Hacker News Reader

Comment by isoprophlex | Hacker News Reader

> To overcome this problem, the MIT team rethought the design of a thin-film loudspeaker. Rather than having the entire material vibrate, their design relies on tiny domes on a thin layer of piezoelectric material which each vibrate individually. These domes, each only a few hair-widths across, are surrounded by spacer layers on the top and bottom of the film that protect them from the mounting surface while still enabling them to vibrate freely.

Whoa, does this mean an array of individually addressable micro-speakers becomes feasible? Like pixels in a computer screen, send different signals to each tiny speaker? That would mean craaaazy spatial audio, if I'm not mistaken.

View on HN

Directional audio is produced by sound interacting with your ear, so this would not produce spatial audio (unless you would wrap your head in the thin film speaker). What you want is 1 speaker per ear and a software modelling of the HRTF: https://en.wikipedia.org/wiki/Head-related_transfer_function

That's not what he means, he means using wavefield synthesis which is actually trying to reproduce the physical sound field with speaker arrays, in that case you don't need HRTF processing because your ears are receiving the waves from the actual positions. It can work for a bigger audience in an larger area without wearing headphones and without using head- and position-tracking but it needs lot of loudspeakers for accuracy.

https://en.wikipedia.org/wiki/Wave_field_synthesis

http://www.eliasmerino.com/structural-wfs.html

https://audioxpress.com/news/holoplot-launches-wavefield-syn...

Yeah, instead of 5.1 or 7.1 surround you could have 500.1 or 1000.1 and have soundscapes that are virtually audibly indistinguishable from actually being in the location.

Maybe overkill for movies but for VR immersion that could be fantastic, especially if the modules could be rigged to use Bluetooth 5.0 so that they would only need to be paired to your computer and mapped out in software the way multiple monitors are mapped out in windows.

For VR you could just use a pair of headphones with HRTF. This is what that can sound like:

https://www.youtube.com/watch?v=WYdIidUIbAs

Those are great for spatial audio, true, but they are simply not capable of making sound seem like it is directly in front of me. Some sound comes in through the nose, not a lot obviously, but enough that headphones are dramatically bad at generating sound in front of you.

Plus, given enough calculation and precision, you could create areas that are nearly sonically separated from each other using wave cancellation, and as people move through the space it would change the sonic characteristics of the space and create new vibes.

There's a lot of untapped awesomeness in the audio world, we just need to tools to craft them

Check out the video I linked then

I did. It never sounds like it's in front of me. It's not capable of making sound in such a way as so the sound feels like it is directly in front of me, it's always to the sides or behind me.

Maybe other people feel it, but I do not.

Well, I tried with some one else who also didn't feel any spatial effect. I think it depends on whether the binaural recording was done with a head/ear shape similar to yours. So you'll just have to take my word for it that the audio positioning is very accurate. A good HRTF implementation should be calibrated to your head, Sony was talking about that when discussing the PS5's spatial audio: https://youtu.be/ph8LyNIT9sg?t=2979

I do feel it behind me, around my head, but I don't hear it coming from in front of me. It never crosses the threshold to be audibly in front of me because microphones cannot send sound in through my nose or vibrate my face.

The sound can sound the same as if it passed through your nose and face though. The model head they recorded this with had a different nose and face than yours, but you face could also be modelled in an HRTF.

I think op is talking about beamforming when he talks about directional audio, where a large array of speakers would allow you to e.g. send audio in one direction only. The more speakers, the tighter the control.

I think OP means e.g. a video call where each person's voice is played back from where their mouth is displayed on screen.

I don’t think OP was saying that, but you are. A video layer could be built on the gridlines between the speaker domes. Might need to be quite a large screen for the spacing to work out right. Not for high def video, but good for large displays like talking billboards (ugh, actually).

urgh actually, indeed

Or as a Center Speaker in a cinema setup where the voice comes not from "the middle" but out the mouth of an actor

That would be great, but doesn't really need new speaker technology, at the scale of cinema screens - not like you need pixel granularity - just the encoding to support it. When I had a Saturday job at a cinema a couple of the screens had the centre speakers behind them iirc, it just wasn't an individually addressable array.

Having said that, I suppose Atmos already implicitly supports it, isn't the idea of it that you can put speakers wherever you want and it remixes appropriately? So if your recording was sufficiently granular that dialogue isn't just all vaguely 'at the front' it could already split between say five (corners & middle) 'centres'?

> isn't the idea of it that you can put speakers wherever you want and it remixes appropriately?

No, Atmos has some tolerance for speaker placement but the general positions are somewhat strict (i.e. 1 Front Center, 2 Stereo Fronts, ...). The idea of Atmos is that given a correct sound setup, a sound can be positioned as an Object in 3d Space within the sound bubble. And that only works to a certain degree. If you want a strong 3D sound effect, the Object pretty much needs to be moving (bullets swishing beside your ear). Static Objects are still somewhat restricted to the edges of the bubble or at least i can hear the 3d effect breakdown when static objects are placed in 3d space as opposed to moving objects

Oh, ok, I see, thanks for the correction.

If you put headphones on you could do that with software without needing special speakers.

You would also need an IMU in the headphones to modify the phase when the listener adjusts or her head. This is an important part of how we discern forward from behind.

Apple already does this via spatial audio for AirPod pro.

And for some reason using that is a pretty weird experience. I've been so conditioned not to expect it that I actually turn it off to make it feel "natural"

It’s a weird experience, a bit uncanny valley like. I was using while watching video, and it’s somewhat eerie to turn my head and still have the audio from the earphones “pinned” to the screen.

While that's certainly one way to synthesize directional audio, the far more common way is to actually have the audio come from the direction it's supposed to and let your ear do its normal thing to sort it out. Imagine you had an entire room wallpapered in individually addressable tiny speakers, you could actually project sound from any angle. The added benefit being that it would work for more than one person in the room. we've gone from 2.1 audio -> 5.1 -> 7.2 -> atmos 11.2. Why wouldn't we want to go to 50000.2 audio as the next extension?

Atmos is not 11.2! Home theater Atmos is 12 statically positioned streams, usually 7.1.4 (horizontally_emitting.low_freq.vertically_emitting) and up to 20 dynamically located streams. The audio is then rendered for the speakers configurations. Highest-end decoder can theorically support 24.1.10 but I never saw a decoder support more than 32 channels.

My receiver has 13 outputs (11 amplified, 2 sub at line level) but I use a 5.1.2 configuration, in a relatively small listening room, and I don't know where I could add more speakers without rebuilding the walls and the ceiling.

That would require a lot of wire and a lot of amplifiers. I bet we'll see an installation or two of the proposed design for proving it can be done.

Still, I bet most spatial audio systems will use software and fewer drivers ( potentially these drivers) to create the intended effect.

It just costs too much to wire all them up.

But at thousands of speakers you don't need as much amplification for each wire and considering how these flat speakers are produced, it's not too unlikely that we could eventually be able to embed chips/controllers into then just like we do with current display tech.

With that setup you could encode hundreds of channels in a single wire and each embedded controller would be responsible do decode it's addressed channel(s) to send to it's respective "speaker(s)". If the signal produced isn't high enough, you may also add in some small amplification stage in the embedded chip.

Very interesting. While I disagree with the implementation details, you do bring up a great point I completely missed. Embedding electronics into these would be trivial, thus enabling some form of smart communication removing the need for discreet amps and removing most of the labor involved in installation. Man I love HN.

Angular resolution is all very well, but what about spatial resolution? Even from the same compass bearing, a noise coming from a mile away sounds very different than someone whispering in your ear.

Playing with phase, I imagine that gangs of speakers could be used to produce bass as well. Don't need to stop at x.2...

Physics/EE students please correct me, I've mostly forgotten this stuff.

Assuming the speaker is many wavelengths away (in the "far field"), the distance between individual speakers needs to be larger (comparable with the wavelength) to make a difference in the radiation pattern in the far field. Speakers which are closer together only make a difference in the near field (meaning the listener is within a few wavelengths of the array).

If you want to understand this, look into https://en.wikipedia.org/wiki/Phased_array

While there is certainly a difference between near- and far-field approximations, the short answer is that no, the speakers can be closer than one wavelength together in order to have a steerable far field. In fact, the link you provided has a 1/4 wavelength spaced array as one of its first examples. You may be confused here because typically it is harder to make lower frequency waves from a single emitter more directional -- but that has more to do with waveguide and aperture geometry.

Looking at the equation for the radiation pattern of the phased array, the angle dependence goes like sin(pi * (N * d/lambda) * sin theta). If N * d (i.e. the size of the array) is much smaller than lambda, there's no interference pattern.

To be more precise, for the radiation pattern to have a null, N * d must be larger than the wavelength.

Yes, but I think that's why you're confused. The distance between the individual speakers would be d, not N*d. Because your original comment was about this spacing, that's what I addressed in my answer.

Obviously a larger effective aperture (either physical or synthetic) would be more effective at beam steering.

It seems like there's no reason for this use case for the speakers to be individually addressable, and that would probably add significant complexity.

That would be pretty nuts, sending out a digital signal and have a built-in DAC for each emitter. Probably way too much bandwidth for any of the standard busses though.

As an example, a thousand channels of 16-bit samples at 48 kHz adds up to 768 Mbits/s, which is well within USB3 bandwidth many times over.

1000 channels is a rather small array, though. 5000 is still not too much.

A lot of the signal is going to be identical, so a hierarchical distribution with in-place modification based on location would be better than calculating the modification of the sound and sending to thousands of channels.

The thing is, don't we have a ton of higher bandwidth buses already available?

I can't imagine the audio interfaces being faster than HDMI 2.0, USB 4/Thunderbolt, DisplayPort 3, etc.

I was thinking more in the line of i2c, i2s, spi, uart, since I think all of those that you listed require very extensive circuitry to parse. (At least that's the motherboard makers' excuse to only include like 2 USB 3.0s and 1 USB 3.1 port and like two bazillion 2.0 ports)

Maybe you could do something like Holoplot[1] does with their full size beamforming speakers.

[1] https://holoplot.com/technology/

> speakers to be individually addressable

Modern audio codecs/standards (Dolby Atmos, DTS:X) have moved 'beyond' individual speakers, and are now using object-based sound:

> Audio becomes an object when it is accompanied by metadata that describes its existence, position and function. An audio object can, therefore, be the sound of a bee flying over your head, the crowd noise, commentary to a sporting event in any language. All this remains fully adjustable on the consumer’s end to their specific listening environment, needs and liking, regardless of the device.

* https://www.thebroadcastbridge.com/content/entry/16347/objec...

> Think about this: imagine the sound of a race car speeding around a track. You can see the car approaching in the distance, off on the right side of the screen. As it gets closer, it gets louder and zooms across the screen from right to left, with the resulting Doppler shift of the sound as it goes past you. It screams off the left edge of the screen and continues down the left wall until it disappears into the distance behind you.

> A sound designer could, in theory, pan this sound carefully from the Right speaker, through the Center speaker, to the Left speaker, and on down to the Left side surround and the Left rear surround before it faded out entirely. That would be the channel-based way of thinking about the task at hand.

> Alternatively, the same designer could associate the sound of that race car with locations (coordinates) that move smoothly across the front of the room and then down the left side of the room. It is the same sound, but now with metadata telling the playback system where it should be from one moment to the next. […]

> The second, object-oriented way, is scalable. It doesn’t care how many speakers you have in your room because it is not referencing a specific speaker – just relative locations. Importantly, these locations can include the space above you and around you, enclosing you in a “bubble” of sound.

* https://www.trinnov.com/en/blog/posts/what-is-immersive-soun...

* https://en.wikipedia.org/wiki/Dolby_Atmos

* https://en.wikipedia.org/wiki/DTS_(sound_system_company)#DTS...

Given the (x, y, x) co-ordinates of an audio signal/object, the codec algorithm figures out which speaker(s) should get what signal: if the system only has two speakers then you'll generally only have stereo, if you have a 7.1.4 setup (four overhead / in-ceiling speakers) it will probably be more immersive:

* https://www.dolby.com/about/support/guide/speaker-setup-guid...

> Modern audio codecs/standards (Dolby Atmos, DTS:X) have moved 'beyond' individual speakers,

What you're describing is representational. You still need individually controlled speakers to position audio in the room according to the positional metadata. The question of representation really isn't relevant to the question I answered, nor is it to my answer, but it's an interesting topic.

> Given the (x, y, x) co-ordinates of an audio signal/object, the codec algorithm figures out which speaker(s) should get what signal

...and for the signals to be distributed accordingly, you need to be able to address each speaker individually.

In the domain of representing positional audio, this is also nothing new. No one mixes surround sound in terms of speakers, and that has been the case for decades. No one would manually pan a sound "carefully from the Right speaker, through the Center speaker, to the Left speaker, and on down to the Left side surround and the Left rear surround before it faded out entirely". What object-based codecs bring to the table is that the positional representation is encoded in the data stream rather than mixed down to per-speaker audio streams during production, which means that the distribution and filtering can be tailored for each setup individually.

To a complete noob such as myself, is this somewhat the equivalent of 3D graphics and lighting or even stuff like raytracing?

After all, 3D graphics create objects and then model their interaction with the world.

> or even stuff like raytracing?

Someone correct me if I'm remembering wrong, "sound raytraycing" was a feature in the game Thief if you had a compatible sound card, right?

I (also not sme) think the best graphics analogy would be holograms in Star Wars - object is at point-in-room, and you can hear/view it there.

The big tech here is the ability to spacially position the audio in the physical space the speakers sit in, by automatically mixing it between the speakers.

Sorry if I misunderstood the question, or a smarter person answered while I was typing.

Could one use something like this to make a stealth suit thing that does noise cancellation like headphones do to completely mask the noises of the wearer? That would be cool.

> To overcome this problem, the MIT team rethought the design of a thin-film loudspeaker.

I wonder if the team has seen Sony's OLED TV's with Acoustic Surface ?

The screen itself is used as a large speaker, and its surprisingly good.

The Sony solution actually has subwoofers hidden inside for the bass. I’m still missing how this paper-thin solution could possibly produce any decent bass.

It absolutely can't. This is a tweeter design. Might be good as that. It's not going to have even enough excursion to do midbass or lower midrange.

Mind you if the description is correct, to hear it properly they need to adhere it to a physical object. Having it dangling in free air like that means it's a lobed omnidirectional radiator: opposite of a dipole, it's putting out the same signal to either side across the whole plane, and the edges aren't putting out anything, and the range isn't low enough to hear it side-on very well: highs are directional.

This is why when he curves it you hear treble louder: it's making a little dish aimed at the mic (roughly). You could easily make a tweeter for 'head in a vice imaging' where the curvature is such that it's aimed only at the ear position, for less near reflections off walls. As described, you'd always want to back it with a physical object.

In a large enough area it's a midrange driver with very high peak output in the highs. You could curve it the opposite direction to make the mids slightly dominate over the highs: slight off-axis will attenuate highs a little more than mids, and the differences in hearing position across the curved surface (it's a smooth radiator without any specific driving points) will cause higher frequencies to cancel, again reinforcing lower frequency stuff.

Reply on news.ycombinator.com