Biometrics is, in the simplest definition, something you are.
It is a physical characteristic unique to each individual and in this case, your voice.

According to Dan Miller, Senior Analyst from Opus Research
Voice Biometrics, otherwise known as Speaker Verification, is an application that works by obtaining voice prints, comparing live utterances to these voice prints to produce a pass or fail response during telephone based transactions.

A multi-factor authentication, as defined by the VoiceXML Forum, is a combination at least two means of authentication which can be split into four groups:

• Something users know (a password or account number)
• Something users have (a token, an ATM card, a device or a secret password)
• Something users are (a biometry feature)
• Somewhere users are (location factor, a GPS position)

The speech recognition engine verifies users’ knowledge and possession while voice biometrics tests who users are. When combined it results in a two-factor authentication that meets the demand for a secure and reliable identification and verification procedure.

Applications that require a specific pass phrase are Text Dependent

• Requires training
• Enrolment involves multiple repetition of a unique passphrase or account number
• Can confuse callers

Text Independent applications can use any utterance

• Multiple prompts
• Simplify enrolment
• Support conversational authentication

Voice biometrics statistically measures the unique traits of human voice, while minimising behavioural aspects of the captured voice, for example an impact of a handset or an accent.
Biometrics always works in two phases: enrolment and matching (also referred to as scoring).

• Enrolment: the first phase, is the process of collecting biometric samples from a user (which leads to creation of a reference biometric model). Unique features are extracted and stored in a non-reversible format. Enrolment might be described as teaching, where a biometric engine learns how to identify a user and how to recognise the user’s characteristics.
• Matching: The second phase is where audio samples are tested against a previously created reference model. The comparison process may be either ‘one-to-one’ (when the user is verified) or ‘one-to-many’ (collected utterances are tested against many voiceprints).

A behavioral component, like a cold, does not affect the vocal tract, so there will be no adverse affect on accuracy levels. Only extreme vocal conditions such as laryngitis will prevent the user from using the system.