How to deal with low end rumble

[PippaPumpkin]

Hello,

I apologize if this is too basic of a question for this forum.

Currently I'm working on a virtual instrument I want to give away eventually. Every sample has this low end rumble and I am afraid that this could be an issue in a mix. Since the instrument is meant to be played like a pad sound with a bunch of notes stacked, I am worried the low end rumble might build up and take space needed for kick and bass. I don't have a subwoofer and it's hard for me to judge from the audio alone.

Here is a picture of the situation:




Is it better to cut it out of every sample? The notes go down to E2/80Hz. Would you recommend a low shelf or a high pass filter and how many db/oct?

Thanks in advance!

[adrian_k]

Just a quick point on not having a subwoofer - do you have some reasonable headphones? Good option for checking the low end.

If it was me I probably would remove anything from the sample that wasn’t needed. I think you might have to try a few things and see how they sound when you stack them.

[Wonks]

You may not have it in reality.

The sample size used as default by most real-time frequency analysers with a waveform display is too small to accurately portray the low frequency content in order to update the higher frequency waveform more often.

To accurately measure 20Hz, you need a full wave period i.e. 1/20th of a second, or the (sample rate/20) samples. To measure subsonic frequencies, you need even longer. Too small a sample size and you get a form of aliasing from higher frequencies folding back as low frequencies and showing up as frequencies that aren’t really there.

Put a high pass 120dB/octave slope filter at 100Hz on the sound and look at the waveform and you’ll almost certainly see a very similar waveform, despite you having filtered it out.

So either

1. use a bar graph type frequency display, where the coders should have used different sample sizes for each frequency.

2. A waveform frequency scope where you can adjust the sample size (e.g. the Melda frequency analyser), and set a large sample size. You should see that phantom low end drop right away (though the treble content will be slow to update so once you know the unwanted bass isn’t there, you can set it back to its original value).

3. Better still IMO, select a small length of the recorded waveform in your DAW and do an offline frequency analysis which should give you a much better analysis of the bass content.

Once you have checked your sounds and are happy there isn’t an issue, then you can start to recognise false bass displays on real-time waveform displays.

If you have got unwanted low frequency noise, then I’d just use a gentle high-pass filter. I think Hugh has mentioned no more than 18dB/octave to avoid filter ringing.

Yes, you may get some phase issues, but only below the frequencies you are interested in. And you won’t hear the phase shifting anyway unless you mix the sound with an identical one that hasn’t got any phase shifting or a different amount at those frequencies.

[adrian_k]

I think Wonks might have meant “use a gentle high pass filter”. And personally I’d start with 12dB/octave.

This all highlights the point about using your ears more than your eyes.

[Wonks]

I think Wonks might have meant “use a gentle high pass filter”. And personally I’d start with 12dB/octave.

This all highlights the point about using your ears more than your eyes.

I did. Thanks. Edited now to avoid confusion.

I continue to get low and high pass mixed up when writing. It’s because a high-pass gets put at the low end of the wanted frequency range (and vice versa).

If you do have low frequency noise (e.g. if using a mic, there might be some traffic noise or ventilation noise in the background), the slope you use will depend on how much there is and if it runs all the way up to the wanted sounds or not.

If theres a gap, and it’s just really low rumble, then a gentle slope should be fine.

If it runs up to the wanted frequencies, then you either have the options of re-recording and trying to eliminate the unwanted rumble e.g. turn off the ventilation or record when traffic is at a minimum or use more aggressive filters.

You can always use two filters in series; an aggressive one at say 25-30Hz, to cut out any subsonic noise, and a gentle one at say 60Hz, to reduce noise levels above that and reduce any ringing from the sharper filter.

[adrian_k]

I continue to get low and high pass mixed up when writing. It’s because a high-pass gets put at the low end of the wanted frequency range (and vice versa).

Me too, that's why I noticed!

You can always use two filters in series; an aggressive one at say 25-30Hz, to cut out any subsonic noise, and a gentle one at say 60Hz, to reduce noise levels above that and reduce any ringing from the sharper filter.

Excellent point.

[Hugh Robjohns]

Wonks makes a very good point about false LF content appearing on this kind of spectral display. You need to configure the analyser to use a big sample size and lots of averaging to really find out what's going on at the subsonic end of things.

But my advice would simply be to build a stack of your samples to create the pad and listen for low-end problems. Experiment with an 18dB/oct high-pass (low-cut) FILTER tuned somewhere around 60Hz to see if you can hear any benefit on headphones.

[resistorman]

I'm one of those people that aggressively cuts unnecessary low frequencies out of every track and channel except bass, drums, etc. In PA work I use the subs on an aux So obviously my advice is to use very good headphones, try different filters and slopes, and run it up until you hear the effect on what you want to keep and back it down.

[Wonks]

Just to highlight what the effect of the sample size does to the indicated low frequency content, here's the same small section of a vocal track displayed in Melda's MAnalyser first at 1024 samples:


And then at 32768 samples.


The greater the number of samples the greater the accuracy of the frequency display at both high and low frequencies, though it does slow down the update of the display considerably. 32768 samples at 44.1kHz is around 3/4 of a second.

But if you want to know visually whether there is significant low end in the samples, then this is something you need to do, or else do an offline analysis.

In this instance, although of a female vocal (so you normally wouldn't expect too much below 200Hz), it was a rough-and-ready recording to capture a song quickly, so the mic was picking up thumps from the keyboard beneath it. So although the low frequency stuff is 30-40dB down on the peak vocal, it still benefited from high-pass filtering.

[PippaPumpkin]

Thank you for all the great answers.

@Wonks

Here is how I did set up SPAN (in the picture above). So it should be fine right?



By the way if anyone can tell me what the drawback of high overlap (makes the graphic move smoother) is, that would be greatly appreachiated as well.

Is there a free tool I can do offline frequency analysis with? As far as I understand it the blue graphic above is infact offline, but I am not sure.

Put a high pass 120dB/octave slope filter at 100Hz on the sound and look at the waveform and you’ll almost certainly see a very similar waveform, despite you having filtered it out.

It does disappear.

If you have got unwanted low frequency noise, then I’d just use a gentle high-pass filter. I think Hugh has mentioned no more than 18dB/octave to avoid filter ringing.

I can set a highpass 6-18db/octave. Since my lower pitch samples go down to 80Hz/E2 should I move the highpass filter for those specific samples since the low end there is actually useful? For the higher pitch samples I would keep the HP constant at a frequency that covers the rumble, no need to micro adjust there. What do you think?

If you do have low frequency noise (e.g. if using a mic, there might be some traffic noise or ventilation noise in the background), the slope you use will depend on how much there is and if it runs all the way up to the wanted sounds or not.

I did record a friends voice. The enviroment was pretty quiet but I did had three preamps (including the audio interface input) to color the sound. I then ran the samples through a noise reduction tool, but appearently it did not get rid of everything. I should have been more careful at this stage. Since I did alot of work on the samples since then, I need to fix this at this stage.

You can always use two filters in series; an aggressive one at say 25-30Hz, to cut out any subsonic noise, and a gentle one at say 60Hz, to reduce noise levels above that and reduce any ringing from the sharper filter.

That is a great idea. Thank you.

But my advice would simply be to build a stack of your samples to create the pad and listen for low-end problems. Experiment with an 18dB/oct high-pass (low-cut) FILTER tuned somewhere around 60Hz to see if you can hear any benefit on headphones.

So obviously my advice is to use very good headphones, try different filters and slopes, and run it up until you hear the effect on what you want to keep and back it down.

I'm going to try this as well. Thank you.

The greater the number of samples the greater the accuracy of the frequency display at both high and low frequencies, though it does slow down the update of the display considerably. 32768 samples at 44.1kHz is around 3/4 of a second.

I have mine at 65536 Block Size at 192kHz.

In this instance, although of a female vocal (so you normally wouldn't expect too much below 200Hz), it was a rough-and-ready recording to capture a song quickly, so the mic was picking up thumps from the keyboard beneath it. So although the low frequency stuff is 30-40dB down on the peak vocal, it still benefited from high-pass filtering.

I suspect that my analog signal chain introduced this noise floor. The voice itself goes down to 80Hz in the lowest sample.

Thanks again. This give me a much better understanding how to approach these issues.

[Wonks]

I'm not 100% sure what overlap does in Span, but I assume with it set to 0%, you get one block of samples 65536 samples wide, and then you get the next block of samples 65535 samples wide and so on, and Span updates the graph at the end of each block. So at 196kHz, that's about 1/3 of a second. At 44.1kHz, that's about 1.5 seconds.

With overlap set at say 50%, Span uses the first block of 65536 samples but the next block starts 50% of the way through that block after 32768 samples and then uses another 32768 samples. The next block starts after the second block has imported 32678 samples. So instead of blocks being end to end,

11111111
0000000022222222
000000000000000033333333
00000000000000000000000044444444

....you now have overlaps of blocks.

11111111
000022222222
0000000033333333
00000000000044444444

With the display being updated every time a block of 65536 samples (or whatever your sample size is) has been collected and processed.

If the maximum overlap setting is roughly the 93.8% shown in your screen shot, then the display should get updated about 16 times in the time it would take the screen to update with 0% overlap.

This obviously takes more processing power and more memory than with no overlap, but with modern processors and loads of RAM, it's not an issue. But back when Span first came out, and computers were a lot slower and RAM sizes a lot smaller, if your computer was struggling to update on the big overlap settings, you could reduce the overlap at the penalty of slower update speeds for the graphics.

Don't forget that 65536 samples is 65535 x 3 bytes with 24-bit word length samples, is 196,608 bytes. With maximum overlap, you need to store 16 separate buffers of 65536 samples at any one time, which is just over 3MB of memory. That may not sound like a lot now, but in addition to all the RAM needed for the FFT processing as well, it was using a fair chunk of the computer's available RAM when Span Plus first came out in 2013 and 1GB RAM was fairly common.

So yes, if I've got it right, then high overlap will give smoother graphics but take up more processing power and RAM. But with 16GB+ RAM being fairly common for DAW computers, these days you can happily set up big overlaps.

[PippaPumpkin]

That is awesome. This way I can use the SPAN VST in the highest resolution running buttery smooth and fast. I was concerned that it might come with drawbacks in accuracy. Voxengo would be smart to hire you for writing their manuals, since close to nothing is mentioned about this feautre in the manual.

Here is what I did.



This is the best case scenario since it is the highest note at D5/590Hz.

My plan is to apply two HP filters as mentioned. (D5/590Hz -> 120Hz 12db/oct & 60Hz 120db/oct)

Those filters I want to lower each note accordingly to keep the distance between the desired signal and the filters constant. (Cis/555Hz -> ???Hz 12db/oct & ???Hz 120db/oct)

This I could manually adjust for each note individually, however that would take alot longer given how many notes I will have to filter. I would like to mathematically calculate the postion of the HP filters, to keep the relative position of the HP filters to the desired signal. Any ideas how to calculate this?

At some point I would have to make some sort of transition, due to the spectrum of note going down to E2/41Hz. At some point I could not move the 120db/oct HP any lower, the limit being 20Hz, making it impossible to keep the relation with the filters constant up to this point. At E2/41Hz I might only be left with a bit of lowshelf to adjust for the added energy by the rumble.

The low end noise floor is constant. My theory is, that it is hum/noise of my analog gear that went through the noise reducer filter threshhold. The notes themselves are very constant as well since the enviroment was very controlled and the sounds are edited to be this way.

[Drew Stephenson]

If the low end noise is constant have you considered trying a noise removal tool before filtering?

[PippaPumpkin]

If the low end noise is constant have you considered trying a noise removal tool before filtering?

Hm... I am not sure how I can make this work. The noise removal tool I have has a learn function. For this I always record silence for the tool to know what to cut out. The product of my editing is a huge number of those recordings stacked ontop eachother. Since the noise floor is so homogenous stacking the noisefloor to recreate this might not be necessery since the sum of many noisefloor tracks is pretty much the same as the original noise floor/silence recording? I going to give it a try.

[PippaPumpkin]

So this is the result:



The signal I am editing here has been denoised, stacked and crossfaded. Unfortunatly I did not have the original denoised silence recording to work with. The original silence recording did however do the trick well enough. I have no idea, maybe treating the audio like this is better than denoising everything in one go. I'd probably try to get it right in the first run the next time around though.

[Zukan]

Band-pass EQ every channel in your DAW to remove redundant frequencies that cause all manner of problems with threshold dependent and air band processes. Most EQs now afford a solo function which helps to audition the offending frequency ranges.

Trust me, this simple process can help tremendously.

[RichardT]

There are some high end sample libraries in which some samples have low end noise - Spitfire comes to mind.

The noise is quite obvious - sometimes I even need to apply HP filtering myself. But it is limited to a few notes.

I would follow the advice above. Stack many samples, put on an LP filter, and see if there actually is a problem by listening. If your speakers aren’t up to it, get some phones.

If so, the simplest approach is the one recommended by Zukan.

To minimise any side effects, particularly when multiple samples are stacked by the end-user, I would use the very best EQ you can find.

For low-end work with incredibly clean results I can recommend the brilliant TBT Kirchhoff.

[tea for two]

With EQ and Filtering I just start with taking out a huge chunk of the frequency thereafter bringing more and more of the frequency back in until it suits what I'm wanting to hear.
Logic's AUBandpass plugin is a doddle to select frequencies, to top and tail a sound, even a mix.

::

TBT Kirchhoff a funny thing happened, I installed the demo was using it in Logic for a piece. Then couple of weeks later no sound from that piece nothing whatsoever no matter what I tried. Other pieces were playing fine. For the life of me couldn't figure ooot.
Then just in the oorrff chance I turned of TBT Kirchhoff demo hey presto sound back on : the demo expired which then caused the whole piece to not have any sound in Logic.

[PippaPumpkin]

Band-pass EQ every channel in your DAW to remove redundant frequencies that cause all manner of problems with threshold dependent and air band processes.

Aren't you concerned about ringing at all?

Most EQs now afford a solo function which helps to audition the offending frequency ranges.

Trust me, this simple process can help tremendously.

Yes, I think you are correct. For me I try to at least get rid of the noise floor of my recordings, especially since I work on virtual instruments at the moment. If there is something going on that deviates from the main signal I am afraid it will clutter up the mix for endusers.

There are some high end sample libraries in which some samples have low end noise - Spitfire comes to mind.

I don't know their products, but I find this disappointing. I want to make samples that you can trust. Unfortunatly I have very limited experience and knowledge. That's why I am thankful for all the help you guys are providing. Thanks for the support.

I would follow the advice above. Stack many samples, put on an LP filter, and see if there actually is a problem by listening. If your speakers aren’t up to it, get some phones.

I don't know how well I can judge this by ear in my situation. But theoretically speaking, who wants noisefloor in their samples? If it is avoidable without introducing to many drawbacks, like excessive filter ringing its fine.

If so, the simplest approach is the one recommended by Zukan.

To minimise any side effects, particularly when multiple samples are stacked by the end-user, I would use the very best EQ you can find.

For low-end work with incredibly clean results I can recommend the brilliant TBT Kirchhoff.

I have been using the Melda Audio band pass but if the Kirckhoff is better I might have to look into it.

With EQ and Filtering I just start with taking out a huge chunk of the frequency thereafter bringing more and more of the frequency back in until it suits what I'm wanting to hear.
Logic's AUBandpass plugin is a doddle to select frequencies, to top and tail a sound, even a mix.

I have so little experience that I often try to understand the theoretical side of things first, because I am so clueless. There are these DJ-Filter plugins with high resonance settings. Is that what filter ringing sounds like?

TBT Kirchhoff a funny thing happened, I installed the demo was using it in Logic for a piece. Then couple of weeks later no sound from that piece nothing whatsoever no matter what I tried. Other pieces were playing fine. For the life of me couldn't figure ooot.
Then just in the oorrff chance I turned of TBT Kirchhoff demo hey presto sound back on : the demo expired which then caused the whole piece to not have any sound in Logic.

Things like this can eat up alot of time.

[Wonks]

Zukan was one of the main studio synth programmers in London in the 80s, then worked for several synth companies on their samples, had his own sample production company (until piracy made it unviable), is a Grammy judge and a Recording Association approved educator.

So, apart from being a lovely person, there isn’t anything he doesn’t know about creating samples.

Use your ears. If the filter is causing audible
ringing then it’s either too steep or set too high/low (or both).

[Hugh Robjohns]

Band-pass EQ every channel in your DAW to remove redundant frequencies that cause all manner of problems with threshold dependent and air band processes.

Aren't you concerned about ringing at all?

There is no ringing with second order (12dB/octave) or third order (18dB/octave) filtering to 'bracket' the wanted audio signal. Youd have to use much steeper filters to risk ringing... but there's no need for anything steeper just to remove a bit of low-end noise.

[James Perrett]

I have been using the Melda Audio band pass but if the Kirckhoff is better I might have to look into it.

There's no need for anything fancy if you are using a proper DAW - the DAW's standard eq should be fine for this.

[RichardT]

I have been using the Melda Audio band pass but if the Kirckhoff is better I might have to look into it.

There's no need for anything fancy if you are using a proper DAW - the DAW's standard eq should be fine for this.

I disagree with you there James! I definitely get fewer unwanted side effects at low frequencies with, say, the Kirchhoff compared even to Fabfilter Pro Q3.

Anyway, I'd encourage the OP to keep things simple. Choose an EQ, apply some filtering, and the job is done.

[PippaPumpkin]

Zukan was one of the main studio synth programmers in London in the 80s, then worked for several synth companies on their samples, had his own sample production company (until piracy made it unviable), is a Grammy judge and a Recording Association approved educator.

So, apart from being a lovely person, there isn’t anything he doesn’t know about creating samples.

I see. He must be really good. I think I got the idea about ringing from an izotope mastering course video on youtube. If I remember correctly the gentleman in that video said that it is better to use shelves if possible, to avoid ringing. Maybe I can find this video again.

While I am looking. I am experincing weird behavior with decentsampler on one note only:



My sample seems to be fine but decent sampler adds something I would describe as a a very slow phaser effect with rumble in the very low end?

I have found the video:
https://www.youtube.com/watch?v=GpLRtFHSA0U&t=1110s

He talks about a 'significant amount of phase shift'. He puts it into general terms so I'm not sure if he is talking about 6-12db/oct.

[James Perrett]

He's using a very steep high pass filter - not the sort of thing that would normally be used in mastering unless you had a very specific problem to correct. I don't know why mastering is being used in the title of that video because that isn't a normal mastering process.

Remember that's a marketing video for Izotope - like most recording related videos on Youtube, you should take it with a pinch of salt.