24-11-2010, 05:16 PM
[attachment=7273]
An Algorithm for Language Independent Speaker Identification
Introduction
Imagine a world where you do not require your wallet to pay your bill just your voice is needed.Speaker recognition will most likely make it possible in near future .
Speaker Recognition is a behavioral biometric in which an individual is identified/verified on the basis of his voice.
In today's world where security is of prime most concern Speaker recognition provides a solution which is user friendly and non-intrusive in nature. Also it is only biometric which has the ability to provide remote access to restricted sites to users.
Application areas
Speaker recognition finds applications in numerous fields, some of them are listed below
Access Control
Transaction Authentication
Law Enforcement
Speech Data management
Personalization
Sources Of Variation
There are basically two sources of variations among speakers:
Differences in vocal cords and the vocal tract shape
Differences in speaking style
So these lead to two types of features in an individuals voice:
Inherent Features (less susceptible to mimicry, based on the anatomy of vocal tract)
Learned Features (susceptible to mimicry, but can be useful for distinguishing between speakers having similar vocal tract)