Train Your Users For Best Results

If you read our voice biometrics tutorial, you are now fully aware of the importance of capturing the cleanest possible audio to get a valid result from a voice biometrics engine.

For a Passive voice biometric system, the person speaking does not know that the voice biometric analysis is happening. Therefore, getting the best audio means attending to the system recording components and creating an experience for the speaker that leads to capturing speech without becoming artificial and obvious that speech is your objective.

On the other hand, for the Active voice biometric use case, in which the person speaking is knowingly participating in the authentication process, it stands to reason that training the user to interact with the system properly will have a large impact on your results. Experience has absolutely shown this to be the case. You can and should be explicit with your instructions and guidance in order to encourage behavior that achieves the best results for both the person speaking and your business. By doing so, you will achieve higher user satisfaction and also avoid the majority of support tickets related to voice biometrics.

In order to help you train your users, we publish a Best Practices Training Guide in our customer help center, available within our dashboard to production customers. Several of the principles to keep in mind as you educate your users are the following:

1. Remind the user that the purpose of the system is to both save the user time and provide stronger security for the user's account.
2. Provide enough detail on the process so the user understands the need to first capture a good voiceprint that is then later used to conveniently verify identity in just a few seconds.
3. Because the quality and clarity of audio are important, tell your users not to use speakerphones. Speakerphones are designed to pick up sounds all around a room and they also create echo. Both of these speakerphone problems strongly interfere with the user's voice.
4. Explain to the user that consistently using the same type of device will give consistent results. Most people don't realize the extent that voice over a mobile phone is compressed compared to a landline phone. Those differences are fairly minor to a human but can show up as larger discrepancies to the voice biometrics algorithms. To address this issue, it is best to always use the same type of device or train the system on all the devices the user will use.
5. Keeping with the theme of consistency, instruct your users to speak naturally at a conversational pace. Voice biometrics examines numerous aspects of how you speak, including pace, so speaking consistently is one way to ensure the system more strongly differentiates the correct user from impostors.
6. Somewhat obviously, because the system is trying to listen to a user's speech, the user needs to be in a place to speak such that background noise is as low as possible. Background noise interferes with the user's voice and will likely lead to a false rejection.
7. Use the error messages from our system to instruct the user. For example, we'll tell you in our API response if the audio has too much noise. You can tell the user that's the problem and ask the user to move to another location.

By keeping these ideas in mind, you will increase the effectiveness and efficiency of your voice biometrics system, achieving both a higher user satisfaction and acceptance rate as well as minimizing your support costs.