Machines that understand speech

Publisher Website
Google Scholar
Abstract
The ability to use language to formulate abstract concepts and to communicate these concepts to another human being must rank as one of the most significant aspects of intelligent behaviour that a human being can demonstrate. Speaking is the technique that has evolved to enable this communication to take place quickly and efficiently between two people in close proximity to one another. Consequently machines that understand speech are usually considered to be intelligent. The designers of the early machines for recognizing speech treated it as a set of patterns that could be processed by conventional pattern classification techniques. This work has led to the design of commercially available machines that respond to a limited range of isolated words in controlled conditions. More recent attempts at designing speech understanding machines have been based upon the linguistic nature of speech and employ linguistic and semantic modelling to aid recognition. Whilst the performance of these systems is vastly superior to the pattern classification type of machine, it is still very limited when compared with the speech understanding capabilities of a human being. This paper reviews the problems and current state of the art in machine understanding of speech.

This publication has 2 references indexed in Scilit: