So, you’ve made the decision to invest in voice biometrics to help improve customer experience and mitigate the risk of fraud. How can you make the most of your investment?
Step one is to have a robust enrolment process in place. Ironically, this starts with being able to confirm the identity of everyone you want to enrol in the first place. Once you know who you’re talking to you’ll need to decide how you will capture the individual voice samples needed for the voiceprint.
Enrolment is an important part of a robust voice biometrics solution. You need to make sure that the reference samples are as “clean” as possible, with little or no background noise. The most common use of voice biometrics is in telephone engagements, so it makes sense to use the phone to capture voice samples.
The phone in question can be a mobile, landline, VoIP or even a softphone using a laptop or PC microphone. However, because quality and clarity are important, you should avoid using a speakerphone. Speakerphones are designed to pick up ambient sounds and create echo, both issues that will interfere with creating a suitable voiceprint.
If possible, stick to the same type of device for creating all your voiceprints. If this isn’t possible, train the system to use all the different devices users may employ for enrolment.
Once a caller has been successfully enrolled and verified, it’s best practice to update the speaker’s model. The decision to update or not will be based on several factors; including verification results, the desired level of security and any additional risk factors. If necessary, authentication by other means should be conducted prior to updating with the verified audio.
You may think that getting each user to record their voice for enrolment in a soundproof room, using a professional quality microphone, will give the best results. However, that will not be representative of real-world use of the system, where users will be calling in over the phone.
Biometric voiceprints are a special category of personal data, as defined in the General Data Protection Regulation (GDPR). This means you cannot capture audio data from your callers, for the purpose of creating voiceprints, without obtaining explicit consent from each user.
Getting that consent provides you with the opportunity to give specific guidance to each user. As callers are participating knowingly in the enrolment process, training the user to interact with the system will have a positive impact on future authentication results.
You should be explicit with your instructions and encourage behaviour that achieves the best results. Adapting your IVR to provide prompts to guide callers through the enrolment process is essential. Also, providing users with a quick reference guide to the process makes a lot of sense. Provide enough detail on the process, so users understand the need to capture a good voiceprint and what good looks, or rather sounds like.
Users should be encouraged to speak naturally, at a conversational pace. Voice biometrics analyses multiple components of a person’s speech, including pace. So, speaking at your usual pace is necessary to ensure the system better differentiates correct users from impostors.
By acting on these tips, you will increase the effectiveness and efficiency of your speaker verification solution and ensure both you and your customers realise the anticipated benefits.
Discover how iTel is leveraging voice biometrics for authentication in a workforce management application.