AFRL/HECP Speaker Recognition Systems for the 2004 NIST Speaker Recognition Evaluation

Abstract

The National Institute of Standards and Technology sponsors a yearly evaluation of speaker recognition systems. This report summarizes the systems submitted by AFRL/HECP for the 2004 Speaker Recognition Evaluation. The evaluation consisted of seven training conditions by four testing conditions for a total of 28 task conditions. AFRL/HECP submitted systems for all 28 task conditions, one of only three groups to do so out of the 24 participating groups from twelve countries. In addition, AFRL/HECP submitted unsupervised adaptation systems for four of the conditions. A total of ten different systems were submitted across the conditions (not counting unsupervised adaptation modes). These systems were various combinations of the scores from eight component systems. Component systems unique to AFRL/HECP's submission included a system based on glottal model parameters; a system based on formant center frequencies, formant bandwidths, and fundamental frequency; and a system based on mel-frequency cepstral coefficients (MFCCs) and phoneme-specific Gaussian mixture models (GMMs).

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 31, 2004
Accession Number
ADA430750

Entities

People

  • Eric G. Hansen
  • Raymond E. Slyh
  • Timothy R. Anderson

Organizations

  • Air Force Research Laboratory

Tags

Communities of Interest

  • Energy and Power Technologies
  • Human Systems

DTIC Thesaurus Topics

  • Air Force
  • Air Force Research Laboratories
  • Bandwidth
  • Coefficients
  • Contracts
  • Detection
  • Detectors
  • False Alarms
  • Frequency
  • Hidden Markov Models
  • Identification
  • Language
  • Markov Models
  • Neural Networks
  • Probability
  • Recognition
  • Standards

Fields of Study

  • Computer science

Readers

  • Speech Processing/Speech Recognition.
  • Technical Research and Report Writing.

Technology Areas

  • AI & ML