Speech Recognition: Statistical Methods

Abstract

Statistical methods for speech processing refer to a general methodology in which knowledge about both a speech signal and the language that it expresses, along with practical uses of that knowledge for specific tasks or services, is developed from actual realizations of speech data through a well-defined mathematical and statistical formalism. For more than 20 years, this basic methodology has produced many advances and new results, particularly for recognizing and understanding speech and natural language by machine. In this article, we focus on two important statistical methods, one based primarily on a hidden Markov model formulation that has gained widespread acceptance as the dominant technique in characterizing the variation in the acoustic signal representing speech, and one related to the use of statistics for characterizing word co-occurrences. This second model acts as a form of grammar or set of syntactical constraints on the language. In contrast to earlier systems that employed knowledge based on linguistic analyses, these data-driven statistical methods have proven to produce consistent and useful results and have become the underpinning technology of modern speech recognition and understanding systems. Such systems are used in a wide range of applications such as automatic telephone call routing and information retrieval.