Removing background noise in Audacity by differencing stereo channels | Hacker News Reader

Removing background noise in Audacity by differencing stereo channels

weblogs.asp.net

65 pointsjohns13 comments

Nice trick, but don't forget that uncorrelated noise is always additive. I know a few people who have built noise cancelling headphones (either for fun or for a project class I once TAed) who discovered (after dealing with lots of fun feedback issues) that their "quiet" output was noisier than the original output.

There's usually a way to make this trick work, but it's important to have accurate models of your various noise and interference sources.

Good tip.

I like to take a mono track, duplicate it, invert one of them, then make both of them into a stereo track.

Now play it and play with the proximity of the left and right speakers. When they're close, they cancel each other out. Similar principle of noise canceling headphones.

Do you do this just for fun? Correct me if I'm wrong but this just seems like a recipe for weird phasing issues.

Yes, it was just for fun, and to show my nieces and nephews that sound can be canceled.

what if you take the windowed cosine transform, and then see where the transform is within 10% of each other (or less than a threshold value). Then zero out all such regions. With the technique you're using I'd worry about the phase happening to be exactly 180 degrees if the frequency and position of the noise source are 'just so' - which might explain your observation of 'pops'.

This is slightly off-topic, I apologize.

Is there a way to run Audacity's noise-removal real-time (or a few second delay)? I've been trying to Google this, but without success.

There is plenty of real-time noise reduction software in VST plugin form. Here's an (expensive, but very effective) example: http://www.sonnoxplugins.com/pub/plugins/products/restore.ht...

Otherwise, you can get a long way with just traditional gating, EQing and maybe a multiband compressor.

Thanks, I should have mentioned that I don't know anything about audio, but I'll look in to it!

A gate only lets sound through when it goes above a certain level, so if you set the threshold just above the level of the noise, the parts of the audio that are just noise will be silenced. This obviously works best if you're treating something intermittent like speech or a single instrument, it wouldn't work on a whole piece of music.

An EQ lets you attenuate certain frequencies, so if there are some parts of the spectrum in your recording which are just noise, you could remove them, eg. for speech you could remove all the information below 300hz and above 12-13000hz

Multiband compression is a bit too complex to explain here :)

The specialist plugins tend to use a combination of these techniques and then their own special secret algorithms, which is why they're so expensive.

Thanks for the explanation! I'm already using Noise Gate in OBS (livestreaming), but the noise is still an annoyance. I'll definitely look in to EQ. Thank you so much. :)

The lowest hanging fruit for you will probably be improving the source. Get a better mic, make a pop shield out of an old pair of tights and a wire coathanger, fill the room with soft furnishings, make yourself a ghetto isolation booth by putting a duvet over your head and climbing into your wardrobe, etc.

That last one might sound silly but many international-pop-career-launching albums have been recorded like that.

I've been thinking about it, but I don't really want to use 100€ on a new mic. I'll try to "fix" this one first. :P Thanks for the advice!

Reply on news.ycombinator.com