Glossary

This topic describes common terms and acronyms.

Accuracy

A measure of the ability to reliably verify or identify speakers. It is a function of the underlying biometrics engine, the configuration used, and the quality of collected audio.

Adaptation

Adaptation is the process of using new media to update an existing print for the purpose of improving it.

AKS

A managed Kubernetes service that lets you quickly deploy and manage clusters.

ANI

Automatic number identification (ANI) is a feature of a telecommunications network for automatically determining the origination telephone number on toll calls for billing purposes.

API Gateway (Ingress)

In Kubernetes, an Ingress is a component that routes the traffic from outside the cluster to your services and Pods inside the cluster.

In simple terms, the Ingress works as a reverse proxy or a load balancer: all external traffic is routed to the Ingress and then is routed to the other components.

Audio clustering

Audio clustering is the process of finding audio segments with similar biometric characteristics in an audio collection. The system assumes audio segments that belong to one cluster were spoken by the same person.

Audio collection

An array of audio segments (audioModels) that can be processed with a single operation. Applications can create audio collections for any purpose.

Audio layers

An audio layer is part of an audio, that has start and end points, and optional fields of meta data.

Authentication & fraud factors

Various authentication mechanisms that return Match or Mismatch results with scores. Factors include voice biometrics, face recognition, behavioral biometrics, and more.

Authentication factors can either contribute to authentication or risk scores, while fraud factors only generate risk scores.

Biometric sample

A segment of media that the system adds to a collection of segments used when training a print profile.

BLL

Business Logic Layer. Refers to the flow that the system executes when a call comes into the contact center (determines when to enroll the caller, when to verify the caller, and so on).

CLI

Caller Line Identity, typically the caller’s telephone number. This is a common, industry term used in call centers and telephony technologies.

Client Request

An external identifier that was created by the client application.

Cluster purity

A score that signifies the quality of the cluster.

Configuration set (configset)

Group of parameters that define system behavior. Often abbreviated as configset. There is one scope configset for each scope (with the same name as the scope), and any number of operational configsets within a scope.

Convoprint

A print representing a person’s unique conversational patterns such as lexicons, specific phrases, and word frequency.

Convoprint (Voice)

Ability to create a conversational model from the audio recording of a caller, which is first transcribed and then modelled. The technology produces an output score/risk that represents the similarity in text with the previously enrolled model of a Person/Fraudster.

This can be used in the contact center interaction.

Convoprint (Digital)

Ability to create a conversational model from the text of a live chat; for example, a user on the chat. The technology produces an output score/risk that represents the similarity in text with the previously enrolled model of a user or fraudster.

This can be used in the live chat scenario.

Deviceprint

A print representing a person’s unique usage of an electronic device such as a mobile phone, touch screen, or keyboard.

DTMF

Dual-Tone Multi-Frequency.

EER

Point at which the false accept and false reject rates meet. This is a standard accuracy metric of biometric systems. Usually, systems do not work at this point and adjustments are made for security considerations.

Engagement

An aggregation of all sessions related to a customer transaction. Engagements enable decision-making that is based on multiple factors used in the transaction, and they enable tracking the activities of that transaction.

Equal error rate

Point at which the false accept and false reject rates meet. This is a standard accuracy metric of biometric systems. Usually, systems do not work at this point and adjustments are made for security considerations.

ETL

Extract, transform, and load.

False accept

An error situation that occurs when an imposter is accepted as a true speaker. All applications expect a small number of these errors known as the FA rate.

False reject

An error situation that occurs when an authentic speaker is rejected as an imposter. All applications expect a small number of these errors known as the FR rate.

Fraud

Fraud is an attempt to misrepresent a speaker’s identity.

Fraudster

Someone who is known to be an imposter and is tracked in a fraudster watchlist.

Grammar

A speech grammar that defines the words and phrases known to a speech recognition engine (Nuance Recognizer). The syntax of the grammar file as well as the lexicon and semantics are determined by the engine used, and is not controlled or specified by biometrics products.

Group

A collection of persons that can be analyzed or processed by a single operation. For example, the application can perform an identification that uses the print profiles of all members of a group. Typically, members of a group share common attributes (for example, a group of suspected fraudsters).

High availability

A deployment architecture that uses redundant hardware and software to avoid out of service situations.

IdP

Identity provider (IdP) is an online service or website that authenticates users on the Internet by means of security tokens, one of which is SAML. For example, NEAM (Nuance Enterprise Access Management) is an IdP.

Imposter

Also spelled impostor. Unknown person who claims the identify of someone enrolled in the system. The system detects imposters by compared their speaking voice to one or more previously enrolled voiceprints.

Ingress

In Kubernetes, an Ingress is a component that routes the traffic from outside the cluster to your services and Pods inside the cluster. In simple terms, the Ingress works as a reverse proxy or a load balancer: all external traffic is routed to the Ingress and then is routed to the other components.

Intelligent detectors

A layer of authentication techniques that consider characteristics of calls or sessions.

Known fraudster

A print for someone who is known to be a fraudster. (The true identity of the fraudster might not be known.)

Kubernetes

An open-source container-orchestration system for automating application deployment, scaling, and management. It was originally designed by Google, and is now maintained by the Cloud Native Computing Foundation.

Mission

A system process that has definitive start and end points. Missions include enrollment, verification, identification, adaptation, fraud detection, training, liveness detection, and print deletion.

Oauth2

An open standard for access delegation, commonly used as a way for Internet users to grant websites or applications access to their information on other websites but without giving them the passwords.

Orphan

A call that is not associated with a Person ID.

Person

Someone whose biometric data is to be analyzed for authenticity. Persons can be added to groups to allow performing identification tasks.

Playback attack

A fraud attempt that uses an audio recording (not a live voice). The biometrics system can detect various forms of playback attacks.

Playback detection

A feature that adds a layer of security by testing incoming audio to see if it represents live speech or if it fraudulently uses a recording of an authorized speaker’s previous interactions with the system.

Print

Processed media that results in a machine-recognizable pattern. For example, convoprint, deviceprint, or voiceprint. Once a print is created for a person, the system compares them to future engagements with that person and adjusts authentication and risk scores based on the result. Prints can be improved over time through adaptation, which uses new media collection to retrain existing prints.

An entity that holds the print information. The print contains metadata information, enrollment segments and footprints, and the print template (binary buffer) generated in the train process. Currently the following types of prints are supported: voiceprints and convoprints.

Profile template

A binary representation of the unique characteristics of a person’s biometric traits.

RCC

Recording controller connector.

Request

A single API call uniquely identified by Request ID.

Risk-based decisions

Risk-based decisions (abbreviated RBD), allow users to tune the system according to the required false acceptance rate.

Risk engine

Returns an overall assessment of all executed factors in the form of session-level authentication decisions (match, mismatch, or inconclusive for voiceprints) and risk levels (low, medium, high) while using fusion mechanisms to determine a fraudster in a session.

Risky sessions

Sessions that received a risky conclusion.

RTP

Communication protocol that defines a standardized packet format for delivering audio and video over IP.

Scope

Logical view of the system, enabling partitioned data access for multiple tenants.

Session

All operations belonging to a single interaction of a person with the system (for example, a phone call).

SIPREC

Communication protocol for recording systems that may be used to acquire audio.

SRTP

Secure RTP.

VDA

Virtual Derived Audio. This is a feature that saves traces of audio used for processing when the original audio contains sections that were not processed. This feature makes it easy to know which sections were used during processing (for example, which audio resulted in a match decision when compared to a print) and which sections were ignored.

Voiceprint

A print representing the unique characteristics of a person’s speaking voice. A voiceprint is uniquely identified in the system by the combination of the speaker ID and a voiceprint tag. A speaker can have multiple voiceprints. A voiceprint can be created before a speaker is enrolled or trained.

VoIP

Voice-over-Internet Protocol refers to the delivery of voice across the internet.

Watchlist

A watchlist is a group of suspected or known fraudsters. You can compare media to the contents of a watchlist to detect fraudsters in realtime.

Wolfpack

A collection of imposter audio used to calculate a false acceptance distribution curve for a target voiceprint. This curve is used during Imposter Detection.