This web site is no longer maintained and the content may be outdated.
Please visit www.cmpe.boun.edu.tr for up-to-date information.
 
CmpE RSS
No upcoming events...

Home / Graduate / M.S. Theses Completed
 
 
 
 
  Ahmet Şentürk, 2004    

Thesis Title

Independent component analysis as a front-end for speaker recognition


Abstract

Current speaker recognition (SR) systems show high performance when high-quality speech samples, usually recorded in laboratory conditions, are used as input. However in real world experiments, interfering sounds in the environment (noise, music, other speakers etc.) degrades the performance and presents a serious challenge. To improve the performance of SR systems, it is important to improve the quality of speech input before being introduced into a SR system by separating those portions of speech that belong to testing speaker from the original sound mixture. The work in this thesis reports the results of experiments on using independent component analysis (ICA) as a front-end processing for SR to decompose sound mixtures into source signals and interfacing source speech signal with automatic speaker authentication system to overcome the cocktail party problem in real-world environments. In experiments, YOHO database [15] is used. The performances with interfering speaker, noise, and music are obtained to show the effectiveness of the proposed method. Also several ICA algorithms are used to compare the performances of different ICA methods. As a result of experiments, it has been successfully shown that ICA algorithms, like the EGLD-ICA and Fast-ICA, can be used as a front-end for speaker recognition systems to separate sound mixtures into source signals therefore increase the performance of the recognition task.
 
 
Boğaziçi University Department of Computer Engineering
Address: 34342 Bebek, Istanbul, TURKEY
Phone: +90 212 359 4523-24 Fax: +90 212 287 2461
general information: infocmpe.boun.edu.tr   webmaster: webmastercmpe.boun.edu.tr