I cloned my voice with free AI tools – and tested a scam call on my family

Your social media posts could give scammers all the audio they need. Would your loved ones spot the fake?

Read full bio October 23, 2026, 8:32 am ET

piyaphun via iStock / Getty Images Plus ZDNET’s key takeaways How easily can you clone a voice from public sources? I put it to the test. AI tools have made it easy, and you need a smaller sample than you think. Here’s how to keep yourself protected from voice-based scams. “Hi, it’s me. I’m on a stranger’s phone, mine’s about to die.”

If you heard this down the line, you’d be unlikely to hang up, and that is what cybercriminals running phone- and voice-based scams count on.

We’ve all seen versions of this play out over text and messaging apps: a plea for help, a story about a loved one in an accident, and an urgent request for money to cover hospital bills. These scams work because they create panic, and panic works best when we believe someone we care about is in trouble.

Also: Chrome warns you about suspicious websites now – how to turn on Google’s new scam protection

Until recently, impersonating a family member’s actual voice was difficult. A convincing clone once needed plenty of source material and specialist tools, but that’s no longer the case. Cheap, free, easy-to-use AI voice-cloning tools are now widely available, and many of us have already supplied the raw audio material by posting videos of ourselves online.

So I decided to find out how easy it is to clone my own voice and whether my friends and family would fall for a scam built on it.

For this experiment, I recorded voice samples and downloaded recordings from my interviews as a cybersecurity host on Spotify. When I created my own samples, rather than uploading raw audio files into each AI model, I deliberately compressed them into.WAV files. The reasoning is that raw audio files can contain far more information for an audio sampler to analyze than.MP3 or.WAV files, which are compressed for sharing, to reduce file sizes, and for use on social media platforms.

Also: Dealing with silent robocalls? This is why scam callers keep quiet

Rather than give myself an advantage, I wanted to use audio samples of a similar quality to those you could rip from the average Facebook, Instagram, or TikTok content that could be exploited by scam artists. If someone tries to impersonate you, these are the sources they will visit first to find samples of your voice.

If we’re using AI for the audio, let’s use AI for the scammer’s script, too. I tasked Claude with developing a short script for a phone-based scam, emphasizing urgency, panic, and a few deliberate pauses for a conversational effect. Here’s the script I developed with a few prompts, a task that took no longer than three minutes.

Also: Your iPhone just got a hidden anti-scam upgrade in iOS 27: How to enable it

“Hi… it’s me. [pause 1.5s] Sorry, I know this is out of the blue. [pause 1s] I’m on a stranger’s phone, mine’s about to die.”

“I’ve had my bag taken… [pause 1.5s] wallet, cards, everything. [pause 1s] I’m stuck at the station and the last train home goes in about forty minutes.”

“I can’t get into my banking without my phone, and I don’t want to worry anyone else. [pause 1s] Please don’t say anything, I feel so stupid.”

“I just need enough for a ticket and to get me through tonight. [pause 1s] I’ll pay you back first thing, I promise. [pause 1.5s] Can you help me?”

“Thank you. I’ll get them to message you the details now… [pause 1s] I’ve got to go, the battery’s going.”

While some scam artists might sign up for premium services, I chose to stick with free options during my testing, which included standalone services and open-source tools. I used the same audio sample across every platform, but on two platforms, I had to record audio samples there and then to create a voice clone.

My first stop was ElevenLabs.io. There’s a free tier, so I uploaded my samples and listened to the result.

Also: A low-tech solution from the past may be your best defense against AI deepfakes

As an AI-based voice cloner, ElevenLabs.io is top-tier. I was shocked by how clean the audio was, but unfortunately (or fortunately?), it wasn’t fit for my dubious purpose: It was too clean, too professional, and made me sound more like a news anchor than a frightened woman seeking help.

I then signed up for alternative platforms, including Kikivoice.ai, Voicebox (using Qwen TTS 1), SoundTools, Epidemic Sound, and voxcpm2.org.

Results varied widely across platforms. Some samples were good enough for a TikTok Reddit story voiceover and nothing more; others made me sound like I was choking; others used a “British” voice that sounded like a high-pitched caricature; and others were relatively convincing but also added strange pauses, heavy breathing, and odd pitch changes.

One common thread, however, was that all of them — with the exception of ElevenLabs — were flat in some way and lacked the emotion you would need in a voice-based scam.

I then visited IndexTTS, an open-source and free AI voice cloner that promised emotive control. It’s an excellent cloner tool that mimicked my voice well, but when I adapted my voice clone using “sad” and “afraid” settings, it turned it into a strangely breathy, husky whisper that I’d rather my editors — let alone my intended victims — never hear.

After I recovered from the embarrassment of hearing that sample, I revisited Voicebox, which lets you pick which AI model you want to use for voice cloning. I opted for Chatterbox, and while it was somewhat flat and had no emotional control options, it provided the closest clone I could use.

I targeted five people in my close circle, including family members and friends. After cloning my voice, I then used a temporary phone number to call each of them, as this more closely mimics what a scammer would do.

Out of the five test subjects, four immediately knew something was wrong and flagged the call as fake. Two didn’t realize it was meant to be me, and two said it sounded like me, but clues in the call suggested something wasn’t right.

Also: How to keep your conversations with ChatGPT, Gemini, Copilot or Claude as private as possible

One did believe it was me. Because she was of an older generation and slightly hard of hearing, it was harder for her to distinguish the familiar clues in a voice that make someone recognizable and trustworthy. (This wasn’t necessarily fair, I know, but it was a scam scenario that could realistically play out, and I was swiftly forgiven.)

I then asked for feedback on the clone scam, and the common indicators that flagged the fake voice were that the voice was too flat; the intonation and rhythm were strange; some words were mispronounced; and it was not my usual phone number.

If you’re wondering at this point how much audio does someone need to clone your voice, the answer is: Far less than you would expect.

With no more than around 10 seconds of audio, I was able to create some almost passable versions of myself, and while they wouldn’t fool most people close to me, they weren’t too far off what could potentially work on a phone call.

Many of us have come across the automated calls and bots that call our phones, flatly promise a new job opportunity or insurance payout, or claim we are in trouble with the IRS. And most of us cut off those calls within seconds.

Also: Listen up: Audio deepfakes are now the leading AI impersonation scam

AI isn’t in a position to make these scams more personal or convincing — at least, not yet, as clones still sound flat and lack the human factor. But given the rapid pace and evolution of AI models today, I don’t think it will be long before AI models can be abused to create emotive, convincing, personalized voice scams (also known as vishing) just by ripping audio from a stray social media video or Spotify track.

AI can also clone your voice from a TikTok, Facebook, or Instagram video. Cocial media platforms are a hotbed of source material, including videos more than a minute long, that can be used by attackers to harvest enough samples for convincing audio. The more voice samples an attacker has, the better, because an AI model can learn to sound convincing — mimicking an accent, particular words and phrases, inflections, and the natural rhythm of someone’s voice.

Also: This simple ChatGPT trick helps you spot scams before you click or respond

If you’re already on platforms like TikTok and frequently post videos of yourself talking, there are already many samples of your voice out there. You may be at higher risk of AI audio cloning than the average person (at least, in the future), so be aware of it.

While it’s becoming harder to distinguish between genuine and fake audio, listen for strange patterns in speech, such as unusual words or phrases that you don’t associate with the person who is on the other end of the phone, incorrect grammar (outside of what you would normally expect), strange pauses or abrupt ends to sentences, and differences in accent.

Also: Fake calendar invites can infect your system, and they’re surging – how to protect yourself

Other indicators include an unknown number, background noise you wouldn’t normally expect, and a lack of conversation cues we normally experience in a human-to-human conversation, such as polite small talk or questions about personal details only a genuine caller would know.

If you suspect that the entity on the line isn’t human, or if they ask for money or payments, hang up. If they claim to be family, friends, or a colleague, ring them back on the number you have for them. If it was a call out of the blue, it was probably a scam.

Make it known. If you believe you have fallen victim to a scam where someone has cloned your voice and is impersonating you, the first step is to tell friends, family, and potentially your colleagues. Report the issue to law enforcement and associated fraud organizations, report any posts or media online that you’ve found using your voice or identity, and consider reaching out to service providers to make them aware of the incident.

Some of the best advice we can give you, however, is to create a password for conversations that could result in financial theft.

AI apps, videos, posts, and online material seem to get more convincing every month. It’s not only for cute cat videos — governments generate their own content for propaganda, bot farms spread political messaging and adapt to conversations based on AI prompts, TikTok is full of AI-generated anime stories and movies, and cyberattackers use AI to streamline and automate traditional cyberattacks.

It might not be long before audio and visual phishing scams are the norm, rather than the exception.

If a scammer is determined enough and has secured a few samples of your audio, that may be enough to clone your voice and impersonate you to service providers, friends, or family.

Also: 1Password’s price hike sent me looking for alternatives – here’s what I chose

If you are concerned about becoming a victim of this type of scam now or in the future, the easiest option is one we talk about at ZDNET a lot — setting a password.

However, we’re not talking about complex combinations of letters, numbers, and symbols. Instead, agreeing on a family-style password shared between trusted contacts could be the simplest option. If a conversation involves money, transfers, or personal information, treat it as suspicious unless the password is used. We’ve already set this up in my family.

This idea can apply to companies, too: If you have a service provider that lets you add password protection to conversations not made in person, ask it to add it to your account. Perhaps this is a concept organizations should adopt today, too, with either two-step verification required to approve invoices and purchases or a password that rotates frequently.