Speech Retrieval Defined

A universally accepted definition of speech retrieval is, unfortunately, still lacking. So here is my personal definition, which I have used to build the Speech Retrieval Resources web page. The goal of what I call a speech retrieval system is to sort through large volumes of dynamically generated information and present to the user those which are likely to satisfy his or her information requirement. Thus, speech retrieval is a special case of information retrieval in which the information is in spoken form.

One important characteristic if this definition is that it does not depend on the way in which the information need is specified, but rather only on the way the information that is sought is expressed. As I use the term, speech retrieval systems may require explicit formulations of a query, or they may seek to infer the information need from a set of training data. Furthermore, when an explicit query is required or allowed, that query can be spoken, typed, handwritten, or even gestured and remain within the scope of this definition.

Some have used the term "speech retrieval" to describe systems which use spoken queries to retrieve information in non-spoken forms (for example, speech interfaces to web browsers). Such systems, although interesting in their own right, would be outside the scope of this page.


Doug Oard
Last modified: Fri Dec 26 22:06:10 1997