Voice Verification catches identity fraud in AI-led interviews by comparing the voice on the AI Call against a voiceprint built from the candidate's earlier qualifying video answers. The result appears as a badge on the candidate profile alongside TalkScore — giving your team a clear, structured identity-confidence signal on every completed interview.

Most candidates completing AI assessments today are doing it on a mobile phone. In markets like the Philippines, that number is close to 90%.

The cheating methods that most proctoring tools are built to catch — screen switching, opening a second tab, toggling to another app — require a desktop. On mobile, there is nothing to switch to. The signal those tools generate in a mobile-first context is not meaningful.

The method that actually works today costs almost nothing: two phones. The candidate holds one up to the camera. A second person answers the AI Call on the other. No screen switch happens. No visual anomaly appears. The recording looks clean.

There was a second constraint we did not want to ignore. A full-session video interview carries roughly 200 MB of data. For a candidate completing an assessment on a mobile data plan, that is a real cost — and it shows up directly in completion rates. Asking every candidate to absorb that overhead to catch a threat the tool is not actually detecting is not a trade-off worth making.

These were the two things we designed around: catch the cheat that is actually happening, without making it harder — or more expensive — for candidates to complete the assessment.

Voice Verification is built around the cheating method that is actually happening. Multi-speaker diarization analyses every AI Call recording for the number of distinct voices present. If three or more are detected — the candidate, the AI agent, and a third speaker — the system hears the handed-off interview and flags it, regardless of what the camera shows. No screen needs to be watched. No visual cue needs to appear.

Enrollment uses the qualifying video the candidate already records as part of your campaign. There is no separate proctoring session, no additional step for the candidate, and no extra bandwidth overhead. The AI Call itself is voice-only. In areas where connectivity is poor, the assessment can also run over the phone. The architecture was chosen specifically so that protecting interview integrity does not come at the candidate's expense.

It is a natural instinct to want both — voice verification and screen capture running alongside each other. We considered it and chose not to ship screen capture as part of this feature.

The fraud pattern screen capture is meant to deter — someone else answering on behalf of the candidate — is already caught by multi-speaker diarization. If a third voice appears on the AI Call recording, the system flags it regardless of what the camera shows. There is no additional signal that screen capture adds for this scenario.

What screen capture does add is cost: bandwidth overhead, friction at the point of assessment, and a privacy concern for candidates. We chose not to ship it because it would ask more of every candidate without giving your team anything that Voice Verification does not already provide.

Voice Verification can flag low-quality voice cloning attempts when the AI Call voice does not sufficiently match the enrolled voiceprint. Comparisons between a real voice and its Text-to-Speech-generated version have resulted in a “no match” outcome under the same verification process.

Voice Verification compares the voice on the AI Call against the enrolled voiceprint captured from the candidate’s earlier qualifying answers. Differences introduced by synthetic or altered speech may reduce the confidence score and lead to a mismatch result.

More advanced synthetic voices or higher-quality cloning techniques may still pass if they closely resemble the enrolled voiceprint. The system is designed to verify voice consistency between the AI Call and the original enrollment samples.

Once Voice Verification is enabled for your tenant, the three steps below run automatically for every candidate who has a voiceprint enrolled.

Note: "Voice verified" means the voice on the AI Call matched the candidate's earlier qualifying answers above the confidence threshold. It does not mean a government ID was verified.

Nothing changes for the candidate. Whether the feature is on or off, the application flow looks identical — no extra prompt, no additional disclosure, no new step.

Once Voice Verification is enabled, you can choose which audio or video questions will be used to build the voiceprint.

Note: The Capture voiceprint checkbox only appears on audio and video question types. Text, dropdown, and other non-speech questions cannot have voiceprint enrollment enabled.

Once a candidate completes an AI Call and a voiceprint exists for their application, the badge populates automatically.

The following attributes are also generated and available in your existing campaign filters, exports, and rules engines. No integration changes are required.