Velvet Noise

[martinvicanek]

I have been reading about a specific type of white noise called velvet noise. Basically it is a train of impulses at random positions. The impulse height is either +1 or -1, the sign being random:

Velvet Noise example

capture.png (8.84 KiB) Viewed 22886 times

There is one free parameter T which is the average spacing between impulses.

Curiously, the human ear perceives velvet noise as pleasant and smooth compared to ordinary white noise, which sounds more grainy. Velvet noise is therefore preferred in audio applications.

Below is a ready-to-use velvet noise implementation along with some demos. Check it out!

VelvetNoise.fsm

(54.99 KiB) Downloaded 1237 times

I have also tried velvet noise in my MVerb 7B and was quite happy with the improvement!

[guyman]

I'm dumbfounded. It has a constant volume, so it's the same as using white noise in application, but somehow "sounds" better? Just when I think I'm on your tail, you go a step further at digging up gems. Now I gotta go back and rework my machines.

Thanks for sharing !

[adamszabo]

Sounds very pleasing indeed!

[Spogg]

Quite a remarkably different sound to white noise.
When I read your explanation my intuition told me it would sound more harsh somehow, but clearly not when I heard it. Maybe it’s the lack of random intermediate levels you get in white noise that explains the smoothness.

I did an A/B comparison with white noise and tried to get a handle on “how” it sounded and then I realised it sounded like a big waterfall, in comparison with the very “electronic” sound of white noise which sounds like nothing occurring naturally. You could call it a Niagara oscillator!

Next time I need white noise for something I’ll try this.

It’s lovely, and thank you so much again Martin.

[tulamide]

For those who want to further research on the topic, velvet noise was invented in 2007, by Karjalainen and Järveläinen, who used it in a new approach to reverbaration, because sparse +1 or -1 values are easy to convolute by (most values are 0, which don't need any convolution to be done). It is now called OVN (original velvet noise), because since then there have been many refinements, variations and other mechanics (mostly regarding the RNG). Velvet noise operating in the frequency domain is called FVN. Velvet noise is nowadays also used heavily in speech synthesis.

A study found out that velvet noise is perceived as smoother versus gaussian white noise up to 3000 pulses per second, with the perfect spot at 2000.

Related to OVN, several other sparse ternary noise sequenceshave been proposed, including Additive Random Noise (ARN),Totally Random Noise (TRN), Extended Velvet Noise (EVN), Random Integer Noise (RIN). ARN randomizes the spacing between pulses rather than distributing a single pulse per window.TRN has a random chance of generating a pulse of a random sign for every single sample. EVN takes OVN and restricts the sample location to only a portion of each window, enforcing a second level of sparsity. RIN is identical to ARN, although both the pulse offsets and sign are read from a precomputed table of integer random numbers rather than independent random numbers.

=======================================================================================
Personal note: I think velvet noise ia a psychoacoustic phenomenon. Remember, we hear air pushed or pulled by a membrane using a magnet (aka speaker). But velvet noise is mostly the membrane not moving at all, and sparsely pushing or pulling to the max. The brain however needs to fill in gaps, that's not a new discovery, but the very nature of the brain. It happens in all our senses, be it sight, audio, taste, feel. Our brain fills the gaps with what it assumes fits best. In that regard it is no wonder, that the sound is pleasant for us! It would also explain, why it is perceived less smooth, the more pulses there are.

[RJHollins]

Excellent stuff !

Thank you Guys

[Halon]

Never heard of Velvet Noise before. Thanks for this

[Duckett]

Fascinating stuff.. noise has held my interest since my early teens, when I'd first heard about various sensory deprivation experiments showing basically how the brain will create meaning when there is none provided. I'd adjusted the brightness and contrast of the white noise/static of a unused UHF channel on my hand-me-down black and white TV at night with the lights off, until the flicker/strobe rate of the static seemed to be the "right" frequency, and while I would sometimes see frantic pinwheel/"carousel" shapes come and go in the static if I watched long enough, there were no wild hallucinations or biological devolution a la "Altered States"

Thanks so much for the additional info and context, tulamide. Noise as relates to information theory and signal processing (outside of the horribly complex wetware of a brain) is worth anyone's time to investigate at least a little bit, IMO. There's sometimes nothing more delightful than when one avenue of inquiry leads to gaining knowledge in unexpectedly related areas.

[francoisreme]

Hi Gang.

Velvet noise is great to create basic reverb algorithm. Using a velvet multiplied noise with an exponential decay (let's say for 100ms) and creating a feedback from output to input, you can simulate an exponentialy decaying noise at low cost.

But it's not really convincing.

[martinvicanek]

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