Working with voiceprint models

A voiceprint File containing a mathematical summary of the vocal patterns of a person’s voice used in the IAFD product to assist with identity verification and detection. model assists with voice biometric detection Function of the IAFD product that compares a caller’s voice to one or more voiceprints in a watch list to detect if a target is participating in a call. and verification Feature in voice biometrics where the speaker’s voice is compared to a collection of employee or customer voiceprints that should match the call. If a match occurs, the speaker is said to be verified.. Use a reliable combination of audio sources from a person so the system can identify the unique characteristics of the voice of a person.

The maximum number of segments per person allowed for a voiceprint model is 10.

Voiceprint model training

Voiceprint model training is the process of using recorded audio of the voice of a person to create (train) or update (retrain) the model. A model trained from multiple samples more accurately represents the voice of a person than a model constructed from a single interaction In Speech Analytics, an interaction represents a single part of the contact between one employee and the same customer. In Text Analytics, an interaction is the communication session between one or more employees and the same customer with a unifying contextual element.. Also, audio sources with higher quality yield more reliable results. The better the quality is, the better the system can identify the unique characteristics for a voice on interactions.

An important choice made when training a voiceprint model is the left or right channel selection. Channel selection enables you to identify to the Risk Management Component of the Real-Time Analytics (RTA) Framework that provides interaction search and replay functionality, combined with integration of the Real-Time Analytics (RTA) Framework. system which person is speaking.

Channel selection

For interactions natively recorded in an environment that supports stereo recording, or stereo clips uploaded to the system, you can select the channel used to train the voiceprint model. Stereo interactions recorded by the system usually have the employee on the left side and the other party (for example, the customer) on the right of the recording.

Stereo Left / Mono: Use the stereo left / mono channel for what is typically the employee voice in stereo recording environments. Always use the left channel for mono recording environments.
Stereo Right: Use the stereo right channel for what is typically the customer voice in stereo recording environments.

Stereo audio uploaded to the system could have either channel configuration, and you must identify the channel to use when uploading the audio.