Live data from Hacker News

Creates hyper-realistic voice clones from just 3 seconds of audio

anyvoice.net

21–30 of 48 posts

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#21

Has anyone tried multiple iterations? That is, upload a real voice, get its synthesized version, upload synthesized version 1 to get synthesized version 2, rinse and repeat...

Perhaps Alvin Lucifer reading his "I am sitting in a room" text would be ideal.

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#23

This was a great way for them to collect a lot of free voice data to train their model.

Every time there's a voice recognition post here someone comments about acquiring data. Why is this method better than having access to all of the video and podcasts sites on the internet?

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#24
post #23

This was a great way for them to collect a lot of free voice data to train their model.

Every time there's a voice recognition post here someone comments about acquiring data. Why is this method better than having access to all of the video and podcasts sites on the internet?

You can get people to utter the same sentence.

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#27
post #4

Sure, just let me submit my voice for cloning to a closed sourced, online service of unknown provenance. What could ever go wrong?

That's why you submit politician's voice instead

It would be fun to have a clone of Majel Barrett's voice for something like Siri or Alexa.

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#28
We've been advertising to get someone to take over the lease on a commercial building. Surprisingly, we've had several of what seem like very obvious scam attempts - people stringing us along, not trying to bargain (we are in a haggling country, people always try to bargain), asking us to wait unreasonable amounts of time, and finally when pressed breaking down into logical inconsistencies. So, not even good scam attempts.

I was wondering, what's the point? I mean, it's a building. You pay money, you sign the lease (in person), you get the use of the building. No money, no building. Where's the scam opportunity?

The only thing I can think of is that they're trying to get enough data and personal info to clone our voices and use that to try and gain access to bank accounts or to scam our relatives. Even if I'm wrong in this case, this seems like a major new vulnerability in society. I mean, if someone who sounded (and with video AI, perhaps even looked) exactly like me called up my mother and pretended I'd been violently robbed or had an accident, she'd transfer money in a heartbeat.

I'm considering that I should set up some kind of code system with my family for this. As in, if I ever end up in a situation where I need help, I'll use a particular code phrase. If I don't know it, assume it's an AI clone.

Re: Creates hyper-realistic voice clones from just 3 seconds of audio

#29

We've been advertising to get someone to take over the lease on a commercial building. Surprisingly, we've had several of what seem like very obvious scam attempts - people stringing us along, not trying to bargain (we are in a haggling country, people always try to bargain), asking us to wait unreasonable amounts of time, and finally when pressed breaking down into logical inconsistencies. So, not even good scam att…

> I'm considering that I should set up some kind of code system with my family for this. As in, if I ever end up in a situation where I need help, I'll use a particular code phrase.

You absolutely should. And include something for videos (like FaceTime calls). Especially if these members of your family are boomers+

Post reply on HN