The Gillick Test: A Method for Comparing Two Speech Recognisers Tested on the Same Data

Abstract

The question of the statistical significance of the difference in error rates of two speech recognition algorithms is almost invariably ignored in the literature. If it is considered, it is usually assumed that the algorithms were tested on two independent test sets, whereas in reality, they are normally tested on the same set. The Gillick Test is a simple and elegant technique for deciding whether the difference between the error rates of two algorithms tested on the same data is significant. Keywords: Great Britain.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Feb 22, 1988
Accession Number
ADA199589

Entities

People

  • S. J. Cox

Organizations

  • Royal Signals and Radar Establishment

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Automated Speech Recognition
  • Binomials
  • Data Sets
  • Equations
  • Machine Learning
  • Normal Distribution
  • Probability
  • Random Variables
  • Rejection
  • Standards
  • Test Sets

Readers

  • Speech Processing/Speech Recognition.
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference