Remove peaks for specific frequencies

I’m trying to remove peaks for selected frequencies only, and not have to apply a constant reduction across the whole audio file.

In the screenshots below there are a couple of peaks at certain frequency ranges, so for example at about 4160 Hz and then a couple more above 11 kHz (with the 4000 Hz one being the most annoying). And I’d like to decrease the volume of only those sections where they reach a certain threshold, while leaving the rest of the audio file untouched.
Applying a notch filter at ~4000 Hz and ~12 kHz makes the audio sound hollow, so I’m looking for a more refined way to do this, but weren’t able to find any plugin for this. Basically something like a “reverse” noise gate or a threshold aware notch filter or a frequency specific compressor.

Is there something available, or is there some other way to achieve what I want?

audacity-file.jpg
audacity-spectrum.jpg

Is this a theatrical presentation?

makes the audio sound hollow

Especially the 4000Hz one. That one is really close to the 3000 center-of-hearing tone. Messing with tones in that range is not easy.

The Q of the notch is the sharpness of the cut. Low values of Q produce broad notches and take some of the tones on each side of the goal, but High Q notches are not a gift, either. As the sharpness of the notch increases, the more likely it is to “ring” and create its own sound.

A bell is a High Q salad bowl.

There is one way to do this with minimum damage if the interfering tone is stable. Generate a new tone at the exact frequency and phase. Change it so its volume and the phase cancels the interference. That doesn’t have Q or slope problems and if it succeeds, the interference will vanish without a trace.

However, hitting that frequency and phase exactly will take you a while and if the tone is coming from a natural generator like a bird or electric motor, it doesn’t stay in one place and the cancellation will fail over time.

So those are the problems. I don’t know of any good Audacity tools to help.

Koz

Because that 4kHz spike is so thing, I’d expect a notch filter to work quite well. If you only want to apply the notch in specific places (and not the whole track) then you could use the “Spectral Multi-tool” on each of the bright lines: https://manual.audacityteam.org/man/spectral_edit_multi_tool.html
Unfortunately I don’t think this can be automated without creating your own plug-in effect.

The higher frequency peaks are a bit more tricky because they are fairly wide (broad frequency range) and there’s a whole bunch of frequencies.
You could try the same approach, though you may find that the spectral edit parametric EQ, or shelf filter work better. You should try each of them - it’s difficult to predict which will work best just from looking at those pictures:
https://manual.audacityteam.org/man/spectral_edit_parametric_eq.html
https://manual.audacityteam.org/man/spectral_edit_shelves.html

For the higher frequencies, a “multi-band compressor” may work, though:

  1. Audacity does not have one built in, but there’s probably a free VST version available if you search on-line.
  2. A multi-band compressor is not likely to work well on the 4kHz tone because a multi-band compressor probably won’t be able to select a sufficiently narrow frequency band.

Actually I’m trying to remove a squeaking brake noise from my Mountainbike videos, so the sound does change depending on how hard I was hitting the brakes. The 4000 Hz noise is one brake, the one > 11 kHz the other one, with the 4k one being much more pronounced. It’d be easier if it was the other way around, I’d probably just use the Low Pass filter to remove/dampen anything above 11k, but for 4k that’s just not possible without the audio sounding like a bad telephone call.

So I was hoping there’s something that will only reduce the volume of a selected frequency range if it’s volume is above a certain level, as most times there isn’t too much going on at that range, except for a somewhat steady lower noise level, as you can see in the first image.

The Noise Gate plugin already has some of the functionality I’d be looking for, unfortunately in the wrong direction - it reduces frequencies below a certain threshold, and not above it. Also it only has an upper limit for the frequency, missing an upper/lower selection (resp. that Q factor you mentioned).


Actually I just gave it another go with the notch filters, identifying a couple of more frequency ranges at ~6000 and ~8000 that belong to the first brake. I can make it somewhat work, but I’d really prefer if only the relevant sections were modified and not the whole audio file.
I’d also prefer not to do this manually for each occurrence, which will be too time consuming. If I had any clue of the Nyquist language I might have given it a shot to modify the Noise Gate plugin, but learning it from scratch is too much of an effort either.

There may be a “semi-automated” way to do this with plug-ins that already exist, but I would need to test on an example of the actual audio.
Could you post a short sample, around 15 seconds, mono WAV format, of unprocessed recording. The sample should contain some of the squeak, and some of the audio without the squeak.
See here for how to attach a file to your post: https://forum.audacityteam.org/t/how-to-attach-files-to-forum-posts/24026/1

Ok, attached is the sample roughly from where the screenshot above was taken.

These de-esser plugins can selectively attenuate the squeak …
https://forum.audacityteam.org/t/updated-de-clicker-and-new-de-esser-for-speech/34283/1
https://forum.audacityteam.org/t/please-help-with-de-sser/48588/14
[Steve’s is much faster].

The higher frequency peaks are a bit more tricky because they are fairly wide (broad frequency range) and there’s a whole bunch of frequencies.

I note the “Size” setting in the spectrum posted is only 1024. That’s “sloppy” and will mis-read the width of the display spikes.

I’m trying to remove a squeaking brake noise from my Mountainbike videos, so the sound does change depending on how hard I was hitting the brakes.

That’s the job from Hell, isn’t it? How are you going to prevent those recording errors from happening in the future, given how hard it is to deal with in post production? You have four different forum elf responses so far and no “push this button and the problems go away.”

Even a cellphone with its world-class dynamic environment suppression and noise cancellation would not be able to deal with this.

If you aren’t already, you should be using a close-talking microphone…

And get good at bicycle maintenance. Brakes should not scream at you when you stop. I bet your bike instructions have something to say about care and adjustments.

Koz

my Mountainbike videos

Did you put the Teflon strip between the tire and tube? Turns out one of the connectors to the Los Angeles Bike Path goes through an oil field. I’m not making that up. In that oil field grows a version of bougainvillea that can fight back. The thorns took out two regular tires (tyres) before I learned the tricks.



Koz

Thanks for the sample. This is tricky. If we look more closely at that 4 kHz squeak, we can see that it is not a constant frequency, but covers a range of frequencies and is not constant.



What we will need to do, is to separate out that entire frequency band from around 3.9 kHz to 4.47 kHz and treat that frequency band.
I’ll continue this, with a list of tools and steps, in my next post…

Continued from previous post…


Tools:

Download, install and enable this “Pop Mute” plug-in:
Popmute.ny (2.89 KB)
Note that this is a slightly modified version of the Pop Mute effect from the Audacity wiki. This modified version allows the “Threshold” to be set lower.

Download this “Eq curve” definition file:
bandpass.xml (204 Bytes)
and import it into the Equalization effect using the “Save/Manage Curves” button, then click the “Import” button.
(see: https://manual.audacityteam.org/man/equalization.html#manage)
You should then see that a new curve called “bandpass” is listed in the “Select Curve” dropdown of the Equalization effect main GUI.
Note: To use this curve correctly, the Equalization effect must be in “Draw” mode.


Steps:

  1. Import the audio file (I’m using “Brake Noise Test Section - Untouched - Mono.wav”)
  2. Select the track (double click on the track, or click on an empty part of the info panel on the left end of the track), and duplicate it (“Edit menu > Duplicate”, or “Ctrl + D”)
  3. Select the second track only, and apply the Equalization effect with the “bandpass” curve (must be in “Draw” mode)
  4. Duplicate the second track
  5. Select the second track only, and apply the “Invert” effect
  6. Add the first track to the selection (Ctrl + Click on an empty part of the info panel) so that tracks 1 and 2 are selected
  7. “Tracks menu > Mix > Mix and Render”. The first two tracks are mixed together, and create a new “Mix” track at the bottom.
    You now have two tracks. The first track contains only frequencies between about 3.9 kHz to 4.47 kHz, and the second track contains all frequencies except for that frequency band.
  8. Select the first track only, and apply the (modified) Pop Mute effect. You may need to tweak the settings, but I’d suggest as a starting point:
    popmute.png
  9. Select both tracks (“Ctrl + A”), and “Mix and Render”. This recombines our processed part with the unprocessed part.

This does not completely eliminate the squeak, but it reduces the squeaks substantially.

I’m running tubeless (or more precisely, tubeless with Procore), which virtually eliminates all problems with reasonably sized holes, e.g. caused by thorns. I’ve read that somewhere in the states goatheads are a problem as well, for which tubeless seems to work fine, too.

Well, I’ve already thought about using a dedicated microphone in the past, although certainly not something like a headset as seen in the image. I think that wouldn’t work too well. However for now I’m not planning to use another mic, that would be too much of a hassle, and since I’m not making any money through Youtube anyway, it’s not worth the additional effort. It’s not that I need the most brilliant audio quality to make a living, I just don’t want it to be awful either. :wink:

As for why the brake is squeaking as much, it really seems to be a combination of the brake pads and the brake rotor. My previous combination was pretty much silent, so I guess I’ll just bear with it until I’ve worn them down. Or if I’m annoyed enough sooner.
Brake performance itself is actually not limited, and the sound does goes away eventually once the brake gets hot enough. The down hill sections here just aren’t not long enough for them to get that hot for the most time though. This and that and I’m not heavy enough. :mrgreen:


Oh wow, that spectrogram of the processed file looks pretty much what I imagined. I was able to reproduce this on my stereo sample with your instructions, so thanks for your effort. Unfortunately it’s no one-click solution either, but it sounds much more promising (literally) than just throwing a couple of notch filters at the audio.

If you have enough of these recordings to justify the effort (or just for fun :wink:), you could make those instructions into a Macro (see: Macros - Audacity Manual)