Speech Perception.

Abstract

A general approach to speech recognition is presented. The objective is to study and investigate the recognition of words as they appear in spoken phrases composed of context-free speech, using a small set of words. This effort is an extension of the work performed under contract F30602-73-C-0049. A method of segmentation is presented based on continuous processing of the input signal to detect plosives, fricatives and vowels. This is coupled with speaker tranformation and pattern matching algorithms to produce a word recognition system. This limited vocabulary word recognition system was used to evaluate various algorithms of consonant recognition and speaker transformations. The speaker transformations that reduce inter-speaker variations were tested using cooperative as well as non-cooperative speakers. Final evaluation was performed for a total of 10 speakers. Using the English digits spoken in a connected manner, speaker transformation and consonant recognition were evaluated. These programs were then combined to form a digit recognition system with an average correct recognition score of 95%. (Author)

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 1975
Accession Number
ADA018442

Entities

People

  • Esber Koprucu
  • Henry Killett
  • Huseyin Yilmaz
  • Leon Ferber
  • William Park

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Automated Speech Recognition
  • Computer Vision
  • Consonants
  • Continuous Processing
  • Contracts
  • Freedom Of Speech
  • Identification
  • Image Processing
  • Image Recognition
  • Perception
  • Recognition
  • Test And Evaluation
  • Vocabulary
  • Word Recognition

Readers

  • Speech Processing/Speech Recognition.

Technology Areas

  • AI & ML
  • AI & ML - Machine Translation