Demucs for audio cleanup

[Tex]

I just realized Demucs (I assume any variation) is extremely good at separating background noise/wind from speech.

I took some video outside when it was windy which had horrible windnouise and rumbling.

I dropped the MP4 video file into x-minus.pro and it totally separated the noise from the voice.

[Kwai Chang]

I just realized Demucs (I assume any variation) is extremely good at separating background noise/wind from speech.

I took some video outside when it was windy which had horrible windnouise and rumbling.

I dropped the MP4 video file into x-minus.pro and it totally separated the noise from the voice.

That's called a Hope Detector(sniffer!)
Will it be any match for the Fairchild 660? It might be!
The problem is Americans won't buy products in generic packaging(without the WIND!).
So, do we really want to end the conversation?
Thank you, for another objective-specific contribution. I am still playing your Baby You're A Rich Man and It's All Too Much!
Keep up the nice work!
KC

[Lord Reith]

Yes I am surprised there has not been more research devoted to this. On mvsep there is a variation of UVR which separates audio into music and hiss but it works only partially. Promising though. There's also a varaiation for echo and reverb which is not too bad but likewise still not completely there.

[Tex]

What they also need is an algorithm dedicated to simply separating the left, middle and right channel information.

[Rave O Lucien]

I just realized Demucs (I assume any variation) is extremely good at separating background noise/wind from speech.

I took some video outside when it was windy which had horrible windnouise and rumbling.

I dropped the MP4 video file into x-minus.pro and it totally separated the noise from the voice.

Do you suppose it could remove the screams from the Hollywood Bowl concerts?

[Lord Reith]

What they also need is an algorithm dedicated to simply separating the left, middle and right channel information.

You can do that with audacity.

[csnyfan]

What they also need is an algorithm dedicated to simply separating the left, middle and right channel information.

You can do that with audacity.

Yeah. But now imagine if AI could bypass the uncertainty principle there and give us crisper center extractions. With standard tools we are stuck with a tradeoff between frequency or time precision - one or the other is always smeared, so especially center isolations sound shitty. That would be a game changer.

[Lord Reith]

What they also need is an algorithm dedicated to simply separating the left, middle and right channel information.

You can do that with audacity.

Yeah. But now imagine if AI could bypass the uncertainty principle there and give us crisper center extractions. With standard tools we are stuck with a tradeoff between frequency or time precision - one or the other is always smeared, so especially center isolations sound shitty. That would be a game changer.

This is true. Still an amazing tool though.

Seamlessly removing hiss would be the big one for me. You can very nicely do this with speech but not music yet.

[alphabeatles]

A website to search for AI apps for Music and other tasks:

hxxps://opentools.ai/