Browse Prior Art Database

Splitting and Routing Audio Signals in Systems with Speech Recognition

IP.com Disclosure Number: IPCOM000117119D
Original Publication Date: 1995-Dec-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 90K

Publishing Venue

IBM

Related People

Choi, W: AUTHOR [+4]

Abstract

Disclosed are methods for splitting and routing audio signals within speech recognition subsystems for the benefit of a system operator who performs repetitive tasks over the phone, such as a stockbroker, a catalogue order taker, or an emergency (911) dispatcher. Preferably, a button or switch mounted on a telephone handset or a personal computer, or a footswitch, is used to reroute outbound acoustic information from the telephone microphone temporarily to a speech-recognition subsystem, while the inbound acoustic information is still routed to the telephone ear piece. In this way, with the rerouting operation being preferably signaled by a switch which does not cause a click or blip, the caller does not know when the person answering the phone is talking to the speech recognition subsystem.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Splitting and Routing Audio Signals in Systems with Speech Recognition

      Disclosed are methods for splitting and routing audio signals
within speech recognition subsystems for the benefit of a system
operator who performs repetitive tasks over the phone, such as a
stockbroker, a catalogue order taker, or an emergency (911)
dispatcher.  Preferably, a button or switch mounted on a telephone
handset or a personal computer, or a footswitch, is used to reroute
outbound acoustic information from the telephone microphone
temporarily to a speech-recognition subsystem, while the inbound
acoustic information is still routed to the telephone ear piece.  In
this way, with the rerouting operation being preferably signaled by a
switch which does not cause a click or blip, the caller does not know
when the person answering the phone is talking to the speech
recognition subsystem.

      Alternatively, the audio signal is split, to be routed to both
the speech recognition subsystem and the caller, so that the caller
continues to hear the conversation when the switch is depressed,
routing the audio signal to the speech recognition subsystem.  The
caller may still not be aware that the speech recognition subsystem
is being used, since it reacts to ordinary words, but he can
acknowledge, and correct if necessary, the information being given by
the system operator.  With this alternative, the system operator can
also let the caller know that he is talking to the computer, for
example by addressing the "computer."

      In a second alternative implementation, the speech recognition
subsystem listens continuously to the system operator, listening
particularly for one or more key words or phrases, such as "code" or
"alert."  These key words or phrases may have to be at the beginning
of a sentence, prefaced or followed by a pause, or they may be
imbedded in continuous speech.  In this application, the reliability
of the speech recognition subsystem is achieved because there is no
need to recognize the speech of people other than the system operator
and because there is no need to recognize speech received over the
switched telephone network.  This "word spotting" technique can be
used along with a button routing the microphone output to the speech
recogni...