An Examination of Some Error Correcting Techniques for Continuous Speech Recognition Technology.

Abstract

The primary purpose of this research was to examine the effects of various input string lengths and error correction methods on the recognition accuracy and efficiency of a currently available continuous automatic speech recognition (ASR) system. The effect of sex was examined also and an estimate of the average recognition accuracy of a continuous ASR system was sought. In the entry of numerical data, the input string length of seven digits at a time proved significantly more efficient than strings of three or five. Although subjects preferred some error correction methods over others, there were no significant differences in error rates or efficiency due to the correction method used. There were also no significant differences due to sex. The average recognition accuracy of the continuous ASR system was conservatively estimated at over 95%. Keywords: VTAG, Voice recognition, Speech recognition, Error correction.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 01, 1985
Accession Number
ADA158001

Entities

People

  • B. J. Martin
  • G. K. Poock

Organizations

  • Naval Postgraduate School

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Accuracy
  • Analysis Of Variance
  • Automated Speech Recognition
  • Automatic
  • Background Noise
  • Computing-Related Activities
  • Data Science
  • Efficiency
  • Errors
  • Experimental Design
  • Identification
  • Information Science
  • Noise
  • Operations Research
  • Recognition
  • Training
  • Vocabulary

Readers

  • Computational Modeling and Simulation
  • Computer Programming and Software Development.
  • Speech Processing/Speech Recognition.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference