The Future Of Sound Is Not Recorded. It is Computed.

This is a new sound synthesis technique and it  can look at objects in a scene without sound,   and after that… create these sounds. Wow! And now, first, here is a real life situation.   And now a simulation…wait a minute. Nothing is   coming. Why? Well, because this was the computer  simulation. When I first saw and heard it,   I couldn’t believe it. It looks and sounds so  real, I refused to believe that this was made   by a computer digitally. Yes, all the sounds here  are synthesized by this technique. And get this,   no AI was used whatsoever. All human  ingenuity. That is unbelievable. So how is that even possible? Well, what it does is that it looks at objects and  breaks them down into little lego pieces. Let’s   call them voxels. And then, it simulates pressure  waves sloshing around to create all this sound. Imagine the scene turned into two voxel molds  - one for the start, one for the end - and the   method smoothly morphs the air between  them while pressure waves bounce around.    Each cell has a tiny slider saying “I’m air”  or “I’m solid,” so when objects move or deform,   the sound updates smoothly instead of cutting  or popping. It’s like a DJ blending one song   into the next one, so you don’t even notice  the change. Unbelievable work. And this means   it does not just play sounds - it understands  the space they happen in. A splash near a wall   sounds different than one in an open field,  and this solver makes that difference come   alive automatically. Think about how many  game developers, film studios will want   this. No more hours of hand-placing sound  effects – the physics does the work for you. But wait a second…so that means that it can take  this geometry into consideration. If it didn’t,   these little M& Ms would sound like this.   But it is now closed between the hands,   so with this new technique,  it sounds like this. Yes,   we still hear it, but it’s a bit muffled.   As physics would say. Incredible result. And when you dig into the paper for details,  the achievements become even more impressive.    If that is even possible. Here are 10 things  I didn’t expect at all. Dear Fellow Scholars,   this is Two Minute Papers with Dr.   Károly Zsolnai-Fehér. Dr. Carroll. One, all this works with pre-recorded  sounds, vibrating shells, sloshing liquids,   even Lego bricks - all inside one unified solver.   You don’t need a bunch of different algorithms for   different kinds of interactions. This alone  can do almost everything. Absolutely insane. Two, it runs on uniform grids only, which  could also be thought of as a limitation,   but it actually makes it super GPU-friendly  and keeps everything simple and fast. And,   yes now hold on to your papers Fellow Scholars,  because thus they run everything on a single GPU,   and it easily beats a high-end multi-core  CPU, typical speedups are at around 140x faster,   reaching up to 1000x faster than  the old solver in some cases. From   just one paper to the next one, we are  jumping a 1000x. I can’t believe it. Three, even at low resolution, some demos like  the cup phone already run faster than real time.    Yes, finally! We’re on the doorstep  of interactive sound simulations. Four, it smoothly interpolates  between animation frames,   avoiding the “popping” artifacts that  plagued earlier methods. Like fading   one movie scene into another - no  hard cuts. It works really well. Five, it can handle crazy geometry changes,   like cavities opening and closing,  without exploding numerically. Six, it even simulates more than 300,000 of these  candy impact sounds, which is incredible. Though   not real time yet, it just needs about 15  seconds of waiting for 1 second of sound.

Seven, it solves the tricky problem where air  appears after an object moves - by filling in   missing pressure and velocity fields with a global  least-squares solution, keeping the simulation   stable. Like a puzzle that magically spawns the  missing piece the moment you notice the hole. Eight, it supports tiny point-like sound  sources for things like debris or splashes,   so you don’t need ultra-fine  grids to hear every little click. Nine, it can even add “phantom” geometry  - just math, not real objects. You can use   these to shape the sound to your liking.   That’s sound design with superpowers. Ten, for moving objects, it resets the  boundary conditions smartly so sounds   don’t suddenly pop when something enters  a noisy area. This is a subtle detail,   but it makes everything  feel physically believable. And perhaps the most exciting part  is that it is not far from real-time,   interactive sound synthesis. Imagine being in  VR, picking up objects, smashing them together,   and the sound you hear is computed on the fly  by physics. This could change how every movie,   game, and simulation sounds - no more  canned audio, just pure physics-driven   soundscapes. The future of sound  is not recorded - it’s computed,   and it’s going to be spectacular. Once again, no  AI needed. Just think about how crazy that is. And once again, almost nobody is talking  about this work. I don’t think you would   hear a word about this incredible  paper anywhere else. Only here,   so if you would subscribe, like and leave  a kind comment, I would be very grateful,   and you would get to hear more  about amazing works like this. Code is available, dataset  is available. All free of   charge. It is incredible. What a time to be alive!