1

I have a stereo microphone recording that features two sources:

  • The signal is considerably more prominent in the left channel than in the right one.
  • The noise is present in both channels but slightly more so in the right one. It’s a combination of different types of undesired sounds including machine sounds, human activity and traffic noise (through a wall). It drastically changes over time.

Here is an close-up of the signals of a section featuring only the noise (logarithmic amplitude scale):

Example

Now, I want to filter out the signal as well as possible exploiting the aforementioned special properties of my recording (as opposed to standard noise reduction). I already tried to simply subtract the right channel from the left one¹. The result was surprisingly good, but I suspect that I can do even better with a more sophisticated tool.

Is there any tool dedicated to this? Or is there a combination of standard tools that I am missing?

I am currently working with Audacity; thus I prefer a solution using it when in doubt.


¹ I later found this blog post, which does something similar plus regular noise reduction.

8
  • The phase-invert trick only works on non-random sources. 'noise' may or may not be random, depending on its source. – Tetsujin Jun 1 at 11:04
  • @Tetsujin: Well, it’s not random (in the sense that it independently acts on each channel) as you can see in the image. – Wrzlprmft Jun 1 at 11:24
  • The waves may be somewhat similar, but they are not phase invertable to zero. – Tetsujin Jun 1 at 11:29
  • … which is why I want something sophisticated than a plain subtraction (phase inversion + mix). – Wrzlprmft Jun 1 at 11:32
  • 1
    …which is what I offered in my answer... – Tetsujin Jun 1 at 12:00
1

These days this is a simple task, using specialist software like Izotope RX. However, simple is not cheap;)

Audacity has something along those lines, but I've no idea how good it is - https://manual.audacityteam.org/man/noise_reduction.html

8
  • That feature of Audacity is just standard noise reduction using a signal-free passage to estimate the spectrum of noise to remove it. It does not exploit the fact that the signal is dominant in one channel only. – Wrzlprmft Jun 1 at 10:18
  • Yeah - that's what you get for 'free' I'm afraid. Try the trial of RX. Stuff like that is pretty much a 1-click fix. – Tetsujin Jun 1 at 10:52
  • Can you name a specific feature that does what I want? Right now, I have to rely on your confidence and that I managed to communicate my intentions to you. Also a name may help to find similar features in other software. – Wrzlprmft Jun 1 at 11:31
  • It depends on what the noise is. De-hum, de-rustle, de-wind, de-crackle, dialog isolate, spectral de-noise… RX also has a 1-button 'fix it for me' mode. – Tetsujin Jun 1 at 11:33
  • The noise is what is recorded on the second channel. That’s the main point of my question. – Wrzlprmft Jun 1 at 11:56
0

Assuming we are talking about the same noise rather than the same kind of noise on both channels, the problem you have here is essentially echo compensation with a comparatively short filter length (hopefully). The common algorithm for that task is called "LMS algorithm". For whatever reason, I've not found implementations in the usual load of free filter plugins but maybe I just have problems looking well.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.