This primer describes the concepts and technology around voice biometrics. It covers biometrics in general; the key elements of the technology of voice biometrics; the key strengths of using voice as a biometric; and finally how voice verification can be used in practical security applications.

Biometrics

The voice, as with other biometric characteristics, is unique to an individual.

Voice biometrics is the technology behind voice (or speaker) verification, which in turn is an application of the core technology.

In voice biometrics, a user (a caller) speaks an utterance that is captured by the biometric system that then compares it to a previously stored voiceprint. This comparison process produces a score of how well the new utterance matches the stored voiceprint.

Voice biometric systems can be classified as either being text-dependent or text-independent. In text-dependent systems, the phrases used during enrolment and verification are specific and predetermined. In a text-independent system, any phrase can be used and the voiceprint is not tied to the specific use of any phrase. Text-independent systems require longer phrases to create voiceprints of equivalent accuracy.

Despite considerable research into text-independent systems, most secure access commercial applications use text-dependent biometric systems because of their higher performance.

The voiceprint (the template) is an algorithmically derived model of both the physical and behavioural characteristics of the users voice. The voiceprint is a unique digital representation of an individual's voice. It is not a recording; or a set of words; or a wave pattern of a voice. It cannot be played back or used for any other purpose than a comparison with subsequent voiceprints using the same algorithm. The algorithm uses over 100 different calculations to generate the voiceprint.

An adaptation process can be used track changes to a user's vocal tract / voice that occur over time so that a customer does not need to constantly update and maintain their voiceprint 'identity'.

Additionally, as the voiceprint is characterised by the vocal tract (a physiological trait) and not specifically the 'sound' of the utterance, afflictions that affect the voice, such as a cold, do not also affect the vocal tract. For this reason, having a cold generally has no adverse affect on accuracy levels. If the speaker can provide enough vocal energy to be heard, enough of their voice will be present for a verification to be possible

Voice Biometrics

The advantages of voice biometrics over other forms of biometrics used in user verification are:

• Easy to Use
• Uses existing phone systems and technology.  No specialist hardware or software is needed.  The environment (a phone) for using the technology is already familiar to users.
• Works with any phone / any telephony system / from anywhere on the globe - this promotes mobility.
• Your voice is always with you.  There is nothing private to carry or remember (such as a PIN, password etc).  Also, your voice cannot be lost or stolen.  It can be recorded but voice biometric technology employs sophisticated algorithms specifically aimed at detecting recordings.
• No training is needed to use a voice biometric system - speaking is a natural activity.
• Easy to Integrate
• Voice biometric-based systems are well suited to securing remote applications.  Users can be located anywhere that they have access to a phone.
• By leveraging existing telephony and security infrastructure, voice biometric systems are cost effective.
• When used together with speech recognition, voice biometric systems can also be used to gather, by the user speaking, additional information such as a PIN numbers; user-acknowledgement of prompted questions; etc. 
• Voice biometrics are culturally non-invasive and are therefore generally seen as non-contentious.
• Secure
• The human voice is harder to use fraudulently than some other biometric techniques.
• Voice biometric systems can be used with existing authentication methods to provide high-security multi-factor user-authentication.