Robust Automatic Speech Recognition

Chapter 2

Fundamentals of speech recognition

Abstract

In this chapter, we introduced the fundamental concepts and component technologies for automatic speech recognition. The topics reviewed in this chapter include several important types of acoustic models—Gaussian mixture models (GMM), hidden Markov models (HMM), and deep neural networks (DNN), plus several of their major variants. The role of language modeling is also briefly discussed in the context of the fundamental formulation of the speech recognition problem.

The HMM with GMMs as its statistical distributions given a state is a shallow generative model for speech feature sequences. Hidden dynamic models generalize the HMM by incorporating some deep structure of speech generation as ...