Year: 2019
Type: dissertation
Abstract: The Speaker Recognition (SR) systems are more accurate than ever in verifying and identifying the human voice which is one of the most convenient biometric characteristics of the human identity. Research and development on speaker recognition techniques have been varied widely in the last decade with an aim to lessen relevant challenges effects such as background noise, poor channel conditions, crosstalk, etc. In this paper, we evaluate two speaker verification (SV) systems, and each one uses an entirely different method to verify speakers: 1) ALIZE 3.0 which is an opensource platform for SR that was successfully passed the NIST Speaker Recognition Evaluations (SREs) implementing Gaussian mixture model (GMM)-UBM speaker. The other one is 2) 3D Convolutional Neural Network (3D-CNN) architecture, which uses a novel method for speaker verification based on Neural Network technique. This paper investigates how challenging it is to implement applications handling tasks in the field of speaker verification.
Author: Mohammad S. Alghamdi
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2985996375