Learn
The Truthring learning centre explains synthetic speech in three parts: how voice cloning and voice detection actually work, how voice fraud is carried out and what to do about it, and how to obtain and read a detection result honestly. Twenty articles, written to be useful rather than reassuring.
If a call is happening now, stop reading. Hang up and ring the person or organisation back on a number you already had, from your contacts, a card or a statement. Whoever was on that line cannot answer a phone you dial yourself. Everything on this page will still be here in ten minutes.
What this section is for, and what it will not do
Synthetic speech is discussed almost entirely in two registers: alarm, and reassurance that a product has solved it. Neither helps somebody who has a recording in front of them and a decision to make. These articles are written for that person, whether they are a parent who took a frightening call, a journalist weighing a leaked clip, a finance controller who nearly authorised a transfer, or a student trying to understand the field.
The rule we hold ourselves to throughout is that an uncomfortable answer is more useful than a confident one. So these pages say plainly that listening does not work, that a clean detection result is not clearance, that phone audio is the hardest case there is, and that recourse for someone whose voice has been cloned is thinner than anybody would like. Where a number would be persuasive and we do not yet have one, there is no number.
What this section will not do is help anyone build a better fake or slip one past an analyser. We describe capabilities and their consequences; we do not publish steps, tool comparisons, settings or evasion techniques, and no amount of framing a request as research will change that. It is a standing editorial position rather than a case-by-case judgement, and it means some reasonable questions stay unanswered here. Our acceptable use policy says the same thing in more formal language.
Nothing here is legal or financial advice, and where an article touches on rights or reporting it says so.
Understanding the technology
Start here if you want to know what is actually happening in the audio. These five build on each other: what detection is, how the field does it, how cloning works from the other direction, what deepfake audio means as a phenomenon, and why your own ear is not a usable instrument.
What is AI voice detection?
The definition, the three verdicts, and why a synthetic flag carries far more weight than a clean result.
Read →How AI voice detection works
What a recording can reveal about its own origin, why perceptual realism stopped counting as evidence, and how compression destroys the answer.
Read →How voice cloning works
What a model takes from a minute of your speech, where the sample almost always comes from, and why a consent checkbox verifies nothing.
Read →Deepfake audio
The phenomenon rather than the product: where fakes actually turn up, and the denial a genuine recording now attracts.
Read →How to tell if a voice is AI
Which listening cues stopped working, which weakly survive, and why the ear is the wrong instrument for this job.
Read →Spotting and handling scams
The practical half. One overview, one consumer guide and one reference page on the terminology, then the individual patterns as they are actually encountered, then what to do once something has happened. Every one of them ends in the same place, which is a phone number you looked up yourself.
How voice scams are run
The field guide. How the trade is organised, which pretexts recur, and what the economics look like from the other side.
Read →Voice phishing: what to do
The consumer guide. What the call sounds like, what it is trying to obtain, and the callback rule that ends it.
Read →Vishing: the term and its history
The reference companion. Etymology, the phishing family of terms, and how security teams classify and simulate the technique.
Read →A relative rings in trouble
The call that comes with crying and a deadline, and the two phone calls that settle it.
Read →A caller says they have someone you love
The ransom call where nobody has been taken, and how to act when you are told not to hang up.
Read →The bank’s fraud team is on the line
Being told to move your money to a safe account, and why no bank has ever needed you to do that.
Read →The executive who rang to approve a payment
Why finance teams are the target, and how a written policy removes the burden of doubting a senior voice.
Read →The voice note is no longer proof
What changed when a recorded voice stopped being evidence that a person exists.
Read →The famous voice recommending an investment
Cloned endorsement at advertising scale, and why it keeps working.
Read →Identity fraud in remote interviews
Verifying a candidate without wronging the honest ones, which is the harder half of the problem.
Read →A recording you never made
Being on the wrong end of audio in a dispute, including the case where the recording is real and nobody believes it.
Read →Someone cloned my voice
For the person whose voice was taken. The honest list of what is available, and where recourse runs out.
Read →What to do after a voice scam
The first hour, in order. Preserve the file, ring the bank, report it, warn the person whose voice was used.
Read →Using Truthring
Two articles about getting a defensible answer out of a file, which is a narrower thing than people expect and depends more on how the recording was handled than on the analysis itself.
Getting a usable recording
Most clips arrive already damaged. How to capture and keep audio so the evidence is still in the file when it reaches a detector.
Read →Reading your result
What the verdict, the confidence figure and the attributed model each mean, and the four ways people routinely misread them.
Read →The evidence behind them lives outside this section. Methodology describes the pipeline; phone audio covers the hardest condition; accuracy is where both error directions are reported, including the false positive rate that most of this industry declines to publish; and limitations is the list of situations in which we expect to be wrong. There are also small tools that run in your browser, including a family safe-word generator and a printable callback card, and a page for checking a recording. Truthring has not launched, so quantitative results are forthcoming rather than published, and this section will say so until they are not.
About this section
Where should I start if something is happening right now?
Do not start with reading. End the call and ring the person or organisation back on a number you already had. Once that is done, the consumer guide to voice phishing and the guide to the first hour after a voice scam are the two pages written for that moment.
Do these articles teach you how to make a cloned voice?
No. We explain what the capability is, what it costs and what it implies, because that is what a journalist, a security lead or a worried parent needs. We do not publish steps, settings or system comparisons, and we do not publish anything whose main value would be helping someone evade a detector.
Is any of this legal or financial advice?
No. Where an article touches on rights, remedies or reporting obligations it says so and tells you to check your own position in your own jurisdiction rather than relying on ours.
How is this different from the research and methodology sections?
The learning centre explains the subject. The methodology, accuracy, limitations and benchmark pages document what Truthring specifically does, what it measures and where it fails. Explanations here, evidence there.
Most read
Evidence and method
Elsewhere on the site
Reviewed