Browse Prior Art Database

Multitasking System for Recognizing Speech Commands

IP.com Disclosure Number: IPCOM000117228D
Original Publication Date: 1996-Jan-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 74K

Publishing Venue

IBM

Related People

Baker, RG: AUTHOR

Abstract

Disclosed is a multitasking system allowing several users in separate but proximate physical locations to use a single speech recognition subsystem. While this system is targeted primarily at the recognition of commands, it may also be used for general types of continuous speech recognition within the constructs of the architecture.

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

Multitasking System for Recognizing Speech Commands

      Disclosed is a multitasking system allowing several users in
separate but proximate physical locations to use a single speech
recognition subsystem.  While this system is targeted primarily at
the recognition of commands, it may also be used for general types of
continuous speech recognition within the constructs of the
architecture.

      The Figure is a simplified block diagram of the system.  The
audio preprocessor 1 accepts audio signals from one or more sources,
such as microphones 2, filters these signals to discriminate between
speech and non-speech inputs (such as extraneous noise), tags each
speech input according to its source, and feeds the signals to the
speech recognition processor 3.  A host processor 4 runs the
application program, and a feedback unit 5 is stimulated to issue
messages to the user as needed.

      Signals may simultaneously come into the audio preprocessor 1
from various microphones 2, as several users attempt to access the
system.  To handle this audio collision, the preprocessor first
filters each signal, determining what is noise and what is speech.
Threshold detection may be used for this purpose.  The filtering
process is not applied to modify the speech signal; for proper
recognition the signal must remain as true to the original speech as
possible.  Therefore, filtering merely determines when a speech
signal is present.  A control signal is generated to coincide with
the detection of a speech signal from each microphone.  These control
signals are passed to the application running in the host processor
4, so that each signal coming from the recognition processor 3 is
tagged with the appropriate s...