Nord User Forum

CountFosco wrote:A high pass filter might do a little to centre the waveform a little, but I don't think it will fundamentally solve the problem. That 2nd note looks like a pretty tricky waveform to normalise in comparison to the first note. I'm tickled by Mr_G's idea of rearranging phase in the frequency domain, but it sounds pretty clunky in practice, but it would be cool to hear if someone has a way to do it. I think it might be useful to mess around with the sound at its source a bit more, maybe tweaking an effect, waveshape, envelope or LFO might minimise the phase additions.

Obviously, it depends on exactly what the problem is I think there are two potential problems that we're discussing. To my eyes, it looks like the waveform of the 2nd note has some strong low frequency components that causes it to be asymmetrical over a period much longer than the base frequency (i.e. over many 100 ms). If you shift the phase of the low frequency component vs the others, it will just move the offset at which it peaks, but using a high pass filter will simply remove those low frequency components. As far as I can see, we're talking about frequencies maybe below 1 Hz which is not audible and should always just be removed.

I've created the following (gross) example to illustrate: The top line is a little more than a second of a 220 Hz tone with a lot of low frequency (1 Hz) content. This causes the "real" tone ("base frequency" here 220 Hz) to be offset from 0, varying over time. In this particular case it is still symmetrical over a longer timeline, but not if you would pick a small portion (like some 100 ms) of it. At the bottom, I've filtered it with a high-pass 10 Hz / 12 dB filter in Audacity:



So in essence, the high pass filter will remove the local "offset" that Mr_G_ refers to above too. And note: this is just to illustrate the concept.

Now, Mr_G_'s idea (from Manfred Schroeder) is indeed interesting, but as mentioned, re-arranging the phase will just move the low frequency content (in the signal above) to some other parts in time, so it may not help as much in this particular situation. Where it MIGHT help would be if you have very peaky signals you want to reduce in amplitude. The paper actually refers to signals with low autocorrelation, which probably isn't true here; he talks about radar signals with few spectral components, and goes on to derive a formulation of the required phase distortion to minimise the peakiness.

Now, I've not tried something like this, but it's not easy to generalise this to work well for all signals; the frequency dependent phase shift needs to be tailored to the signal. And it's easy to come up with examples where it will cause challenges as some signals may increase in peakiness in some parts while reducing at others with the same treatment.

Also, I'm not sure I would really want this. We're basically talking about using a non-linear phase-shifting filter -- normally when you design filters, you would like to avoid phase distortion.

In fairness, it is somewhat open to which degree phase distortions are audible, but audiophiles (of which there may be some here too) sometimes pay enormous sums of money to avoid phase distortion too

CountFosco wrote:Nice work, looks you've almost reached the limit of what could be possible. The idea of skipping the specific peaky notes should work

@baekgaard, nice post, and I've always enjoyed your signal analysis insights, but in this case, I'm inclined to disagree. I don't believe those signals are influenced, or modulated by, a HP filterable low frequency. My impression is there are two closish frequency sources interacting, and in the particular example of note 2, happen to interact with phase such that the addition is more extreme than for note 1. Why it happens inconsistently from note to note could be related to something at the source - eg waveshape modulated by LFO, slow chorus or detuning, randomly picking a dodgy spot in an LFO cycle, maybe a velocity sensitive effect (not sure if the notes were played by fingers or midi), I dunno, could be lots of things. That's why I suggested playing more with the source.

And coming back to Mr_G's frequency domain phase reshuffle, on rethink it might be difficult to find a way to get accurate enough phase resolution on two closely spaced frequencies to do this in a controllable way - but maybe some random tweaking could spew out an acceptable result. I still like the idea of trying it.

Thanks for the kind comments -- and likewise!

I agree that the offsets we're seeing here could either be an overlaid low-frequency signal OR a mix that also involves some superimposed waves close to one another in frequency (even if this could also have resulted in some areas with very low amplitude when they are out of phase) -- maybe from a chorus effect with multiple feedback parts? And indeed, based on the results that the OP mentioned, I look to be wrong in my interpretation since the HP filter didn't help. It's sometimes hard to tell only from looking at the waveform, of course, and your hunch was more correct here, it seems

EDIT: I just noticed the sample is there already, so I'll have a look at that too.

I had a look at the sample (available further up), and it looks like this may be caused by either a couple of similar waveforms (oscillators) being summed with some phase modulation of one of them, and/or perhaps one or more of the harmonics being slightly "out of tune" vs the lower fundamentals -- both of which are common situations for these types of sounds, I guess

I tried applying a strong notch filter at the 3rd harmonic (around 98 Hz) and that seems to "nuke" some of the spikes and make it more "even", but of course it also changes the sound a lot, so that's not a useful approach either.

So the best remedy is likely what the OP did, using a limiter to take care of the spikes. An maybe consider if there are effects on the original sample that could be switched off and applied in the instrument afterwards (e.g. chorus of phaser).

Interesting case

WannitBBBad wrote:

baekgaard wrote:I had a look at the sample (available further up), and it looks like this may be caused by either a couple of similar waveforms (oscillators) being summed with some phase modulation of one of them, and/or perhaps one or more of the harmonics being slightly "out of tune" vs the lower fundamentals -- both of which are common situations for these types of sounds, I guess

I tried applying a strong notch filter at the 3rd harmonic (around 98 Hz) and that seems to "nuke" some of the spikes and make it more "even", but of course it also changes the sound a lot, so that's not a useful approach either.

So the best remedy is likely what the OP did, using a limiter to take care of the spikes. An maybe consider if there are effects on the original sample that could be switched off and applied in the instrument afterwards (e.g. chorus of phaser).

Interesting case

Thanks, I agree with your thoughts on addressing the sound at the source if possible. One thing for sure is rolling off the bass a little that I had boosted. I'll blame my headphones The sound is a variation of "Analog Funk" I edited. It's not using Chorus, LFO or detuning, but the MONO mode along with Pulse wave set an octave lower than the Saw wave are variables to look at. Thanks again!

Analog Funk W.jpg

Ah, yeah - a sub osc could explain it too, as it's usually slightly detuned to fatten up the sound, and the overtones will then clash as the phase varies, similar to what we discussed above.

Thanks for filling in the background!

Sent from my phone in brevity