Removing noise from sample

[Alandalton]

Hi,

Would anybody have a technique for removing background noise from a sample?

Thanks

[madwurmz]

there is this handy thing in Audacity (free wav editor) , selecting noise to make a noise profile and tweak the impact. https://filmstro.com/blog/how-to-remove ... n-audacity

but you can also just find and filter the noise with graphic eq like synapse GQ-7 or
use Softtube dyna-mite noise gate patches.
or just leave it in and call it Lo-Fi

[ideaB2]

[Alandalton]

there is this handy thing in Audacity (free wav editor) , selecting noise to make a noise profile and tweak the impact. https://filmstro.com/blog/how-to-remove ... n-audacity

but you can also just find and filter the noise with graphic eq like synapse GQ-7 or
use Softtube dyna-mite noise gate patches.
or just leave it in and call it Lo-Fi

I'll try these.
Thanks.

[Alandalton]

Thanks.

[Alandalton]

I'm actually trying to remove the sound of skiing from the part of this clip where Homer says "stupid sexy Flanders"
It's a hissing noise so I guess some sort of EQ might do.

[joeyluck]

iZotope RX could do it

[challism]

Or...... just say it's Lo-Fi and call it a day.

[Loque]

Or...... just say it's Lo-Fi and call it a day.

Thats not HQ.
You need to have it in 198khz, 32bit, remove noise and than use the best analog noise generator plugin to add some hizz and noiz. Than it sounds vintage/lofi/analog and not like cheap crap

Back to topic:
* Use noise gate
* Use a LPF or EQ
* Mix it with an phase inverted version of only the noise (record it quiet)
* There are repair plugins/tols availabel. Try them...
* Improve recordings
* Live with the noise

[challism]

Or...... just say it's Lo-Fi and call it a day.

Thats not HQ.
You need to have it in 198khz, 32bit, remove noise and than use the best analog noise generator plugin to add some hizz and noiz. Than it sounds vintage/lofi/analog and not like cheap crap

Back to topic:
* Use noise gate
* Use a LPF or EQ
* Mix it with an phase inverted version of only the noise (record it quiet)
* There are repair plugins/tols availabel. Try them...
* Improve recordings
* Live with the noise

That's a great tip, right there!
* Mix it with an phase inverted version of only the noise (record it quiet)

Try to isolate the noise with an EQ or filter, but instead of cutting it out, record it or bounce to track. Then invert the noise and mix it with the original signal. With any luck, you will get desired phase cancelation of the troublesome noise.

[jam-s]

I'm actually trying to remove the sound of skiing from the part of this clip where Homer says "stupid sexy Flanders"
It's a hissing noise so I guess some sort of EQ might do.

For this clip you could try to throw spleeter at it to isolate the voice.

[selig]

]
That's a great tip, right there!
* Mix it with an phase inverted version of only the noise (record it quiet)

Try to isolate the noise with an EQ or filter, but instead of cutting it out, record it or bounce to track. Then invert the noise and mix it with the original signal. With any luck, you will get desired phase cancelation of the troublesome noise.

That will never work, since we know that even a slight level or timing difference of even a sample will prevent a full null from an EXACT copy of the audio. But an exact copy of the isolated noise doesn't exist here. At best you could possibly find a similar noise print, which more sophisticated software could use to reduce the noise (anyone remember DINR?).
Inverting “similar” random noise is going to be pretty much exactly the same as not inverting random noise, as noise is random by definition. And involving a filter/EQ will introduce phase issues that are all but guaranteed to prevent a null, all other things being equal.

[crimsonwarlock]

Maybe Cockos ReaFir might do the trick (and it's free):

ReaFIR features:

- FFT based dynamics processor (requires host plug-in delay compensation).
- Supports FFT sizes of 128-32768.
- Useful in/out frequency response display.
- Supports defining response curves both using any number of points, or freehand mouse.
- EQ - can be used as a linear phase mastering EQ, or as a super-effective surgical EQ.
- Compressor - can compress at a fixed ratio with a per-band threshold.
- Gate - can gate with per-band threshold.
- Subtract - can build noise profiles and subtract noise from the signal.

https://www.kvraudio.com/product/reafir_by_cockos

[moneykube]

(anyone remember DINR?).

used it a lot working on feature films and tv shows when I worked in a studio... it worked pretty good most times> but don't overdo the reduction... many artifacts are the result. As you said... noise changes, especially in long film clips, where a fan turns on and off in the distance, generators and the like... I found removing fan noice quite a challenge... many frequencies involved> I was never able to remove such sounds 100 percent. Often those areas of a show would have to be revoiced with ADR> very time consuming and much work syncing back then and likely now... there may be better tools for that now.
Someone here may even have a list hint hint

[challism]

]
That's a great tip, right there!
* Mix it with an phase inverted version of only the noise (record it quiet)

Try to isolate the noise with an EQ or filter, but instead of cutting it out, record it or bounce to track. Then invert the noise and mix it with the original signal. With any luck, you will get desired phase cancelation of the troublesome noise.

That will never work, since we know that even a slight level or timing difference of even a sample will prevent a full null from an EXACT copy of the audio. But an exact copy of the isolated noise doesn't exist here. At best you could possibly find a similar noise print, which more sophisticated software could use to reduce the noise (anyone remember DINR?).
Inverting “similar” random noise is going to be pretty much exactly the same as not inverting random noise, as noise is random by definition. And involving a filter/EQ will introduce phase issues that are all but guaranteed to prevent a null, all other things being equal.

I don't have much experience trying to remove noise, and haven't ever tried the method I described. On paper, it seems like it would work; Giles corrected that assumption reminding me that noise is random, so it wouldn't work. He has forgotten more than I've ever learned, so I'd go with his advice and ignore my thoughts.

Giles, what about splitting the signal, then filtering out all but the noise (on one channel), inverting it, then mixing it back into the other channel? Perhaps even applying a micro delay to one channel (100% wet, no feedback) and trying to fine tune the timing. Could something like that work?

[Creativemind]

Try Izotope RX8. Loads of things in there to denoise / dehiss etc audio.

[Baylo]

Giles, what about splitting the signal, then filtering out all but the noise (on one channel), inverting it, then mixing it back into the other channel? Perhaps even applying a micro delay to one channel (100% wet, no feedback) and trying to fine tune the timing. Could something like that work?

Surely if you could filter out “all but the noise”, then you could equally just filter out the noise to start with, without worrying about inverted signals. But we can’t and so end up back at square one again.

[selig]

Giles, what about splitting the signal, then filtering out all but the noise (on one channel), inverting it, then mixing it back into the other channel? Perhaps even applying a micro delay to one channel (100% wet, no feedback) and trying to fine tune the timing. Could something like that work?

1: "filtering out all but the noise".
If that was possible we'd not be having this discussion! You can only filter out (actually, just reduce the levels of) frequencies, you can't target noise and voice separately that are at the SAME frequency. There will always remain some voice in the "noise" and vice-versa.
2: Also as I mentioned earlier, the phase shifts from the filtering would become obvious when summing the two channels back together, so the results would not be as expected. Trying filtering white noise on a parallel track so there's just a few octave or so in the mid range - when inverting one channel you won't get a few octave wide dip as expected. Instead you get a narrow notch and obvious comb filtering artifacts.
3: delay won't help here since there is no overall delay introduced by any of the processes.

There are always new ideas to be found, but this is an old one that's been tried many times to no avail, similar to wanting to use polarity inversion to isolate the center channel of a stereo mix.

[challism]

1: "filtering out all but the noise".
If that was possible we'd not be having this discussion! You can only filter out (actually, just reduce the levels of) frequencies, you can't target noise and voice separately that are at the SAME frequency. There will always remain some voice in the "noise" and vice-versa.

Surely if you could filter out “all but the noise”, then you could equally just filter out the noise to start with, without worrying about inverted signals. But we can’t and so end up back at square one again.

Geeze. A guy has to pick his words REALLY carefully around here, doesn't he? Obviously, I didn't literally mean "all but the noise" but rather meant ranges around troublesome frequencies. But the point is well-taken.

external-content.duckduckgo.jpg (56.84 KiB) Viewed 3733 times

.
It also seems that the type of noise we are talking about matters. If it can be isolated to a frequency range or not.

There are always new ideas to be found, but this is an old one that's been tried many times to no avail, similar to wanting to use polarity inversion to isolate the center channel of a stereo mix.

I was mentioning the adding of delay in case the filtering and polarity inversion of one channel caused any latency, which seems possible (unless I'm wrong again!). So it seemed a delay with such accurate precission as Normen Hansen VMG-01 Sample Delay would be suitable. But if all of this isn't even possible... it's a moot point, isn't it?

[jam-s]

I just threw RX8 on it using spectral De-noise trained on a section of the isolated skiing sounds. Seems to work quite OK: https://we.tl/t-w6sGJMLyvu

[ravasb]

Now all you need is something to remove the copyright.

[jam-s]

Now all you need is something to remove the copyright.

Depending on jurisdiction it might/should fall under fair use.