Voice and caller security
Voice deepfake detection
Autessa scores every caller utterance on phone calls and in voice chat from 0 to 1 for how likely it is to be a synthetic or cloned voice, labels it Real, Synthetic or Unsure, and applies your action at 0.5 or above, without ever storing a voiceprint.
The problem
A few seconds of audio from a podcast or a voicemail is enough to clone a voice. Your support line, your IT help desk and your payroll agent all trust that the voice on the phone belongs to the person it says it does.
How voice deepfake detection works
Score each utterance
Everything the caller says that is longer than about a second gets a voice score, alongside the call. It never delays the agent’s reply.
Label it
Below the threshold (0.5 by default) is treated as real; at or above is synthetic. Each score is labeled Real, Synthetic or Unsure.
Apply the action
Flag, Ask to verify, or Block. With Block, tools that need verification stay off for the rest of the call, so the agent can transfer to a person or end the call.
Tighten while verifying
While a caller is proving who they are, the bar is stricter (0.35 by default), so a borderline voice isn’t enough to pass.
Every setting
Who can do what
Admins
Set thresholds and actions for the company or per agent.
Agents
Read the score through the Get Caller Info tool.
Reviewers
See the score on every utterance in the call’s details.
What to know
- It’s a signal, not proof. A high score means the audio has the marks of computer-made speech; a low score doesn’t prove identity. Pair it with verification for anything sensitive.
- Phone audio is harder. Phone lines carry less sound than a microphone, so a bad line, a speakerphone or heavy background noise can raise a real voice’s score.
- Short replies (“yes”, “okay”) and anything said while typing on the keypad aren’t scored.
- In voice chat, moments when the caller overlaps the agent speaking are skipped, so echo doesn’t count against them.
- No voiceprints: only the score is kept, and scores are deleted after 30 days.
More in Voice and caller security
FAQ
Voice deepfake detection: questions
Can Autessa detect AI-generated voices on phone calls?
Yes. Each caller utterance is scored for synthetic speech in real time, on phone calls and in voice chat, and labeled Real, Synthetic or Unsure.
Does voice deepfake detection slow down the agent?
No. The check runs alongside the call and never delays the reply. A score applies from the caller’s next turn.
Does Autessa store voiceprints or biometric data?
No. It scores each utterance and keeps only the score, never a profile of anyone’s voice. Scores are deleted after 30 days.
Can I score recordings outside a call?
Yes. The voice score API scores an uploaded clip and also reports whether it carries the Autessa voice watermark.
Turn on voice deepfake detection.
Every thing a caller says on a phone call or in voice chat gets a synthetic-voice score from 0 to 1, labeled Real, Synthetic or Unsure, and a policy action applies at 0.5 or above. Start in Monitor and see what it would catch on your real traffic.
Starts in Monitor · nothing changes for your customers until you enforce
100% customizable, code and all
Make Autessa AI Agent Firewall fully your own.
Use it out of the box, then change anything. Ask Autobot, Autessa’s AI builder, for a new field, screen, report, workflow or agent and it builds it into your copy of the app. Or open the code and edit it yourself. Nothing is locked: the app, its data and its code are yours.
Start free and make it yours