Authentication and risk factors

Gatekeeper combines positive and negative factors to confirm the authentication of people and prevent access by fraudsters as follows:

  • Your application uses the API to control risk factors.
  • The underlying configuration tunes the performance of the factors.
  • The web applications expose factors as needed. For example, when fraud analysts search, view, and investigate sessions.

Authentication factors

Authentication factors return Match or No Match results and scores:

  • A Match is a strong indication of an authentic session. For example, confidence is high when an audio matches the voiceprint of a claimed ID.
  • A No Match has little effect on risk levels. For example, a caller isn’t necessarily a fraudster just because the audio doesn’t match.

The following table describes various authentication factors:

Authentication factors
Authentication Factor Description
IdentifyConvoprint Compares audio to a collection of previously trained convoprints. This factor requires a group of print profiles. Applications can build groups of print profiles at runtime.
IdentifyVoiceprint Compares audio to a collection of previously trained voiceprints. This factor requires a group of print profiles. Applications can build groups of print profiles at runtime.
VerifyConvoprint Compares incoming audio to a previously trained convoprint. This factor requires an enrolled person convoprint.
VerifyVoiceprint Compares incoming audio to a previously trained voiceprint. This factor requires an enrolled person voiceprint.

Fraud factors

Fraud factors help to confirm the authenticity of a session or to raise suspicions. No single fraud factor is definitive. Instead, the factors accumulate during calls and contribute data points towards a decision. Typically, your application combines the results of as many fraud factors as needed depending on the acceptable risk.

The following table describes various fraud factors:

Fraud factors
Fraud Factor Description
ChannelPlaybackDetection Confirms that audio files are live recordings, and not the fraudulent playback of previously recorded audio.
FootprintPlaybackDetection (Deprecated) Detects playback of a surreptitiously recorded audio. Gatekeeper implicitly identifies suspected audio segments during the verification process.
KnownFraudsterDetectionConvoprint Detects when a fraudster with a pre-enrolled convoprint accesses the system.
KnownFraudstersDetectionVoiceprint Detects when a fraudster with a pre-enrolled print profile accesses the system.
LivenessDetection (Deprecated) Ensures the audio is live speech and not a recording.
SuspiciousSession Marks suspicious sessions for further investigation by fraud analysts.
SyntheticSpeechDetection Detects audio tampering by a variety of electronic manipulations.
ValidateCallerId Validates the authenticity of caller telephone numbers.
ValidateText Uses speech recognition to confirm the spoken text during identification and verification missions.

There are multiple factors to detect known fraudsters, and if more than one of them returns a match, the system fuses the results to identify the most dominant fraudster in the session. (This requires fraudsters to belong to the same watchlist.)

For more information: