Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Impressive.

But also enabling the next gen of "Mom, I'm in Mexican jail. Quickly wire me $2,000 so I can get out." scams.



And Black mirror's "would you like to speak with your dead husband again?"


Forget dead husbands, with this tech, it will be hard to trust anything a politician said. Basically, once they master adding this to video, ANYTHING could be construed against anyone.

Want a video of a politician saying "Hitler was right" to cheering masses? Want a video about a president saying it's time to start Nuclear War One?

You can make that.


We already couldn't trust anything a politician said! But indeed, this is the moral equivalent of how Photoshop has nearly undermined photographic evidence. That leaves video, and it's starting to succumb. Synthetic video AI politician, well, Robert Heinlein wrote _The Moon is a Harsh Mistress_ in 1966.


> Photoshop has nearly undermined photographic evidence

No it hasn't, not at all.


> No it hasn't, not at all.

What do you mean? it most certainly has.

It is much harder to look at a photograph now and say for sure it's real. Before photoshop, there were ways of manipulating photos, but they were much harder and did not yield nearly as good results. You can photoshop people into photos they were not originally in, doing things they have never done. Even Trump is having his hands enlarged in photos.

Maybe you're speaking specifically about "evidence" in the legal sense. I can't speak to that. But there are countless examples where you can't (and shouldn't) believe what you see, because it's not real.


Certainly it raises the bar for skepticism, but I can think of relatively few cases where a photoshopped image has driven a news cycle/key event because someone famous/important believed it to be real.

Most doctored images/videos get snuffed out by the media very quickly if they grow viral.


So because you can think of relatively few cases where key figures were duped, it therefore must not be a problem? That hypothesis fails to account for the rest of humanity.

I have a friend who believes in aliens. They showed me a video that someone had put together of "footage". They were seriously showing it to me as evidence, to convince me that they are real, based on this video. I was shocked they were so serious - and having a hard time containing my laughter.

I later found a number of other videos debunking the videos and clips that video was based off of - but my friend still believes in aliens, based on those videos. And this is someone I regularly have "intelligent" conversations with.

This change, the ability to put words in someones mouth, is the next photoshop. And it WILL have consequences to it, good and bad. And we are not prepared for that, as a society - we haven't gotten over photoshop yet.


From the "Ethics" section of the Lyrebird site:

"Voice recordings are currently considered as strong pieces of evidence in our societies and in particular in jurisdictions of many countries. Our technology questions the validity of such evidence as it allows to easily manipulate audio recordings. This could potentially have dangerous consequences such as misleading diplomats, fraud and more generally any other problem caused by stealing the identity of someone else.

By releasing our technology publicly and making it available to anyone, we want to ensure that there will be no such risks. We hope that everyone will soon be aware that such technology exists and that copying the voice of someone else is possible. More generally, we want to raise attention about the lack of evidence that audio recordings may represent in the near future."

I'm glad the authors addressed this issue pretty forthrightly, but part of me wishes they'd written a bit more about exactly your point. Whether or not recorded speech will continue to be legally binding evidence, I think it's just as important to point out that many people are normally quite happy to take what they hear as solid evidence, especially when it aligns with their prejudices.



Hi.

This is your first and only post, you have no submissions or favorites, and your account is 193 days old.

I'm very curious (and perplexed) as to why you have linked a video from elsewhere in this thread with no supporting context regarding its relevance other than "hf".


I can't wait to hear the new cassetteboy songs.


In the future: PGP signed speeches.


You joke, but I strongly suspect Authenticated Personal Trust is absolutely going to have become a thing.


And any genuine recording can be plausibly denied.


"Forget dead husbands, with this tech, it will be hard to trust anything a politician said."

Like we can trust anything a politician says/has said today anyway!


I meant it as you can't trust recording of what someone said. Basically any audio could be faked.

Which would be great for people like Trump. "Trust me, that audio is fake news. Fake!"


I was thinking the same when I saw an Adobe demo about a similar product. But then I realize that I have listen voice imitators mocking politicians very accurately. Maybe having this technology at hand will make everyone realise that we can't trust an audio recording.


The advantage of this method is that you can train on random speech datasets, and then using only a minute or two, find the "voice embedding" of a person, and generate anything with his/her voice. This voice embedding is much like the word2vec word vectors, and have similar arithmetic properties.

Responding to multiple sibling comments.


Wouldn't you need tons and tons of training data ? Obama, Clinton and Trump are public personalities, so it's easy to have many hours of recording of their voices.

A random relative; not so much.


Do you know the voices of your second or third degree relatives? Telephone scams are quite a well organized crime by now in Germany. If they can extend the target audience from elderly people to the general public by this means, they dont really care about some negative contacts.

http://www.aarp.org/money/scams-fraud/info-2016/how-to-beat-... https://de.wikipedia.org/wiki/Enkeltrick

If they really need a voice sample from a relative they could still impost, with voice or not, as insurance person, HR or police and acquire it by simple means if they have any idea about the social graph of their victim.


They claim to only need 1 minute of recording. In this age of all the kids sharing everything all the time that shouldn't be too hard to acquire.


Even better, targeted attacks against a person to collect their voice could involve contacting them for an opinion survey regarding a product, survey, or political opinion they value. Gleaning something like that from social media profiles is fairly easy.


"My voice is my passport. Verify me."


It claims that it only needs a minute of audio.


Or 1) planting evidence (no need for extended wire taps, 2) voice recognition software for access control, and 3) nearly any other fraud that will exploit the acceptance/authorization of verbal consent.


"But you just went out the door 10 minutes ago, and we live in France! How did it happen?! I'll wire it right away!"



First thing that came to my mind. We're assembling the pieces one at a time.


If this is anything like adobe's system then you need a fairly extensive library of the target speaking naturally (~20 hours), so to pull off a scam like that you'll want to try and get into alexa, siri, cortana or Google's library and find a heavy user.


Interesting that companies are trying to get us used to taking to our computers.


It states that it only needs about a minute of someones voice.


I wonder how long until we have an audio file of Paul Ryan saying something horrendous.


<lowhangingfruit>Just switch on CNN</lowhangingfruit>




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: