🔉 Introducing SAM Audio, the first unified model that isolates any sound from complex audio mixtures using text, visual, or span prompts. We’re sharing SAM Audio with the community, along with a perception encoder model, benchmarks and research papers, to empower others to explore new forms of expression and... build applications that were previously out of reach. 🔗 Learn more:show more

SAM Audio represents a significant advancement in audio separation technology, outperforming previous models across a wide range of benchmarks and tasks.

Discover what’s possible with SAM Audio, SAM 3D, and SAM 3 in the Segment Anything Playground:

please don't ask us why we're so good at extracting voice from noisy audio clips

This is wild for creators, suddenly you can just pull the sound you want without fighting the mix.

Bringing ideas to life should be more intuitive🔈

Meta really said segment *anything*

open source?

Sure is, find all the details here:

So you could integrate this in your goggles and just make me hear what i'm looking at?

it's time for dingband @yacineMTB

>"Creators, musicians, audio engineers, and tinkers" Awwe. That's really sweet. But don't forget about these customers too: "National Security Agency", "Law Enforcement Agents", and "PsyOps Deep Fake specialists". They need stuff like this too.

Glad to see Meta is not tempted with other competitors and focusing on some different and good area.

Sounds like a great listener

We live in the craziest timeline I can’t believe all these companies release this stuff OSS for free wtf @finkd thank you, seriously. This (the fact that you altruistically release OSS AI Models for the world) is your best achievement imo

And this, ladies and gentlemen, is how the CIA releases 20 year old technology to the general public :)

Mark is trolling Sam Altman 😂

Tried 4 different files and failed to isolate. Nice start.

Was this to be funny and name Meta’s audio AI after @sama ?

This is fantastic, and we get all this for free. Thank you so much! 🙏

Audio has been waiting for a 'moment' like this. The ability to clean up crazy mixtures with just a prompt is a big win for everyone.

Ok this is really impressive

This is a massive milestone for audio AI! 🔉 To support the community in exploring SAM Audio, I’ve built an open-source GUI called AudioGhost AI. Since the original model can be heavy on VRAM, I implemented a "Lite Mode" that optimizes the pipeline for consumer GPUs (reducing requirements from 30GB+ to ~6GB-10GB VRAM). Also included a 1-click installer to fix the Windows dependency hurdles. Let’s make this incredible research accessible to every creator!

this is ready for karaoke apps 🎤

Whatever is happening in the LLMs domain at Meta, SAM team has been consistently delivering. Congratulations. Give SAM team a raise and maybe put them in leadership positions of Meta AI.

I wonder how well this handles speaker diarization

Wow, I was blown away by SAM3 for videos already, this is next level - can't wait to play with it

bravo

Brilliant.

Hyperagents for the win

@peace_node you seeing this shit?

This is a great model and so far so good on some early test.

Huge help for independent filmmaking & shooting on location in general 🔥

"Failed to separate audio."

fuckuuu

Very promising @tomislav_rupic @YakkStack

Is there an api endpoint to the playground? Or would I have to self host for that kind of interoperability ?

The playground provides a quick preview of what SAM can do, you’ll have to host for API level interoperability.

Fresh out of GPU resources. Got any clusters to spare?