SPEECH RECOGNITION BY FEATURE-ABSTRACTION TECHNIQUES.

Abstract

A speech-analysis system using analog-threshold logic (ATL) for feature abstraction has been developed to recognize consonants in utterances of CVC words by a number of talkers. The feature-abstraction networks use a single ATL element for most of the logic functions. The ATL element, originally modeled after the biological neuron, has an output which is linearly proportional to the net sum of excitatory and inhibitory inputs, provided that this net sum is greater than some adjustable threshold. Using networks of ATL elements, both the presence and magnitude of significant features can be abstracted in real time from the speech signals. The recognition equipment is capable of abstracting these features over a 60-db dynamic range from the logarithmitized outputs of 19 low-Q, band-pass filters. The speech-recognition equipment contains more than 500 ATL elements and was designed to operate in real time, to utilize parallel processing in the featureabstraction networks and not to require segmentation. (Author)

Document Details

Document Type
Technical Report
Publication Date
Aug 01, 1964
Accession Number
AD0604526

Entities

People

  • A. L. Nelson
  • H. J. Zadell
  • T. B. Martin

Organizations

  • RTX

Tags

DTIC Thesaurus Topics

  • Automated Speech Recognition
  • Computational Processes
  • Computer Vision
  • Computing-Related Activities
  • Consonants
  • Dynamic Range
  • Identification
  • Image Processing
  • Image Recognition
  • Parallel Computing
  • Parallel Processing
  • Processing Equipment
  • Recognition
  • Speech Analysis

Readers

  • Computer Engineering
  • Neural Network Machine Learning.
  • Speech Processing/Speech Recognition.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks