Can AI voice detectors run on a live phone call, or only on recordings?
ok so this came up after a few of you on the forum were talking about family emergency scam calls. i make a podcast and i've played around with voice cloning for fun, so i know how convincing it can be, and it freaks me out a bit.
Cite
Global 100 Forum, "Can AI voice detectors run on a live phone call, or only on recordings?", https://global100.org/forum/q/can-ai-voice-detectors-run-on-a-live-phone-call-or-only-on-recordings/, accessed 2026-11-10.what i can't work out is how detection would even work in real life. all the voice detector demos i've seen are "upload a file, get a score." that's fine for a clip someone sent me. but a scam call is live. you pick up, someone talks, you've got maybe a minute before you decide whether to panic.
so is there anything that can analyse a call while it's happening? like an app on the phone, or something the carrier does? or are the detectors really only for recordings, after the fact?
and the second bit: if there are live ones, do they cope with phone line audio, which is squashed and noisy compared to a studio recording? i'd love to hear from anyone who's tried one, or who knows how the real products work, because the marketing pages all sound the same to me.
Write something first.
Give people something to work with: at least 30 words on what happened and what you tried.
That is too long. Keep it under 6,000 characters.
Write the question as the title, 15 to 140 characters, no links.
Pick a category.
Add a name (2 to 40 characters, no links).
That email address does not look right.
That was quick. Read the thread, then try again.
The form expired. Reload the page and post again.
Something went wrong with the form. Reload and try again.
Please complete the check and post again.
Limit reached for now. Try again later.
This thread is closed to new replies.
This thread no longer accepts replies.
Something went wrong with the form. Reload and try again.
Post a reply or question first (name and email), then this browser can vote, edit and accept answers.
You cannot vote on your own post.
Only the author (within 30 days) or the forum team can do that.
That email already has an account. Use your sign-in link instead.
New accounts are paused for the moment. Try again later.
That was already posted.
6 answers
- Hana SatoMember ·
Honest answer: some products claim to run in near real time, so it is technically possible to score audio as it streams. What I would doubt is how well it works on the call you are describing. Phone audio is heavily compressed, and detectors are known to struggle when audio is degraded. Add background noise and a short window of speech and the evidence available to the model is thin.
I would not build a safety plan around a live detector. I would build it around what you do when the call comes in.
- JulesMember ·
@Hana Sato that's what i was afraid of, but better to know. the call back rule is simple enough to teach my parents tonight. the voice detector accuracy thread has more detail for anyone else who wants it
- Sam WhitlockMember ·
Agree with Hana, with a threat model addendum. A live detector has to decide fast, under noisy conditions, and the attacker gets to try again. If I know a detector is on the line, I can iterate until I pass. Defences that only work when the attacker is not looking at them are weak.
Procedural defences do not care how good the clone is. A family code word, hang up and call back on a number you already have, never act on a payment request that arrives on the same call.
- Rob M.Member ·
I asked about this in another thread after my mother got one of these calls. The answer I was given then, and it stuck, was that you do not need a gadget, you need a rule. She now has a rule: if anyone asks for money on a call, she hangs up and rings that person on the number she already has. It cost nothing and it would have worked every time.
- Dan OkaforMember ·
I thought phone companies already flag AI calls automatically these days. Isn't that what the "scam likely" label is for? If so you do not need your own app.
- Ingrid V.Member ·
@Dan Okafor I would be careful there. A "scam likely" label is usually based on the calling number and the pattern of calls from it, such as reported numbers or high volume, not on analysing whether the voice is synthetic. A cloned voice calling from a fresh or spoofed number may not be labelled at all. So the label is worth noticing and worth nothing as proof either way.