Performance study for multimodel client identification system using cardiac and speech signals

A person's physiological or behavioral characteristic can be used as a biometric and provides automatic identification. There are several advantages of this identification method over the traditional approaches. Overall, biometric techniques can potentially prevent unauthorized access. Unlike...

Full description

Saved in:
Bibliographic Details
Main Authors: Hussain, H., Salleh, S.H., Ting, C.M., Norman, F., Mohammad, M.M., Latif, A.Z.A., Al-Hamdani, O.
Format: Conference or Workshop Item
Language:English
Published: 2018
Subjects:
Online Access:http://eprints.unisza.edu.my/1212/1/FH03-FP-19-24431.pdf
http://eprints.unisza.edu.my/1212/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A person's physiological or behavioral characteristic can be used as a biometric and provides automatic identification. There are several advantages of this identification method over the traditional approaches. Overall, biometric techniques can potentially prevent unauthorized access. Unlike the traditional approaches which uses keys, ID, and password, these approaches can be lost, stolen, forged and even forgotten. Biometric systems or pattern recognitions system have been acknowledged by many as a solution to overcome the security problems in this current times. This work looks into the performance of these signals at a frequency samples of 16 kHz. The work was conducted for Client Identification (CID) for 20 clients. The building block for these biometric system is based on MFCC-HMM. The purpose is to evaluate the system based on the performance of training data sets of 30%, 50% and 70%. This work is evaluated using biometric signals of Electrocardiogram (ECG), heart sound (HS) and speech (SP) in order to find the best performance based on the complexity of states and Gaussian. The best CID performance was obtained by SP at 95% for 50% training data at 16 kHz. The worst CID performance was obtained by ECG achieving only 53.21 % for 30% data training.