I needed until "Everybody’s talking about it" to hear it in his voice :)

Please no spoilers!

It does have actual genuine uses. I'm in the process of recording a series of tutorials for my peers but I'd like them to hear things in my voice so it doesn't sound like I have offloaded the work to someone else.

I don't know if this helps or harms the credibility but I can't really talk more than an hour without seriously straining my voice. So cloning it sounds like a great use-case for someone with a similar problem.

Looking forward to trying this.

It's useful for some things, like satire. Presidents Play is a good series in YouTube where it uses US presidents' cloned voices for comedic satire.

proudly trumpeting the ability to co-opt other people's identities without their permission

EXACTLY. Clone the wrong person's voice and it's game over.

It's really easy for a technical person to do as well.

I use Coqui TTS[0] as part of my home automation, I wrote a small python script that lets me upload a voice clip for it to clone after I got the idea from HeyWillow[1], and a small shim that lets me send the output to a Home Assistant media player instead of using their standard output device. I run the TTS container on a VM with a Tesla P4 (~£100 to buy) and get about 1x-2x (roughly the same time it'd take to say it, to process) using the large model.

Just for a giggle, I uploaded a few 3s-5s second clip of myself speaking and cloned my voice, then executed a command to our living room media player to call my wife into the room; from another room, she was 100% convinced it was myself speaking words I'd never spoken.

I tried playing with a variety of sentences for a few hours and overall, it sounded almost exactly like me, to me, with the exception of some "attitude" and "intonation" I know I wouldn't use in my speech. I didn't notice much of an improvement using much longer clips; the short ones were "good enough".

Tangentially, it really bugs me that most phone providers in the UK insist you record a "personal greeting" now before they'll let you check your voice mail box, I just record silence, because the last thing I want/need is a voicemail greeting in my voice confirming to some randomer I didn't want calling me, who I am and that my number is active, even more so knowing how I can clone any voice to a reasonably good accuracy with just a few seconds of audio.

[0] https://github.com/coqui-ai/TTS [1] https://heywillow.io/

Technically, wouldn't a simple "Hold on, I'll call you back" test call stop that?

I'd be interested as well. This is where I imagine the space is going - particularly as the potential for litigation increases around cloning.

Game studios will spin up a bunch of unique virtual voices for all the dialogue of extras. It'll probably be longer before we see replacements of main characters though. There's been some research in speech-to-speech transference as well - this means that company employee A records the character B's line with the appropriate emotional nuance (angry, sad, etc.) and the emotional aspect is copied on top of the generated TTS.

Have you tried eleven labs? I used that. Had to record 3 hours of training audio reading books and and news articles. But the result was really good.

StyleTTSv2 is pretty good and open source, you can easily traverse its latent space for voice

I’ve used tortoise tts before and trained it on my voice and a mix of voices. It’s not perfect but still impressive.

Yeah they made an easy to use frontend. Don't be the dropbox guy

Yeah I also noticed the install instructions is run this batch file that gets administrator access and starts downloading things…

It’s weird to me that people look at a technology and then assume from their reckoning that they know all the uses for that technology immediately. Most technological progress happens because someone notices a creative use for something that already exists which nobody else has noticed.

I'm absolutely using celebrity voices for my Home Assistant voice. Amazon has spent the last couple years removing the voices for Alexa that people had paid for.

To be fair, they’ve got pretty serious potential for letting tech companies get paid for a seasoned voice actor’s unique delivery, tone, inflection, etc rather than the voice actor themselves.

They are pretty good for leaving messages for my blind friend. I generally find calling / voice texts a waste of time (I type and read far faster than I talk or listen, not to mention the ability to reread etc), but my blind friend prefers getting voice messages when on his phone and this works for us. I type and send and when he comes back with something, Whisper makes it into text for me.

I like tools like these cause they make zero trust default even more obvious, and their "pretty limited use" is saving people hours of work.

Gen AI space to everyone else: “Your computer scientists were so preoccupied with whether or not they should, they didn’t stop to think if they could just do it anyway”

Bulldozing grandma is just the cost of technological progress /s

How many victims will it take for lawmakers to do something about this?

Quit being a doomer or keep it to yourself. This reminds me of the sound boards that were popular in the early 2000s except way more versatile. Some things are just good for people to have fun and THAT'S OKAY.

I haven’t looked at the code, but can you just patch out the 30 minute limit?

It’s a bit of a hassle, but after closing the Windows command, you can restart the program and use it indefinitely. The results you worked on will still remain in the workspace folder.

Yeah, felt like it positions itself as open source project here and on GitHub, but buries the cost in other pages... Doesn't even say the subscription cost anywhere I could find (in English). Not a huge fan of this advertising model.

Do you mean Inflection's Pi?

I'd claim the way most people imagine it being used for scamming, cold-calls impersonating someone the victim knows, doesn't really end up working out in practice because scam operations dial numbers at a huge scale expecting most not to pick up a "scam likely" call (or be away, or a dead number, etc.). Having to find a voice clip prior to each unanswered call would tank the quantity they're able to make.

For spear-phishing (impersonate CEO, tell assistant to transfer money) it's more feasible, but I hope it forces acceptance that "somebody sounds like X over the phone" is not and has never been a good verification method - people have been falling for scams like those fake ransom calls[0] for decades.

Not that there aren't potential harms, but I think they're outweighed by positive applications. Those uncomfortable with their natural voice, such as transgender people, can communicate closer to how they wish to be perceived - or someone whose voice has been impaired (whether just a temporary cold or a permanent disorder/illness/accident) can use it from previous recordings. Privacy benefits from being able to communicate online or record videos without revealing your real voice, which I think is why many (myself included) currently resort to text-only. There's huge potential in the translation and vocal isolation aspects aiding communication - feels to me as though we're heading towards creating our own babelfish. There's also a bunch of creative applications - doing character voices for a D&D session or audiobook, memes/satire, and likely new forms of interactive media (customised movies, audio dramas where the listener plays a role, videogame NPCs that react with more than just prereccorded lines, etc.)

[0]: https://www.fbi.gov/news/stories/virtual-kidnapping

You do realise this is not the first AI release to clone voices?

There is nothing in here that precludes you from running this on any OS that supports python + CUDA. They use miniconda for installation of python and python packages, but this could just as easily be a venv + system CUDA install or even better: a container. This is only one tiny Dockerfile away from running anywhere.

"If, by whiskey...."