Background

Speech recognition is an important area of research today because of current trends in user interfaces and applications that advertise speech as an alternative to other forms of interacting with a computer. The scope of speech recognition extends beyond just providing an alternative to the keyboard to the concept of "universal access" that enables a caller to carry out functions such as making travel reservations, manipulate financial accounts, access information from his/her personal organizer, etc., at any time (flexibility) , and without the need for any other human to be involved in the transaction (privacy)

Another key role of speech recognition lies in data mining and information extraction. Call centers today record hundreds of thousands of hours of speech between customers and customer service representatives. These conversations represent conversational telephone speech. Another major source of conversational telephone speech is voicemail. Developing speech recognition methods to automatically transcribe these conversations would enable the conversations to be indexed and also enable information to be extracted from the transcriptions.

The first steps towards achieving these capabilities is to (i) develop the speech recognition technology that transcribes conversational telephone speech with reasonable accuracy, and (ii) develop algorithms for extracting meaning or information from the text. We will address some of these issues in the following pages.