Enrollment

Enrollment is the process of registering an individual in the voice biometrics system. A person must be enrolled before voice biometric analysis can occur.

The result of enrollment Process in the IAFD product where an employee, customer, or target is registered and the system is then able to assist with identity verification and watch list detection when that speaker is on a call. is a trained 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. The model represents the voice of a person. Voiceprint models apply to employees, customers, and targets.

Enrolling a person into voice biometrics requires:

Unique identifier for an employee or a customer

The identifier is one or more data items for both employees and customers. The identifier provides a unique “key.”
- For employees, the unique identifier is the employee ID. Some areas of the user interface use the label Employee Name when referring to this identifier. The Employee Name value is the employee ID, which is unique for each employee in the system.
- For customers, the exact fields used vary from one deployment to another.
Trained, biometric voiceprint model:

A trained voiceprint model requires at least one 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. containing the voice of the person. You can select interactions for enrollment using rules-based, Automatic Enrollment or by Manual Enrollment. You can also import audio recorded outside the suite to use for enrollment.

Trained, voice biometric voiceprint model

The objective of enrollment is a trained voiceprint model that represents the voice of a person. Voiceprint models apply to employees, customers, and targets.

Audio used for enrollment

You can use recorded interactions or imported audio for enrollment. Recorded interactions are contacts captured by the Recorder.

However, the following types of recorded interactions cannot be used for enrollment:

Mono interactions with more than two people speaking.
Stereo interactions where two or more people take turns speaking on a single channel.

Imported audio

Imported audio originates from a source outside the suite. Before importing, edit the audio so it has only the voice of one person. You can use a third-party audio editor to eliminate unwanted parties from the recording.

Stereo: Use for manual enrollment, provided only one person is speaking per channel.
Mono: Use for manual enrollment, provided only one person is speaking on the recording.

Amount of audio needed

Ideally, three minutes of speech trains the voiceprint model. However, more speech by a person yields a higher-quality voiceprint model.

The length of an interaction can last seconds, minutes, or hours. However, the system uses net speech. For example, on a three-minute interaction, the person whose voiceprint model you are training speaks for only 70 seconds. Therefore, only 70 seconds of speech is available for training. The length of an interaction can have little bearing on how much net speech is available for a person. The maximum number of interactions allowed per person for a voiceprint model is 10.

Multiple voice samples result in a more accurate representation of the voice of a person. The more accurate the voiceprint model, the better the system can identify the unique characteristics of a voice. An accurate voiceprint model results in better Verification and Detection.
Whenever possible, and to get the best voice biometric results, obtain voice samples from recordings created by the same system where voice biometric analysis occurs. In addition, try to get voice samples from multiple device sources, such as a landline and a mobile device.