Voice Verification Use Case

This section describes a voice verification use case as an example. The das-Peak service can receive queries for both enrollment and verification. These processes are explained below.

Simple Infrastructure Proposal

Enrollment

During the enrollment process, the user's voice is recorded. This is a critical step, as all future verifications will be compared against this initial sample. For that reason, it is strongly recommended to ensure high quality in the recording, minimizing noise, distortions, or any condition that may degrade biometric performance.

Once recorded, the audio must be transferred from the user's device to the infrastructure of the Veridas client. This infrastructure is responsible for:

• Sending the enrollment audio to das-Peak.
• Storing the RBR (Renewable Biometric References) returned by das-Peak.
• Storing the original enrollment audio, which will allow regenerating the RBR if the biometric engine is updated in the future.
• Storing the hash of the biometric model version used to generate the credential, which helps determine whether an update is needed later.

Verification

In the verification process, the user's voice is recorded again and sent to the Veridas client’s infrastructure.

This infrastructure must then:

• Retrieve the RBR associated with the user (obtained during enrollment and stored previously).
• Send both the newly recorded audio and the stored RBR to das-Peak.

das-Peak performs the biometric comparison between the incoming voice sample and the stored RBR, and returns a result in the form of a decimal score between 0 and 1. The higher the score, the greater the confidence that both samples belong to the same person.

Each Veridas client must define the appropriate calibration of their audio input according to their specific use case (e.g., microphone or telephone channel), as this can have a significant impact on recognition performance.