Hierachical methods for large population speaker identification using telephone speech

Lerato, Lerato

Hierachical methods for large population speaker identification using telephone speech

Master Thesis

2003

Publisher

University of Cape Town

Department

Department of Electrical Engineering

Faculty

Faculty of Engineering and the Built Environment

Abstract

This study focuses on speaker identificat ion. Several problems such as acoustic noise, channel noise, speaker variability, large population of known group of speakers wi thin the system and many others limit good SiD performance. The SiD system extracts speaker specific features from digitised speech signa] for accurate identification. These feature sets are clustered to form the speaker template known as a speaker model. As the number of speakers enrolling into the system gets larger, more models accumulate and the interspeaker confusion results. This study proposes the hierarchical methods which aim to split the large population of enrolled speakers into smaller groups of model databases for minimising interspeaker confusion.

Keywords

Electrical Engineering

Reference:

Collections

Masters

Full item page