Browse Prior Art Database

Phonetic Speech Recognition using Spectrum Pairs

IP.com Disclosure Number: IPCOM000115361D
Original Publication Date: 1995-Apr-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 2 page(s) / 60K

Publishing Venue

IBM

Related People

Destombes, F: AUTHOR

Abstract

The disclosed device allows to perform phonetic recognition of running speech. It consists of: 1. A microphone linked to an acoustic component. 2. An acoustic component to determine the frequency spectrum of successive sound "frames" (short duration segments). For example, the technique used to compute such a spectrum might be Linear Predictive Coding. 3. A memory device holding model spectrum pairs obtained in a training phase. 4. A comparison mechanism to find which model spectrum pair is closest to an incoming unknown spectrum pair, thus yielding a phonetic label for that unknown pair. 5. A decision mechanism to group successive identical labels to yield a phoneme label. 6.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 53% of the total text.

Phonetic Speech Recognition using Spectrum Pairs

The disclosed device allows to perform phonetic recognition of
running speech.  It consists of:
  1.  A microphone linked to an acoustic component.
  2.  An acoustic component to determine the frequency spectrum of
       successive sound "frames" (short duration segments).   For
       example, the technique used to compute such  a spectrum might
be
       Linear Predictive Coding.
  3.  A memory device holding model spectrum pairs obtained in a
       training phase.
  4.  A comparison mechanism to find which model spectrum pair is
       closest to an incoming unknown spectrum pair, thus yielding a
       phonetic label for that unknown pair.
  5.  A decision mechanism to group successive identical labels to
       yield a phoneme label.
  6.  An interface to other devices used to store, display or use
such
       phoneme labels to perform other speech recognition tasks.
  7.  A control mechanism to coordinate the device operation.

Under control of the control mechanism, the following operations are
performed:
  o  Speech sounds are captured by the microphone.  The acoustic
      component computes spectra for successive frames (short
duration
      segments).  It then provides spectrum pairs to the next stage.
A
      spectrum pair is composed of the spectrum for the present
frame,
      plus the spectrum for a previous frame, at a fixed,
predetermined
      time delay from the present frame.
  o  The comparison mechanism searches the model spectrum pairs,
      c...