Quick answer
PlayHT was the strongest of the five deepfake tools tested against VoiceCheck, yet it was blocked within VoiceCheck's recommended configuration. The only PlayHT clones that crossed the 60 test threshold were low landline-quality attempts scoring 60 to 65, well below the 79 to 89 range genuine users reach, and referred to a Help Desk in a real deployment. On higher-quality laptop and cell phone audio, PlayHT deepfakes did not pass.
Verdict
Blocked, with Help Desk check
Laptop (best 55) and cell phone (best 34) clones stayed below the pass line of 60. Low-quality landline (best 65) passes go to a Help Desk, not straight in.
AI voice cloning tools like PlayHT can copy a person's voice from seconds of audio, so a fair question follows: can a cloned voice simply walk past VoiceVantage? To answer it with evidence rather than claims, the Institute for Cyber Security and Resilient Systems at Ontario Tech University (UOIT) tested VoiceCheck against the five most popular deepfake tools, including PlayHT. This page shows PlayHT's results in full.
How the test worked
1. Clone the voice
PlayHT was used to create deepfake copies of enrolled users' voices.
2. Try to log in
The clones tried to pass as the real user over a laptop microphone, a cell phone and an 8kHz landline.
3. VoiceCheck decides
Scores below 60 are rejected. A low pass under the 79 to 89 genuine range goes to a Help Desk.
Inside the 533-test study
The Ontario Tech study ran in two phases across all five deepfake tools. In every case, a score of 60 that stays below the 79 to 89 genuine range is referred to a Help Desk, not accepted.
Phase 1 · True identities
178
deepfaked pass-phrases (veriphrases)
1 passed the score of 60 · 0.56%
Phase 2 · False identities
240
deepfake attempts across every attack scenario
Internal attacker: 6 of 120 · 5%
External attacker: 4 of 120 · 3.33%
Genuine users
79-89
the score real users reach with their own voice
Prior UOIT tests show valid users consistently score 79 to 89, and often higher. The pass line of 60 sits well below that.
PlayHT vs VoiceCheck: the scores
PlayHT's highest deepfake score against VoiceCheck, by audio channel
Pass threshold set to 60 for single-factor testing. Genuine users typically score 79 to 89. Scores from independent UOIT testing.
PlayHT deepfake score Pass threshold (60) Genuine user range (79 to 89)
55
Laptop microphone (high quality)
best deepfake score: 55 / 100
Blocked (below pass)
34
Cell phone (medium quality)
best deepfake score: 34 / 100
Blocked (below pass)
65
Landline 8kHz (low quality)
best deepfake score: 65 / 100
Help Desk referral in production
What these results mean
The pattern is consistent and easy to read. Even the strongest deepfake tool in the study only approached the pass line at the lowest audio quality, a landline call. Those few passing scores landed between 60 and 65, which sits far below the 79 to 89 band that genuine users reach with their real voice. In a live VoiceVantage deployment, any score in that low range is not treated as a clean pass. It is referred to a Help Desk for a second check, or the passing threshold is raised for low-quality channels.
On the audio quality most customers actually use, a laptop or a mobile phone, PlayHT clones did not pass at all. In the earlier laboratory phase, PlayHT reached the 60 mark only once across 359 attempts, and scored zero on every VOIP test.
Summary by audio channel
| Audio channel | PlayHT best score | Genuine user range | Outcome |
|---|---|---|---|
| Laptop microphone | 55 | 79 to 89 | Blocked |
| Cell phone | 34 | 79 to 89 | Blocked |
| Landline (8kHz) | 65 | 79 to 89 | Help Desk referral |
Where this protection matters
Contact centers
Callers are verified by voice in about 1.4 milliseconds, so agents are not left to judge a cloned voice by ear.
Banks and payments
A voice check before a payment or an account change adds a step that a cloned voice has to beat, and in testing it did not.
Help desks and resets
Password resets are a favourite target for voice fraud. Low scores go to a person for a second check instead of passing.
About PlayHT
PlayHT is a widely used AI voice generator that offers instant voice cloning from a short clip and a higher-fidelity clone trained on longer audio. In this study it produced the most convincing clones of the five tools tested, which is why its results are the most demanding test of VoiceCheck. That makes PlayHT the right benchmark: if VoiceCheck holds against the strongest tool, the weaker tools follow.
Why VoiceVantage blocks PlayHT deepfakes
VoiceVantage validates a voice in as little as 1.4 milliseconds and is built on many years of signal-processing research. Its recommended configuration pairs the score threshold with a Help Desk referral for low scores, so the small residue of low landline passes seen in testing is caught before it reaches an account. The result is a system that stays secure against advanced voice cloning while keeping the experience fast for genuine users.
Every attempt is scored
Each voice is scored from 0 to 100. Genuine users land at 79 to 89, far above the pass line of 60.
Low scores never pass quietly
A score that clears 60 but stays under the genuine range is sent to a Help Desk for a second check.
Fast for real callers
Verification takes about 1.4 milliseconds, so genuine users are not slowed down.
Compare VoiceVantage with the other tools
Frequently asked questions
Can PlayHT voice clones bypass VoiceVantage?
In independent UOIT testing, PlayHT was the most capable of the five deepfake tools tested, yet VoiceCheck blocked it within its recommended configuration. The only scores that crossed the 60 test threshold were low landline-quality attempts scoring 60 to 65, well below the 79 to 89 range genuine users reach, and referred to a Help Desk in production.
What was PlayHT's highest score against VoiceVantage?
65, on a low-quality landline attempt. On higher-quality laptop and cell phone audio, PlayHT deepfakes did not pass the threshold.
Is VoiceVantage safe against AI voice cloning?
Yes, within its recommended configuration. Across 533 independent attempts using all five leading deepfake tools, the pass rate was very low, and passing scores were confined to low landline quality and referred to a Help Desk.
How was this tested?
By the Institute for Cyber Security and Resilient Systems at Ontario Tech University (UOIT), using PlayHT to generate deepfake voice clones and attempting to authenticate against VoiceCheck across laptop, cell phone, and landline channels.
See VoiceVantage block deepfakes for your organization
Independently tested against the five leading AI voice-cloning tools.
Source: VoiceVantage VoiceCheck Verification Capability Analysis using five top deepfake apps, Ontario Tech University (UOIT), 2025. Read the announcement: VoiceVantage blocks the top five deepfake voice products across 533 independent tests.