David Bridges
After previewing its new Audiobox generative AI mission final month, which may replicate your voice utilizing simply seconds of pattern audio, Meta has now launched a publicly accessible demo of the method, which lets you create customized audio samples based mostly by yourself voice and/or textual content prompts.
As defined by Meta:
“Audiobox is Meta’s new basis analysis mannequin for audio technology. It may possibly generate voices and sound results utilizing a mix of voice inputs and pure language textual content prompts – making it straightforward to create customized audio for a variety of use instances.”
And it’s fairly good, although additionally fairly freaky.
First off, the Audiobox demo supplies entry to a variety of components, together with voice descriptions, sound impact technology, audio modifying, and extra.
The primary pattern use case is customized audio technology based mostly on textual content prompts
As you may see on this instance, Meta has offered a number of primary Audiobox check components, together with a text-to-speech course of that lets you generate customized audio based mostly on any textual content enter.
There are solely two system voices to make use of within the demo, “Alice” or “Emily”, nevertheless it offers you an concept of what the method can do, by way of translating customized textual content into various audio streams.
You may also add customized sounds to your pattern, once more based mostly on textual content prompts.
However the freaky half is the choice to generate your personal voice:
It undoubtedly gave the impression of me, and I undoubtedly didn’t prefer it. This, after all, might find yourself being a major concern, by way of potential misuse, and Meta does make you comply with a variety of phrases and situations of use earlier than attempting it.
Nevertheless it’s fairly good, fairly correct. Uncomfortably so.
It’s the most recent in Meta’s rising slate of generative AI instruments, that are set to see various use instances in its apps over the subsequent 12 months. And whereas Meta’s nonetheless constructing in safety parameters, it’s a concern that it’s making such instruments extra broadly accessible, even with the assorted phrases of service situations hooked up.
I’m unsure the world is prepared for it, in all honesty, and with varied elections developing, I can think about a variety of points arising from potential misuse.
However AI growth is shifting quick, and Meta needs to maintain up.
If you wish to see how good it’s at replicating your voice, you may try it out for your self right here. Name your accomplice and play the snippet down the road, see if they will decide that it’s not really you.
Maya Kensington
Cameron Steel
