

Abstract

In this thesis, we will discuss why speech recognition is still a very challenging task. Then we will go through the concepts of machine learning and deep learning necessary for speech recognition systems like Hidden Markov Models(HMMs), Gaussian Mixture Mod- els(GMMs), Deep De-Noising Auto-Encoder and Convolutional Neural Networks(CNNs). We will also discuss how Deep De-noising Auto-Encoder and Convolutional Neural Net- work will increase the reliability and robustness. The main target of this project is to implement the architecture of automatic audio speech recognition system. We will discuss this implementation using Kaldi framework. Further, on the same dataset, we will discuss the implementation of speaker identification system using k-nearest neighbor algorithm.