Voice and signature for the identification of persons
Biometric systems are automatic systems that measure the physical characteristics and behaviour of persons. The aim of such systems is to differentiate between the characteristics and behaviour of each person and, thus, identify a person immediately.
Identification of physiological traits is based on the measurement of certain parts of the body; amongst other that are used as working tools are fingerprints, facial factions, the iris, the geometry of the hand, DNA or the retina. However, with identification though behaviour, certain activities of the person are used as parameters, such as, for example, the voice, handwriting, the signature, walking gait, the manner of using a keyboard or of moving a PC mouse.
Whatever the bio-measurement, the system requires a suitable sensor in order to read such biometric data. Just as important is having a tool to contrast measured data and previously stored data, i.e. having a suitable database. Given that, in order to repeat the tests reliably over and over again and to compare the results of the algorithms, it is essential to have a well-equipped database. It is also important that the sensors are capable of detecting attempts of forgery.
With this requirement for a biometric database, some years ago the Department of Electronics and Telecommunications at the School of Engineering in Bilbao started joint work with a number of universities in Spain in order to design a database that would provide the biometric characteristics of hundreds of people. Currently, in order to complete the database, University of the Basque Country (UPV-EHU) researchers are focusing on analysing the voice, signatures and handwriting.
As regards voice detection, traditional techniques use segmental characteristics in order to differentiate between persons. One of these is the timbre of the voice. Although good results are obtained, there is always room for improvement. To this end the UPV-EHU researchers set out to enhance this system and, moreover, to measure the rhythm and intonation of the voice. All these parameters are introduced into the database. In fact, it has been shown that, when certain biometric systems are crossed or the parameters of the same bio-measurement summed, in general the average error is less than the error produced when each system is measured independently.
Not only do the research team want to incorporate the voice into the database, but also the signature of the person. Depending on the manner of data collection, the automatic recognition of the signature can be carried out in two ways: on-line and off-line.
Off-line signature recognition is based on a document which is scanned for subsequent processing. All the characteristics of a signature depend on spatial parameters and so it is easier to forge, the forger only having to imitate the way of writing the signature.
However, what is put forward by the research team for on-line signature recognition is not only the study of the spatial form of writing, but also the dynamic data – the manner of movement while writing. Using a digitalisation table and a digital pencil, data is collected in an ongoing way as the pencil writes, the pressure or force exercised when writing with the pencil, the inclination of the pencil, and so on. When registering the on-line signature recognition data, the signer has to be present in order to store in the database the movements and manner of signing and writing.
But, how is a database drawn up? In such a database own signatures as well as imitated ones of other users have to be entered for the system to function.
A biometric system should not admit erroneous data but they often do. With on-line signature recognition there is a 4 % margin of error, more or less. That is, the system rejects 4% of correct signatures and accepts the same amount of false signatures. With off-line identification, the error is significantly greater - 20 %.
As regards voice identification, quite a small error margin was found. In any case, these are provisional results. The research is ongoing, currently looking into the combination of certain voice parameters in order to obtain more precise results.
Last reviewed: By John M. Grohol, Psy.D. on 21 Feb 2009
Published on PsychCentral.com. All rights reserved.