sramder@lemmy.world
on 10 Oct 2023 23:44
nextcollapse
Anyone know how many hours of training data it takes to build up a convincing model of someone’s voice? It was 10’s of hours when I did a bit of research a year ago… the article says social media is the likely source of training data for these scams, but that seems unlikely at this point.
treefrog@lemm.ee
on 10 Oct 2023 23:53
nextcollapse
I don’t remember the exact number but I did see an article recently that said it was videos on social media like you surmised.
And it was a pretty minimal amount of data needed. Definitely not tens of hours. Less than one hour iirc.
Rozz@lemmy.sdf.org
on 11 Oct 2023 00:57
nextcollapse
Is it safe to assume that if you don’t have any family that posts videos to Facebook/socials you are in a safer place?
treefrog@lemm.ee
on 11 Oct 2023 01:15
nextcollapse
I certainly am hoping so myself.
NeoNachtwaechter@lemmy.world
on 11 Oct 2023 03:25
nextcollapse
if you don’t have any family that posts videos to Facebook/socials you are in a safer place?
You are safe only if you don’t have any people at all whom you trust.
The technology has clearly come a long way in a short time, really fascinating.
I remember the first examples I read about being trained with celebrity read audiobooks because they needed so much audio data. I want to say Tom Hanks or Anthony Hopkins but I could have that confused with something else.
Even_Adder@lemmy.dbzer0.com
on 11 Oct 2023 00:25
nextcollapse
TorToiSe can work off of just three ten second clips when you’re using a pre-trained model. No telling if that’ll sound any good.
CrabLangEnjoyer@lemmy.world
on 11 Oct 2023 00:29
nextcollapse
A current state of the art ai model from Microsoft can achieve acceptable quality with about 3 seconds of audio. Commercially available stuff like eleven labs about 30 minutes. But quality will obviously vary heavily but then again they’re using a low quality phone call so maybe not that important
sramder@lemmy.world
on 11 Oct 2023 02:42
nextcollapse
That’s downright scary :-) I think it took longer in the last Mission Impossible.
30 minutes is still pretty minimal for the kind of targeted attack it sounds like this is used for. I suppose we all need to work with our families on code words or something.
I went in thinking the article was a bit alarmist, but that’s clearly not the case. Thank for the insight.
madsen@lemmy.world
on 11 Oct 2023 05:53
nextcollapse
With that little, they may be able to recreate the timbre of someone’s voice, but speech carries a multitude of other identifiers and idiosyncrasies that they’re unlikely to get with that little audio, like personal vocabulary (we don’t choose the same words and phrasings for things), specific pronunciations (e.g. “library” vs “libary”), voice inflections, etc. Obviously, the more training data you have, the better the output.
waylaidwanderer@lemmy.ca
on 11 Oct 2023 18:26
collapse
ElevenLabs only needs 1 minute, but it also works with even shorter clips.
DontMakeMoreBabies@kbin.social
on 11 Oct 2023 00:44
nextcollapse
I literally just cloned someone's voice for a presentation on AI and did it using maybe 30 total minutes of audio....
Took me about an hour and it was free. Hardest part was clipping the audio to get the 'good bits.'
The voice was absolutely convincing.
sramder@lemmy.world
on 11 Oct 2023 02:32
nextcollapse
Wow! That’s really impressive.
theodewere@kbin.social
on 11 Oct 2023 03:29
collapse
it's no wonder actors are taking an interest, given the level of tech Disney and everybody else must have access to
Someone else mentioned that Microsoft has one capable of working with far less material.
But 30 minutes is definitely short enough to make this sort of scam/attack feasible in my mind.
Drusas@kbin.social
on 11 Oct 2023 00:46
nextcollapse
As someone who has an uncanny ability to recognize voices, I'm skeptical about how good these really are. Of course, most people don't share that ability.
Meanwhile, I could probably be fooled by a picture.
Rozz@lemmy.sdf.org
on 11 Oct 2023 00:50
nextcollapse
Yeah, does a familiar voice mean a famous person or personal friend?
PilferJynx@lemmy.world
on 11 Oct 2023 02:53
nextcollapse
Hmm, I understand your sentiment, but how would you know. Of course you’d pick out the bad dupes but this technology is getting really good that I fear it would go unnoticed, especially if they keep the detectable ones to reinforce bias
I don’t have examples but having listened to some samples of various Ai generated clones (the one paper had samples of I believe 10s, 30s, 1min, 5 min) and all of them progressively sounded better. The 10 second one basically sounded like a voice call whose bit rate dropped out mid word. And the voice so long as you used words that were similar in phoenix sounded pretty close. Although this is just my experience, but to you it might sound pretty bad while to me it sounded pretty reasonable if under bad audio conditions.
This is the main one I’ve seen examples of. You’ll have to find the samples yourself, I believe it was in the actual paper?
cypherpunks@lemmy.ml
on 11 Oct 2023 18:29
collapse
That code was state of the art (for free software) when the author first published it with his master’s thesis four years ago, but it hasn’t improved a lot since then and I wouldn’t recommend it today. See the Heads Up section of the readme. Coqui (a free software Mozilla spinoff) is better but also is sadly still nowhere near as convincing as the proprietary stuff.
Wait it’s been 4 years? Time really goes by. Yeah with most Ai things I assumed those with more time and resources would create better models. OS Ai is at a great disadvantage when it comes to data set size and compute power.
just_another_person@lemmy.world
on 11 Oct 2023 00:48
nextcollapse
Whomever is stupid enough to think that Tom Hanks is calling you personally probably needs a court appointed guardian.
TheFriar@lemm.ee
on 11 Oct 2023 01:19
nextcollapse
Did you read the article? It’s talking about taking kids voices from TikTok and shit. Social media. People have been posting videos of themselves talking for years. That’s enough data to train an ai to leave a message saying, “mom, I lost my phone and I’m in trouble. I need some money.” Or something of that sort. It’s been happening for a long time. This is only making it more confincing
just_another_person@lemmy.world
on 11 Oct 2023 03:07
nextcollapse
The reference of the entire article is talking about scammers using AI models of voice you know and understand. None of these scam rings have the time to break it down to your family.
You sure? It’s very easy for these scammers to make a bot to trawl those “address/people lookup” sites, get family names and numbers, and then search for anyone in there’s public social media, and compile that footage. It wouldn’t be much work at all after creating the bot. Those creepy people lookup sites list an absurd amount of information. It would make doing this very easy. And think of how much work already goes into scams that use sheer numbers to boost likelihood of working with a basic ruse. If they can trim that list of available phone numbers down to—even if it were just 30%, or 15% of available phone numbers now with personal information and an in by imitating someone they know and love? That’s still a fuck load of people. And the likelihood of success would shoot WAY up while actually cutting down on the amount of work they’d need to do. So I’d argue you have that backwards.
PlantJam@lemmy.world
on 11 Oct 2023 13:20
nextcollapse
enough data
To be clear, about three seconds of your voice is “enough”.
Same. I managed to stay off of social media, and I was the prime age for it at every turn. MySpace came around when I was in middle school/early high school. Facebook was opened up to everyone in late high school. Instagram came around when I was in college—and when I was traveling. I’m so glad I was that super annoying kid calling everything a conspiracy to steal my likeness/steal my data…who knew my need to be a contrarian as an anarchist teen would be so helpful?
I mean…I also grew up into an anarchist adult. So I just got lucky that I found the right books and music to push me in that direction young.
ChickenAndRice@sh.itjust.works
on 12 Oct 2023 04:19
collapse
A lot of crimethinc., Emma Goldman, and adbusters in high school (adbusters isn’t a book, but it was still deep in my repertoire). From there, Hannah arendt, Chomsky, etc. in late high school/college. I also listened to a lot of anti-flag, against me!, propagandhi, strike anywhere…all of my media was very anarchist/anti govt/anti capitalist. I stood no chance lol.
And as someone who was young enough to feel angry (and justifiably so…bush/Cheney and the patriot act were all happening. I had plenty reason to be wary of spying), admittedly I was following these things and knew what was happening, but I was still just a contrarian at heart, I could yell and argue with my parents friends, but I probably sounded like an ass. I didn’t fully know how to hold these beliefs. They were more knee jerk reactions fueled by hormones and an insane set of circumstances in the world. A lot of my embarrassing memories that come to me randomly when I’m trying to fall asleep have to do with being up in arms about something I wasn’t really qualified to speak on lol
I’m sure I was more annoying than I was inspiring
MargotRobbie@lemmy.world
on 11 Oct 2023 02:20
collapse
Unless you actually know Tom Hanks personally and are expecting a call from him, of course.
TheFriar@lemm.ee
on 11 Oct 2023 01:20
nextcollapse
The Industrial Revolution and its consequences have been a disaster for the human race.
d4rknusw1ld@lemmy.world
on 11 Oct 2023 02:04
nextcollapse
BADONK!
flossdaily@lemmy.world
on 11 Oct 2023 02:17
nextcollapse
I cloned my own voice to prank a friend, and… Wow, it was a gut-dropping moment when I understood just how dangerous this tool is for precisely this type of scam.
It’s one thing to hear about it, but to actual experience it… Terrifying.
qooqie@lemmy.world
on 11 Oct 2023 12:34
nextcollapse
Mind sharing more info about the prank? Sounds like an interesting story
flossdaily@lemmy.world
on 11 Oct 2023 12:52
collapse
Oh, it was nothing more than just showing off the technology, really. It wasn’t a committed bit.
I cloned my voice then left a voicemail that said something like: “hey buddy it’s me. My car broke down and I’m at… Actually I don’t know where I’m at. I walked to the gas station and borrowed this guy’s phone. He said he’ll give me a ride into to town if I can get him $50 bucks. Could you venmo it to him at @franks_diner? I’ll get you back as soon as I can find my phone. … By the way this is really me, definitely not a bot pretending to be me.”
Catfish@lemmygrad.ml
on 11 Oct 2023 18:26
collapse
Yeah but they’ll call your family. A friend of mine was recently affected by this, a scammer had a clone of her voice asking for around $300 to fix their car because they got stranded in the middle of nowhere. So they call up your parents and to your mom it’s like “Oh no! My baby! Of course I’ll help you!” and your mom gives them $300 thinking it’s you.
Yeah my family knows better. I don’t call anyone either plus I’ve got all of my family on DEFCON 1 when it comes to asking for money. Had someone try and scam my mom via Facebook pretending to be my sister. I have family members contacting me ALL the time with issues with their stuff so they don’t trust anything at all.
This all stems from myself getting scammed nearly 20 years ago via email so I’ve educated everyone immensely.
threaded - newest
Anyone know how many hours of training data it takes to build up a convincing model of someone’s voice? It was 10’s of hours when I did a bit of research a year ago… the article says social media is the likely source of training data for these scams, but that seems unlikely at this point.
I don’t remember the exact number but I did see an article recently that said it was videos on social media like you surmised.
And it was a pretty minimal amount of data needed. Definitely not tens of hours. Less than one hour iirc.
Is it safe to assume that if you don’t have any family that posts videos to Facebook/socials you are in a safer place?
I certainly am hoping so myself.
You are safe only if you don’t have any people at all whom you trust.
But then you are having some other problems…
Make sure to use a fake accent when talking to strangers on the phone
The technology has clearly come a long way in a short time, really fascinating.
I remember the first examples I read about being trained with celebrity read audiobooks because they needed so much audio data. I want to say Tom Hanks or Anthony Hopkins but I could have that confused with something else.
TorToiSe can work off of just three ten second clips when you’re using a pre-trained model. No telling if that’ll sound any good.
I’ll have to check that out, thanks for the link.
A current state of the art ai model from Microsoft can achieve acceptable quality with about 3 seconds of audio. Commercially available stuff like eleven labs about 30 minutes. But quality will obviously vary heavily but then again they’re using a low quality phone call so maybe not that important
That’s downright scary :-) I think it took longer in the last Mission Impossible.
30 minutes is still pretty minimal for the kind of targeted attack it sounds like this is used for. I suppose we all need to work with our families on code words or something.
I went in thinking the article was a bit alarmist, but that’s clearly not the case. Thank for the insight.
With that little, they may be able to recreate the timbre of someone’s voice, but speech carries a multitude of other identifiers and idiosyncrasies that they’re unlikely to get with that little audio, like personal vocabulary (we don’t choose the same words and phrasings for things), specific pronunciations (e.g. “library” vs “libary”), voice inflections, etc. Obviously, the more training data you have, the better the output.
ElevenLabs only needs 1 minute, but it also works with even shorter clips.
I literally just cloned someone's voice for a presentation on AI and did it using maybe 30 total minutes of audio....
Took me about an hour and it was free. Hardest part was clipping the audio to get the 'good bits.'
The voice was absolutely convincing.
Wow! That’s really impressive.
it's no wonder actors are taking an interest, given the level of tech Disney and everybody else must have access to
The most advanced Model I know just needs half an hour of your voice or sth.
Someone else mentioned that Microsoft has one capable of working with far less material.
But 30 minutes is definitely short enough to make this sort of scam/attack feasible in my mind.
As someone who has an uncanny ability to recognize voices, I'm skeptical about how good these really are. Of course, most people don't share that ability.
Meanwhile, I could probably be fooled by a picture.
Yeah, does a familiar voice mean a famous person or personal friend?
For me, it could be either. Some of us recognize people by their voices more than by their faces.
I always thought being able to recognize voices is a common skill? Is it not?
Very much not, in my experience.
Hmm, I understand your sentiment, but how would you know. Of course you’d pick out the bad dupes but this technology is getting really good that I fear it would go unnoticed, especially if they keep the detectable ones to reinforce bias
I don’t have examples but having listened to some samples of various Ai generated clones (the one paper had samples of I believe 10s, 30s, 1min, 5 min) and all of them progressively sounded better. The 10 second one basically sounded like a voice call whose bit rate dropped out mid word. And the voice so long as you used words that were similar in phoenix sounded pretty close. Although this is just my experience, but to you it might sound pretty bad while to me it sounded pretty reasonable if under bad audio conditions.
github.com/CorentinJ/Real-Time-Voice-Cloning
This is the main one I’ve seen examples of. You’ll have to find the samples yourself, I believe it was in the actual paper?
That code was state of the art (for free software) when the author first published it with his master’s thesis four years ago, but it hasn’t improved a lot since then and I wouldn’t recommend it today. See the Heads Up section of the readme. Coqui (a free software Mozilla spinoff) is better but also is sadly still nowhere near as convincing as the proprietary stuff.
Wait it’s been 4 years? Time really goes by. Yeah with most Ai things I assumed those with more time and resources would create better models. OS Ai is at a great disadvantage when it comes to data set size and compute power.
Whomever is stupid enough to think that Tom Hanks is calling you personally probably needs a court appointed guardian.
Did you read the article? It’s talking about taking kids voices from TikTok and shit. Social media. People have been posting videos of themselves talking for years. That’s enough data to train an ai to leave a message saying, “mom, I lost my phone and I’m in trouble. I need some money.” Or something of that sort. It’s been happening for a long time. This is only making it more confincing
The reference of the entire article is talking about scammers using AI models of voice you know and understand. None of these scam rings have the time to break it down to your family.
You sure? It’s very easy for these scammers to make a bot to trawl those “address/people lookup” sites, get family names and numbers, and then search for anyone in there’s public social media, and compile that footage. It wouldn’t be much work at all after creating the bot. Those creepy people lookup sites list an absurd amount of information. It would make doing this very easy. And think of how much work already goes into scams that use sheer numbers to boost likelihood of working with a basic ruse. If they can trim that list of available phone numbers down to—even if it were just 30%, or 15% of available phone numbers now with personal information and an in by imitating someone they know and love? That’s still a fuck load of people. And the likelihood of success would shoot WAY up while actually cutting down on the amount of work they’d need to do. So I’d argue you have that backwards.
To be clear, about three seconds of your voice is “enough”.
I’m so fucking glad that I’ve hardly ever had my voice and likeness posted publicly on the internet
Same. I managed to stay off of social media, and I was the prime age for it at every turn. MySpace came around when I was in middle school/early high school. Facebook was opened up to everyone in late high school. Instagram came around when I was in college—and when I was traveling. I’m so glad I was that super annoying kid calling everything a conspiracy to steal my likeness/steal my data…who knew my need to be a contrarian as an anarchist teen would be so helpful?
I mean…I also grew up into an anarchist adult. So I just got lucky that I found the right books and music to push me in that direction young.
What books did you read?
A lot of crimethinc., Emma Goldman, and adbusters in high school (adbusters isn’t a book, but it was still deep in my repertoire). From there, Hannah arendt, Chomsky, etc. in late high school/college. I also listened to a lot of anti-flag, against me!, propagandhi, strike anywhere…all of my media was very anarchist/anti govt/anti capitalist. I stood no chance lol.
And as someone who was young enough to feel angry (and justifiably so…bush/Cheney and the patriot act were all happening. I had plenty reason to be wary of spying), admittedly I was following these things and knew what was happening, but I was still just a contrarian at heart, I could yell and argue with my parents friends, but I probably sounded like an ass. I didn’t fully know how to hold these beliefs. They were more knee jerk reactions fueled by hormones and an insane set of circumstances in the world. A lot of my embarrassing memories that come to me randomly when I’m trying to fall asleep have to do with being up in arms about something I wasn’t really qualified to speak on lol
I’m sure I was more annoying than I was inspiring
Unless you actually know Tom Hanks personally and are expecting a call from him, of course.
The Industrial Revolution and its consequences have been a disaster for the human race.
BADONK!
I cloned my own voice to prank a friend, and… Wow, it was a gut-dropping moment when I understood just how dangerous this tool is for precisely this type of scam.
It’s one thing to hear about it, but to actual experience it… Terrifying.
Mind sharing more info about the prank? Sounds like an interesting story
Oh, it was nothing more than just showing off the technology, really. It wasn’t a committed bit.
I cloned my voice then left a voicemail that said something like: “hey buddy it’s me. My car broke down and I’m at… Actually I don’t know where I’m at. I walked to the gas station and borrowed this guy’s phone. He said he’ll give me a ride into to town if I can get him $50 bucks. Could you venmo it to him at @franks_diner? I’ll get you back as soon as I can find my phone. … By the way this is really me, definitely not a bot pretending to be me.”
We neeeed a guide 👋
Check out ElevenLabs.
Do you guys remember when the T-1000 did this?
What’s wrong with Wolfie? I can hear him barking…
Your parents are dead
In Terminator 1 the T-800 made a scam call to Sarah in order to find out where she is. He deepfaked the voice of Sarah’s mother, and she fell for it.
Good luck criminals. I ignore nearly every call.
Yeah but they’ll call your family. A friend of mine was recently affected by this, a scammer had a clone of her voice asking for around $300 to fix their car because they got stranded in the middle of nowhere. So they call up your parents and to your mom it’s like “Oh no! My baby! Of course I’ll help you!” and your mom gives them $300 thinking it’s you.
Yeah my family knows better. I don’t call anyone either plus I’ve got all of my family on DEFCON 1 when it comes to asking for money. Had someone try and scam my mom via Facebook pretending to be my sister. I have family members contacting me ALL the time with issues with their stuff so they don’t trust anything at all.
This all stems from myself getting scammed nearly 20 years ago via email so I’ve educated everyone immensely.