Feature extraction and normalization in SVM-based speaker recognition

Thembisile Mazibuko, Daniel Mashao

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Feature extraction techniques have the capabilit to improve the performance of pattern recognition systems by providing a more relevant representation of the data. Principal Component Analysis and Independent Component Analysis are two linear feature extraction techniques which seek to find new representations of the multivariate data. We compare the performance of ICA and PCA on a speaker verification task based on a subset of the NIST 2000 speaker recognition evaluation task. The classification is performed using a Support Vector Machine classifier. The SVM is a discriminative classifier which has been successfully applied to several pattern recognition tasks. We find that the ICA representation outperforms the PCA representation on the speaker verification task.

Original languageEnglish
Title of host publicationWMSCI 2006 - The 10th World Multi-Conference on Systemics, Cybernetics and Informatics, Jointly with the 12th International Conference on Information Systems Analysis and Synthesis, ISAS 2006 - Proc.
Pages260-264
Number of pages5
Publication statusPublished - 2006
Externally publishedYes
Event10th World Multi-Conference on Systemics, Cybernetics and Informatics, WMSCI 2006, Jointly with the 12th International Conference on Information Systems Analysis and Synthesis, ISAS 2006 - Orlando, FL, United States
Duration: 16 Jul 200619 Jul 2006

Publication series

NameWMSCI 2006 - The 10th World Multi-Conference on Systemics, Cybernetics and Informatics, Jointly with the 12th International Conference on Information Systems Analysis and Synthesis, ISAS 2006 - Proc.
Volume5

Conference

Conference10th World Multi-Conference on Systemics, Cybernetics and Informatics, WMSCI 2006, Jointly with the 12th International Conference on Information Systems Analysis and Synthesis, ISAS 2006
Country/TerritoryUnited States
CityOrlando, FL
Period16/07/0619/07/06

Keywords

  • Feature extraction
  • Speaker recognition
  • Support vector machines

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Information Systems

Fingerprint

Dive into the research topics of 'Feature extraction and normalization in SVM-based speaker recognition'. Together they form a unique fingerprint.

Cite this