Hacker News Reader

Comment by shreyshnaccount | Hacker News Reader

I don't think it can be done.. active noise cancellation needs to be really really fast and is done using dedicated hardware afaik.

View on HN

We programmed this at a company hackathon a while back. We pivoted and presented it as an "open office eavesdropper" because our naive algorithm amplifies distant speeches pretty well unexpectedly. https://github.com/Quantisan/nullwave

Can't be done because the sound from source propagate as a spherical wave. To cancel it with destructive interference you would need to make a wave which focuses to the source and would only work if you are in between the noise source and the counter wave generator. Unfortunately to make a plane wave you would need infinite number of point sources or need the counter source further away. An alternative is to track listener and cancel the noise around that region but you would only be able to do that with very low frequencies with wavelength larger than distance between your two ears.

I suspect it could be done for sort of periodic noise (or possibly with a predictive net). I've always wanted to implement this on the web but never got around to it.

As far as I can remember, even some of the special-purpose-hardware based solution like you'd find in a pair of noise canceling headphones do better on periodic noises. Not needing to race the soundwave seems like it should make the latency problem much more forgiving.

And the most annoying noises tend to be periodic anyway.

I think you should give it a try... let us test it, if you can cancel out my air conditioner noise you'll be my favorite person...

True, and shows how far we have fallen in the war against latency.

In theory it should be possible with a standard sound card, or a sound chip in the motherboard.

I'm neither a physics nor sound processing expert, but it seems like the microphone creates the signal to invert, and the speakers emit the inverted signal before the initial sound hits your ear. So I reckon the sound must first hit the microphone, then the speakers, then your ear. With headphones, the speakers cover your ears and the microphone is outside the enclosure, so that's easy. In a larger controlled environment with reliable sound dampening on 3 sides, maybe? My inexpert intuition says that doesn't stand a chance of working in an open environment with one microphone and two speakers usually closer to each other than the user's ears. It seems like having a door on a building with no walls.

And for some reason I thought it required triangulation, in that noise cancelation can't be done with sound in a space because the inverse of the sound you hear is relative compared to the person a foot or two away from you.

It does only if the microphone is far away from the ear canal.

If it's close to the ear canal, all you need is low latency hardware (or hardware+software), responding in a few milliseconds thereabouts.

> is done using dedicated hardware afaik.

Agree with the first part of your answer, but not sure what you mean by this second part.

There is dedicated hardware for it, then there are solution using your generic gpu (ex Nvidia RTX Voice), then there are solution using your generic cpu (ex whatever the name of the one discord desktop client provides).

isn't that different? i thought that's noise isolation that runs using a Neural Net. i don't think it's as real time as you'd need for noise cancellation

I meant circuits specifically designed for noise cancellation. like in headphones

ASUS makes a noise canceling headphone adapter, and I have a chip built into my Asus motherboard that does it.

https://shop.asus.com/us/90yh02l1-b2ua00-ai-noise-canceling-...

there's not too much detail on that page but what I see is that it "reduces EMI interference" from a mic. that's a different thing than noise cancellation (remove the noise from your background so that you hear the audio better)

That appears to be a "noise canceling" microphone adapter. It doesn't even claim to offer headphone noise canceling, it's just for microphones. There are two dots that might possibly be microphones for active noise cancellation, but I suspect that they would be too far away from your face to be effective. Most likely there is just a DSP inside that tries to filter out background noise heuristically.

Yea. Relatively straightforward on an MCU that supports audio, and running non-OS firmware. I think anything that runs through an OS or schedule would introduce too much latency.

So, it can be on a generic MCU, but not one where the algorithm is competing for CPU time with other processes.

you prolly need a dac or smth, because embedded devices usually don't have the compute needed for real time noise cancellation

You could do it on STM32H7, and probably slower ones too.

Supporting audio kinda implies having a DAC...

Unless you really like the sound of square waves I guess.

How fast can it need to be? The output frequency would be below 20kHz so latency of 50us or so, probably too fast for a userland app but I'd expect it to be doable in a driver or in the OS?

I get a similar answer but through a wildly different path:

I think you have to detect the sound with a microphone, then come up with the canceling waveform before the sound hits your headphone drivers (as they will be generating the canceling waveform). Given the typical size of headphones, it would be something like

2cm/(34300cm/s) ~ 60us

Although, sound travels faster through solids, so might want to be a little quicker. And 2cm thick headphones would be pretty thick.

wonder if there's some kind of "virtual microphone" like device that let's you hear sounds further away than a mic in the headset. Google fu showed some ultrasonic virtual microphone but idk if it is useful here..

Hmm, could be. I don't know much about this.

I imagine you want a very good approximation of the soundwave that will hit the ear (so you can cancel it out very accurately). The farther that your 'virtual microphone' is from the ear, I guess the harder it will be to figure out exactly how the sounds it is picking up will combine to hit the ear.

I mean, imagine the extreme case -- your 'virtual microphone' is (virtually) sitting in the path of some directed audio beam which won't hit the headphone wearer at all. Now you do the signal processing and generate a signal to cancel out that directed audio beam, which, because the user isn't in the path of it, causes them to hear it!

yeah, it would have to be super directional..

Interesting! I guess our approaches are linked by the fact that the size of our ears is connected to the wavelengths we want to pick up (or cancel out)… and to the size of headphones. :)

it effectively needs to be fatter than the speed of sound. sound needs to hit the mic of the noise cancelling device, its inverse needs to be found before the sound hits your ear drums so it can be played at exactly the same time. all this needs to happen in the time it takes for the sound to travel from the devide to your ears. super fucking fast.

Reply on news.ycombinator.com