Text-D Dependent Speaker Verification System: An Experiment Using Hidden Markov Model (Hmm)
No Thumbnail Available
Date
2004-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
As the popularity of biometrics increase as a means of securing data and locations, the
area of speaker recognition is emerging as an important fie ld of current research. Since
speech is our most natural means of communicate ion having enough identifying
characteristics, speaker recognition has become a desirable method of determining a
person’s identity. This thesis examines the field of speaker verification in which a
speaker' s identity claim is ,'verified from a sample of speech, The experiment mainly
focuses on HMM-based text-dependent speaker verification system experimented by
using a word from the Amharic language. In the course of building the verification
system, the popular Hidden Marcov Model Toolkit (HTK) is used. For each client, six different utterances are taken in two sessions and phone-based
HMM is developed, Speech of each client is collected at different sessions for testing
purpose. Away from the client set a speech (of the word used in the experiment) is
collected for testing purpose, The results obtained are promising in that. only two false acceptances out of the ninety
nine false claims, and one false reject out of the ninety true claims, errors are
c01l1mitted, giving a 0,02 FAR and 0.01 FRR. The details of the research and findings
arc discussed in the paper. Conclusions of the findings as well as future directions are
recommended
Description
Keywords
Information Science