Velvet Noise

[Spogg]

Indeed, velvet noise is great for reverbs. I have uploaded MVerb 7B+, check it out.

That sounds really nice Martin, thank you!

[HughBanton]

Mmmm, that's nice Another great reveal Martin.

Here's a thing. I recently became aware that noise generators, in general, tend to 'over-add' when played in chords.

Wozzat? Well I'm probably more sensitive to this than most because of my erhemm .. 'electric organ' background. Electric organs, if poorly designed, were very susceptible to over-adding.This was because they invariably involved a lot of borrowing of generators, and if you simply just make 2+2 = 4 it doesn't sound right: tones and harmonics tend to get over-emphasized.

So basically here, with just noise on its own .. play a note, add a second, add a third, add a fourth. It gets louder than you'd naturally expect.

This came up a couple of years ago (not with noise) when we were discussing Hammond organs, and Martin himself - who else! - immediately came up with the solution of squaring the control signals, summing them, and then taking the square root.

Anyway, that's just background. Why noise should over-add I cannot fathom; you'd think that adding random spectrums would behave by the proverbial textbook.

Anyway, try my modification; I imagined this would need to be fixed in poly but evidently not.

If you set up a tone, add a bit of noise, and try the test above. With the 'noise level compensation' engaged the noise portion is progressively attenuated, and the 0.9 factor seems about right to me to keep the balance right. Disengaged, the noise begins to swamp the tone.

Happy to be shot down if no-one agrees! (Might just be lockdown insanity kicking in )

H

VelvetNoise_level_comp.fsm

(371.28 KiB) Downloaded 1019 times

[martinvicanek]

... noise generators, in general, tend to 'over-add' when played in chords.

They certainly do if the noise pattern is exactly the same for each note. (I had this very issue with sibilants in MIDI Choir.) In that case the noise amplitude doubles for two notes played simultaneously, hence the power quadruples. Two flute players, on the other hand, would generate completely uncorrelated noise, hence the power would only double. With that in mind, I thought each note should come with a new seed to generate uncorrelated noise with respect to the noise of any other note.

Apparently this model is still not quite right for electric organs (and likely in other contexts, too). I suspect it might be because there are more noise sources involved, so the resulting noise power is not simply proportional to the number of notes played in a chord. Surely there is already some amount of noise even without any key pressed. So by all means, it is good to have a knob to adjust these things!

Thank you, Hugh, for this sharp observation!

[HughBanton]

It can drive you nuts after a few minutes of trying to evaluate this stuff! But I do still get the impression that generated Noise seems to inherently add more sound level, per note, than regular musical tones do. Even when seeded as here.

By messing with the seeding you can certainly hear when two noise trains sync up - they clearly go into mono for starters - but I can't say that the 'addition question' changes noticably either way, it's most weird. (The true-stereo effect is a striking feature of this dem btw, like it!)

But I have to confess I can't detect any difference whether the 'new seed for every note' element is connected or not. But I'm thinking .. assuming the random sequences don't repeat until what .. 7 weeks on Tuesday? .. then is it not true that playing a new note is never going to sync with any other note anyway, with or without being re-seeded ?

Anyway, I'm nitpicking without a doubt, sorry . Lockdown insanity m'lud, as charged. And noise is usually added in small proportions anyway; I only spotted the (alledged) issue here because I was listening to the noise in isolation. A bit unrealistic!

H

[HughBanton]

I can't detect any difference whether the 'new seed for every note' element is connected or not

Oops, I've missed the point I think. The 'new seed for every note' catches the situation where, for example, you play 5 notes together (or very close together), where notes 1 & 5 will likely produce phasing effects if they are following an identical noise pattern. Fairly easy to demonstrate if you disconnect it and do that.

Meanwhile the noise over-addition thing leave me puzzled still. It evidently doesn't happen in real life, in real space, e.g. with a 30-piece violin section. Violins certainly have a substantial noise content but I'd bet that the spectrum remains much the same for 1 or for 30.

So I'm wondering if it's a speaker thing, when the noise elements are forced to mix. To investigate this I could set something up on 4 speakers .. I'll start a new thread if I get any where. QI, imho!

H

[martinvicanek]

Your observation about the first and fifth simultaneous notes interfering noise-wise when the seed input is left unconnected is spot-on. That is because I already supplied a different seed internally for the 4 SSE channels. The first 4 notes are assigned to the 4 SSE channels of an oscillator, no problem there. When you play the fifth note, FS' voice management will create a new oscillator object and assign the voice to its first SSE channel, that's where you can have interference with the first voice. Of course you know that.

If I understand you correctly, you say that even in the analog world, 30 violins will generate less than 30 times the noise power of one violin? Not sure, why, but as one slogan on KVR says: If it sounds right, it is right.

[tulamide]

If I understand you correctly, you say that even in the analog world, 30 violins will generate less than 30 times the noise power of one violin?

Not sure if it helps you, but I found this on the internet:

Each violin in a section will be out-of-phase with the others. The peaks and troughs of the various sound waves will not occur at the same time. Thus N violins sound about Sqrt(N) times louder than one violin.

[adamszabo]

I think what Hugh means is that the noise seed should be per note and not per sse channel. If you have 30 violin players starting at the same time, its human nature that they will never start at exactly the same time, there will be a subtle difference when they hit the strings, therefore each will start at a different "phase". So i think per note random seed value would solve the problem

[DaveyBoy]

As well as the differing start times I would suggest that because the violins won't all be perfectly in tune with each other they will also 'drift' in and out of phase as time progresses.

[HughBanton]

Martin has totally taken care of seeding by note && by SSE channel. It works perfectly, no problem there.

However, to my ears anyway, I still hear over-addition of noise when you build up a chord, so it can't be to do with correlation. Now that I'm alerted to this I've tried other types of noise source too, and they all seem to do it. If you have a tone plus some noise, the relative tone-to-noise ratio seems to change as you add more notes to a chord, noise getting louder than tone.

Since regular tones don't appear to exhibit over-addition, (so long as they're not phase-locked), I'm thinking it's perhaps just a feature of broad-band noise, so maybe you won't get this when it's filtered. Some experiments to do. As mentioned I'll also have the opportunity next week to try a four-speaker system, so I'm also wondering if I'll still hear over-addition if a 'four-noise-chord' is directed to separate speakers??

I'll report; if it turns out that my issue is a result of electronic mixing, as opposed to real-life acoustic mixing, that has some interesting implications ...

H

DaveyBoy - yeah, and violinists also habitually apply heavy vibrato most of the time, that's how a string section sounds so big & lush. Absolutely no chance of any correlation between two instruments in the group for more than a nano-second or so