Print profiles

Print profiles are a key feature of biometric processing. After you create one, you can compare it to new data, existing data, and other print profiles. The comparisons result in scores and decisions (match or mismatch). For example, if someone calls your application and claims a particular identity, you can determine if their voice matches a voiceprint profile for that identity.

Create print profiles

The following is a high-level procedure for creating print profiles:

  1. Acquire biometric samples.
  2. Associate samples with identities.
  3. Add the samples to enrollment collections.
  4. Train collections into print profiles.

Adapt print profiles with new media

Use adaptation to maintain the most accurate print profiles possible. Adaptation is the process of updating an existing print profile with new media.

Adaptation improves accuracy as follows:

  • Adjusts to regular background interruptions (for example, the sounds of someone’s habitual calling locations).
  • Tracks shifts in biological characteristics (for example, as the tone of someone’s voice changes with age).
  • Adjusts to various channels (for example, an individual may enroll by landline, but subsequently use a mobile phone).
  • Takes advantage of data collected through the newest input technologies (for example, improved microphones on smartphones).

Near-term adaptation of voiceprints

Near-term adaptation is an automated feature to improve the quality of new voiceprints by training additional audio as soon as possible. The name “near-term” refers to this immediate goal of improving a basic trained print.

Typically, this feature improves voiceprints that were created with a minimal amount of media. This improves performance by reducing the likelihood of false rejections.

The following table describes the near-term adaptation of voiceprints parameters:

Near-term adaptation of voiceprints parameters
Parameter Description
NearTermAdaptationInterval Minimum time between the training and a near-term adaptation.
NearTermAdaptationWatchlist Watchlist to detect if the audio enrolled for near-term adaptation might be spoken by a fraudster.
NearTermAdaptationProfileTag Profile tag to use for fraudster detection in near-term adaptation.

When a session ends, the system detects if there is enough collected audio that match the enrollment criteria (length, time, and so on), and uses it for offline adaptation. These conditions must be met for automatic near-term adaptation:

  1. Session involved the speaker (it was not an offline activity). For example, the speaker engaged with a contact center, IVR, or website.
  2. Session decision must be Authentic (the speaker successfully verified against the current voiceprint)
  3. Session has a VerifyVoiceprint factor where the factor decision is Authentic. (If there’s more than one factor, use the newest.)
  4. Voiceprint has a basic trained status.

Create multiple print profiles per person

You can add a dimension of accuracy to your system by creating multiple print profiles per person.

For example, create a different profile for each channel used by the speaker (landline telephone, mobile telephone, and so on). Because the profiles are specific to the channels, they avoid generalizations that might lower the accuracy of verification decisions.

Lock and unlock prints

You can set print profiles to automatically lock after a number of consecutive mismatches (determined by mismatchesBeforeLock).

The profile locks for a period of time (determined by lockDuration).

For more information, see the configuration parameters.