How does voice authentication compare against other biometrics?
Voice authentication uses both physical and behavioural characteristics, unlike other biometrics which use only physical characteristics. Voice authentication is more user friendly in that it can be used over the telephone. This is one of the major advantages for voice, over fingerprinting and other biometric technologies; it can be used remotely, and does not need the user to be physically present for verification to occur.
While all biometric technologies have to deal with 'noise' of some kind, voice authentication has less interference than other biometrics. For example, fingerprint recognition has to deal with sensitivities of the technology to dry skin, dirt and oil, and with damage to fingerprints by cleaning chemicals. The complexities for face recognition are lighting conditions, and the use of glasses, hats, and facial hair which limit the face area for recognition.
The software that VerifyMe uses to authenticate voices has been shown to provide greater than 99.9% security with 97% call completion under real deployment situations.* According to the Dialogues Spotlight Technology Report from The Centre for Communication Interface Research at The University of Edinburgh, in their May 2000 publication 'Large Scale Evaluation of Automatic Speaker Verification Technology', the software provides two orders of magnitude increase in security for typical banking applications (in other words, 100 times more secure than current typical banking applications).
*Performance measurements based on US English and UK English
The elements of a voice that make up a 'voice print' are created by the unique physical make-up of a user's voice tract (their vocal cords, articulators, size and shape of the areas of the mouth and nose). These elements can be recognised and verified even if the users voice 'sounds' different (to the human) due to a cold. This allows speaker authentication to proceed even in the face of colds, extraneous background noise and other types of 'noise' that might be imposed on the users voice. Although the voice can be altered so that another person may no longer be able to recognise it, the voice and its unique voice print can still be recognised and verified by the voice authentication technology.
Even though voice authentication relies on the sound of someone's voice to verify them, the effect of recording a voice and playing it back via a tape or some form of recordable medium can be detected. When these elements are recognised by the user authentication software, the claim to a particular identity is rejected. In addition, it is recommended that a system illicits a random digit string, or has a range of questions that are randomly asked.
No matter how good an impressionist is, they cannot reshape or redesign their vocal tract in order to replicate another person's voice. The elements that make-up a voice print cannot usually be heard by the human ear, nor can they be copied by an impressionist.
Does is matter what type of phone I call from or how much background noise there is?
All speech recognition technologies are trained to recognise speech and noise. The same applies to speaker recognition technologies. In addition, speech models are trained using different channels (different phone types including mobile phones).
