Modeling Single-Class Data from Multi-Class Data

Modeling Single-Class Data from Multi-Class Data

Method of Modeling Single Data Class from Multi-Class Data

This extensible technology isolates data (text, image, and voice) representing a target class from heterogeneous data representing multiple data categories of the same type. The method may be applied to identify speech from a speaker of interest in an audio stream containing speech from several other speakers, and extends naturally to language and gender identification and more generally to image and text applications. By automatically selecting data representing a particular class from multi-class data, nonessential artifacts may be removed from models trained on multi-class data, thereby enhancing detection and identification capabilities. The complexity of statistical models is also reduced, eliminating the need for costly memory swapping.

applications: 
  • Speech applications such as speaker recognition, gender identification, or language recognition
  • Improved classification accuracy in pattern recognition scenarios
  • Biometric data discrimination
  • Patent Number: 
    7,454,337
    Patent Issue Date: 
    November 18, 2008
    Email: 
    tech_transfer@nsa.gov
    Lab Representatives
    Share to Facebook Share to Twitter Share to Google Plus