TL;DR: A modern voice cloning model needs around 30 seconds of clear speech to produce a voice that holds up on a phone call, and most people have far more than that online already. Callstrike's free AI Voice Cloning Test lets anyone feel that number. You chat with an AI for about a minute, it clones your voice from the conversation, then plays a short workplace vishing call in your own voice and explains how to catch one. The clone is deleted when the session ends and nothing is downloadable.
Why we built a test you run on yourself
I have spent years turning attacker techniques into training, and one thing holds every time. People believe a threat once they have felt it, not when they have read about it.
Telling a finance team that "voice cloning needs very little audio" changes nothing. Letting a finance lead hear their own voice ask a colleague to approve a supplier payment changes how they answer the phone the next morning.
So we built the AI Voice Cloning Test. It is free, it runs in the browser, and it needs no account. It takes about two minutes.
How much audio does it take to clone a voice?
About 30 seconds of someone talking. Not an hour of studio recordings, not a podcast back catalogue. A current cloning model needs roughly half a minute of clear speech to produce a voice that survives a phone line, where the audio is already compressed and nobody is listening for artefacts.
Now count how much of your own voice is public. A conference talk is 45 minutes. A webinar recording is an hour. A single podcast episode is 30 minutes. A two minute WhatsApp voice note is already four times what a clone needs, and the greeting on your voicemail can be enough on its own.
Attackers rarely have to record anyone. The audio is already out there.
What happens in the test
You press start and allow the microphone. An AI answers in a stock voice and asks you a few ordinary questions, about your weekend, your work in general terms, a trip you enjoyed. Nothing sensitive, and it is told never to ask.
While you talk, a call panel shows three lanes on one timeline: the AI, you, and your clone. A counter shows how many seconds of your voice it has collected. Only your own speech counts, so about a minute of conversation gets it to 30.
Then it switches. There is no loading screen and no warning. The clone is built while the conversation carries on, and it takes the AI's next turn in your voice.
In your voice, it plays the kind of call a colleague of yours could receive tomorrow. "Hi, it's me. I'm about to go into a meeting and I'm a bit stuck. Can you approve the supplier payment I just sent over? I'll explain later." Then it tells you how you would have caught it.
How to catch a cloned voice call
You will not hear the clone. That is the uncomfortable part of the test, and it is the point. The defence has to be a habit that works whatever the audio sounds like.
Call back on a known number. Hang up and dial the number you already hold for that person, never the number that called you and never one read out on the call.
Agree a code phrase. A short phrase shared with your finance team, and separately with your family, ends a cloned voice call in one question. Rotate it the way you rotate a password.
Treat urgency as the signal. Pressure plus secrecy is the attack pattern itself. "Before my meeting", "the auditors are waiting" and "don't loop anyone else in" are the scam, not the context. A genuine request survives a five minute delay and a check on a second channel.
What happens to your voice
We built this to forget you. The clone is made only from the live microphone, so there is nothing to upload. It exists only for the length of the session and is deleted when the session ends. The cloned voice only ever speaks the fixed demonstration and the advice, so it cannot be given a script, and nothing is downloadable.
If you want to hear a different version of the same trick, our Voice Phishing Simulator turns your voice into a boss or a colleague live, and the deepfake security training demo does the same for a face on a video call.
The email inbox has had twenty years of filters and training. The phone has had almost none, and a familiar voice is now the cheapest thing in the world to fake. Two minutes with your own clone is the fastest way I know to make that real.
Frequently Asked Questions
Q: How much audio does it take to clone a voice?
A: Around 30 seconds of clear speech is enough for a current model to produce a voice that holds up on a phone call. Most people have far more than that public already, in conference talks, podcasts, webinars, social clips and voicemail greetings.
Q: Is the AI Voice Cloning Test free?
A: Yes. There is no account, no card and nothing to install. It runs on your computer in the latest Chrome or Edge, or on your phone (Safari or Chrome on an iPhone, Chrome on Android), and takes about two minutes.
Q: Do you keep a copy of my voice?
A: No. The voice clone exists only for the length of the session and is deleted when it ends. There is no recording kept and nothing to download.
Q: How do I protect my company from voice cloning scams?
A: Make verification a habit rather than relying on anyone to spot the audio. Call back on a number you already hold, agree a code phrase for payment and access requests, and treat urgency plus secrecy as the warning sign.
Train your employees before an attacker does
The free test shows one person what a clone sounds like. Callstrike runs consented AI voice phishing simulations across your whole workforce, then trains the employees who fall for them, as part of a deepfake security awareness training programme. Try the AI Voice Cloning Test yourself, then see how a vishing simulation trains employees.
Share this article
Found it interesting? Don't hesitate to share it to your friends or colleagues
