Thresholds are used to make decisions based on verification scores. They are not explicitly represented in the API because they are not part of the results, but of their interpretation.

Thresholds are implemented in the client systems to allow for maximum flexibility in the business logic that reacts to the outcome of the verifications performed.

​
Threshold selection

​
0.5 threshold — Balance point

Since a score of 0.5 represents an equal probability of a positive an a negative result, it is the point of maximum uncertainty and the natural choice for the threshold when the correct classification of positive and negative audios are equally important.

The 0.5 threshold is a good starting point when there is no clear bias towards security or usability.

​
Thresholds above 0.5 — Higher security

Higher thresholds are more strict and increase system security at the cost of usability.

  • Higher security: less negative audios will be falsely accepted as positives.
  • Lower usability: more positive audios will be falsely rejected as negatives.

​
Thresholds below 0.5 — Higher usability

Lower thresholds are less strict and increase system usability at the cost of security.

  • Higher usability: less positive audios will be falsely rejected as negatives.
  • Lower security: more negative audios will be falsely accepted as positives.

​
Verification model accuracy

When a system has a strict requirement for either the False Acceptance Rate (FAR) or the False Rejection Rate (FRR) the threshold must be selected considering the overall accuracy of the verification model, introducing a bias towards security or usability to match the requirement.

The model index contains detailed information of the models. Learn more.

ScoresList