I struggle to find non-evil applications of voice-cloning. Maybe listening to your dead relative's voice one more time? But those use-cases seems so niche to the overwhelming use this will likely have: misinformation, scamming, putting voice actors out of work.
I would clone my own and do things like create scripted tutorials/presentations and audio books.
I do not personally prefer it, but a non-trivial number of individuals like video/audio presentations over writing.
I’m currently recording my books into audiobooks the old fashioned way. I wonder good indistinguishable this would be.
Video game mods are the first use case that comes to mind for me. If you want to add new voice lines for a character, your options are:
1. don't (keep it silent);
2. recruit the original VA somehow;
3. recruit an alternative VA (who hopefully sounds close enough to the original to not be jarring);
4. splice voice lines together from existing voice lines; or
5. use text-to-speech.
Voice cloning is just Option 5 but with results much closer to Options 2 and 3.
Parody, but it toes the line of evil in being fake.
The possibility to continue to sound like yourself after permanently losing your voice (e.g. from motor neurone syndrome) is one. Perhaps almost the only one.
At my workplace, a colleague in another team used an AI tool to voice/video clone my companies CEO, CRO and CTO (I assume with their permission) and created a mandatory 30 minute training video that they expected us to watch with these monotone fake company leaders doing the presentation. It wasn't even a joke.
Voice acting for procedurally generated games?
They said non-evil.
Selling a voice profile for procedural/generated voice acting (similar to elevenlabs "voices") of a well-known person or pleasant sounding voice could be a legitimate use-case. But only iif actual consent is acquired first.
Given that rights about ones likeness (Personality rights) are somewhat defined there might be a legitimate usecase here. For example, a user might prefer a TTS with the voice of a familiar presenter from TV over a generic voice.
But it sounds exceedingly easy to abuse (similar to other generative AI applications) in order to exploit end-users (social engineering) and voice "providers" (exploitation of personality rights).
Eleven Labs pays the estate of the people's voices they use, correct?
I have their app on my phone and it will read articles in Burt Reynold's voice, Maya Angelou's voice & etc. I'm under the impression that they consented to this and their estate's are being compensated (hopefully).
If you want to make a podcast but don’t want to spend the time reading out the script and then painstakingly audio engineer every part of it.
Why would you want to make a podcast then? You don't need to offer a sub-optimal product if you don't want to make it.
I like writing out the scripts and doing the research. I don’t enjoy cleaning up the audio.
Black mirror episode
In Descript it is used to patch words in videos.
Last year I proudly said it was "two thousand and five" during a video take, and didn't notice it at the time. I was able to add the "twenty" using Descript.
Anyone Can Clone Your Voice Now
A good reason to let any caller ID one does not recognize go to voice mail. Phones need an app that does voice-to-text-to-voice to prevent capturing the voice. I want mine to sound like Smeagol (LoTR).
I'm pretty sure Caller ID is forgable and would lead to false negatives.
Then the only winning move is not to play. Everyone goes to voice mail and or all communication is in person. Perhaps these fondle slabs were indeed a fad.
Or just not use PSTN.
Yes, but can they be as annoying as I am? I think not.
They couldn't already do that? Or is this new Qwen model just that much significantly better?
It is significantly better.
Is there a demonstration or a free way to try it without having to pay for a single test or anything out there?
Official demos:
https://huggingface.co/spaces/Qwen/Qwen3-TTS
https://modelscope.cn/studios/Qwen/Qwen3-TTS
You can even do it on your PC. I did it via ComfyUI and this Plugin:
Thanks! I can't get it to speak in an African voice. Is this only possible through cloning?
Thank you, I'll have to give it a try.
Yes -- you can run inference on HF directly [0]. If you want to hear the demos that Simon Willison recorded, he has a nice blog post about it. [1]
Thank you!
This looks super interesting. Thanks for sharing.