Don't worry everyone, everything I talk about in this video is 100% applicable to Engine Simulator and needed to get done at some point. I figured it was best to do this in Steam Engine Simulator since it will soon have a major DLC release and has an active player base on Steam. Hopefully you all enjoyed the video! Thanks again to today's sponsor, Brilliant! To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/AngeTheGreat . You’ll also get 20% off an annual premium subscription.
Speaking about sound generation. Are you familiar with this book: "Designing Sound" aspress.co.uk/ds/about_book.html ? It has a few examples of mechanical sound generations metal clicks, bell sound - that kind of stuff. (aspress.co.uk/sd/). It has even an engine sound generator .
He took the engine and made it sound like a fudging 4-8-4 Nothern type locomotive. Maybe we did get an official train for the dlc after all. (Even if this isnt the dlc and is just an update)
What's crazy, is *_both_* sound good, even when compared back to back! Or at least I thought so... 😊 But yea, the new sound is definitely a marked improvement.
I never realized just how difficult simulating the sound of a steam engine would be until these devlogs and the devlogs for Century of Steam. In that regard, pretty incredible that both games have come to simulate them so well.
I think it could be worthwhile to remember what the components of the steam engine are made of in the real world because the material will have an effect on the sound produced. Scraping metal doesn't really make a white noise sound, it has weird harmonics and other vibrations which give it a distinct character. I'm a musician, so maybe I'm just hyper-aware of this stuff, but the low-passed white noise sounds very weird and not like anything you'd hear in real life.
I've built engines and have spun them over by hand while they're off. It very much just sounds like a normal sliding sound. There are definitely other forms of mechanical noise and the exact frequency spectrum of that noise might differ from exactly what I'm using here, but this is just a first approximation. Thus far I've been unable to find any research about the exact frequency distribution of sliding noises for different materials.
@@AngeTheGreatby looking at the new visuals of the brake, it seems it uses rubber, if that's the case, it would make a very different sound when braking. Also, maybe taking the materials properties into consideration to generate the friction sound, is way more difficult
@@AngeTheGreat That’s very interesting, I suppose my experience has been the resonances of various metals, resonances of the whole system, and the nonspecific distortion of other mechanical noise. If you’re interested in emulating the internal resonances of metal components, I believe it can be calculated with the general rules for acoustic analysis, but I am not sure exactly where to find specific resonance tendencies of the atomic structures of different metals.
Yeah, usually metals have a quite annoying and high-pitched whining or squeaking sound. Perhaps this is caused by some resonance mode and the stick-slip effect, a bit similar to how a violin works?
Bro casually unravelling a bunch of black magic to get awesome physically based sounds, it's like when physically based rendering came into full force in game engines. Congrats. I hope you might be able to work with BeamNG to get some form of this tech in their game eventually, assuming you both appreciate the partnership. Would be an absolutely masterful tech demo.
Its probably been recommended already, but you should look into the techniques used by convolution reverb audio plugins. They can handle minute-long impulse response without a bit CPU hit, such that you can run several instances (though they still are not the cheapest thing to run). I believe the technique is call Overlap-Add (or Overlap-Save) FFT Convolution, where both the input audio and the impulse response are processed in chunks and in parallel, and the algorithm complexity is reduced from O(n^3) to O(n^2) by distributing the work over time. The autoregressive solution to combat audio glitches is very interesting, and i think the audio world (and at the very least I myself lol) would love a more detailed explanation of the process. I can see game engines and maybe music applications benefiting from this, depending in how applicable the solution is to music.
Yeah, there are a few options, but using a frequency delay line is a pretty simple. In overlap-add you segment up your output into blocks, calculate their FFT, convolve in the spectral domain with a multiplication, and calculate the inverse FFT, then perform the overlap-add with the last block. An FDL is just the observation that the signal and the IR are interchangeable in the algorithm, and you can in fact chunk BOTH the signal and the IR. So as the blocks of sound come in you FFT those and put them in a circular buffer that you convolve with all the IR chunks. Now the cost to process the reverb is on small FFT/iFFT and a bunch of multiplies once per block instead of per sample. My crap home baked implementation of this was ~50-100 lines of single threaded, scalar C code, and handled a multi-second reverb on a Pi 4 with low single digit CPU usage! Neat! If you want to go _really_ crazy with performance, you can make a hierarchy of FDLs of increasing block sizes. ericbattenberg.com/school/partconvDAFx2011.pdf
Could be cool if you also added ringing to the friction noise to make different parts have that 'metallic' clang to them. Also it could be fun to add backlash noises when forces change direction in linkages and stuff. No idea how computationally expensive it'd be, but it shouldn't be bad on the surface of it.
arguably using a different convolution reverb for every part is actually pretty realistic, given that every part is going to be in a slightly different environment and have its own internal resonances, which would involve taking apart an actual engine and hitting every part with a stick to capture its resonant impulse. this could be aproximated with possibly less computation by using an inharmonic FM synth to make metalic "ting" tones and then use those to create filter functions by aproximating their spectra. also the brake pad could use some heimholtz resonance to create the sqeualing sound accossiated with real brakes.
In some sense, yes it is more realistic, however, I'm not referring to minor tweaks to each impulse response. I'm talking about certain parts sounding completely wrong without the impulse response applied which isn't right. Fine tuning the base impulse response for individual parts makes sense but that can't be done until the base signal is at least reasonably correct.
This almost feels more like a 300-level college lecture than a devlog But that's awesome! It's crazy how much work and thought and science you're applying to this project.
Yeah the main part of his work is related to understanding concepts of engineering, physics and math. Programming ist just a tool to implement these ideas.
There's an old audio VST developer called xoxos, he made a few physical modeling inspired synthesizers (at least), one of which is devoted to stick-slip friction models I believe just called "friction". They have some great presets for various mechanical objects like doors creaking or a chair sliding against a floor. You couple that with some resonance models and you can get a huge amount of depth beyond just filtered white noise. I'd highly recommend you check it out!
Funny it's just like improving graphics over time, when it first came out i thought it sounded epic. but now it sounds even better, i can't believe I thought the old sound was so great haha.
This feels like the same kind of leap in tech that real time ray traced lighting was. Really awesome to see how you’ve worked through each issue methodically.
A cool feature would be something like a listening tube, you can drag it around and hear what the machine sounds like at particular positions instead of as a whole. Could be useful for debugging too!
@@TaylorfromPapaLouie charade you are, who said I jest? I of course look forward to the future where all sounds may be faithfully recreated via physical means
@@TaylorfromPapaLouie There's already a pretty rudimentary vocal tract simulation called Pink Trombone you can find online, it's far from capable of speech, but it's fun to play around with.
Sweet! It's amazing the stuff you're working on. For too long, simulations have ignored how important sound design is, and how everything shouldn't just be sampled. Incredible stuff.
Reminds me of my days playing Quake on a 486DX/33. It ran so poorly that not only was the game a slideshow, but the sound broke up as well. If you want to try it for yourself on a modern machine, set DOSBox to run at a fixed 10-15k cycles. Around the 20k mark, the emulated PC is fast enough that the sound stops breaking up, though the graphics are still a slideshow.
I could see this entire simulation and its audio be usable as a means to generate the sounds of, say, a steam loco cab simulator in a museum or the like. If the control inputs were adedd via audrino sensors, a person pulling the whistle cord of a loco in said museum would have realtime audio of the whistle in the sim played throgh speakers in the cab. As someone working with steam trains, I think your simulation is very well-made and sounds very good! 👍🚂
I've said this months and months ago on one of your videos! People don't realize that a big part that's lacking in the realism is every other sound that *isn't* coming out of the exhaust :D
Absolutely incredible work! Several times throughout I was thinking "yeah, but what about X...", and almost every time you already thought about it. Regarding convolutions for every part, a thing I felt was missing was the whine of f.ex a brake. The noise is good, and the low-pass is exactly what I was thinking (although, is the filter too steep? It sounds a bit dark to me IDK), but the "individual convolutions" would be the parts themselves resonating, no? Calculating resonating frequencies of metal parts seems doable. Love the thinking in the distance parameter! A thing that's also pretty important to how something sounds up close vs from afar is panorama. Up close, you would be able to hear that the piston and the wheel are in different places in your field of hearing, but from afar the entire machine becomes monophonic. Man this is so impressive, it really sounds dynamic and has a real feel to it.
If I got it right, the individual convolutions are because the different outlets' sound went through different parts, so the convolution for each should be different. You can aggregate for the path they took and have a single one for each outlet. That is, ignoring the vibration of the parts themselves.
I cant imagine how good this would sound with the resonance of the objects like the brakes resonating with the wheel when it comes to a stop or the joints squeaking when it starts to spool up, etc Insane job
9:36 those squeaks sound so real. Not the popping, but as the engine winds down the squeaking. When I was playing around with Stirling engines I heard them all the time, and no amount of oil made them go away. They're kind of annoying so I'm not sure you want them there, but even if they are artifact errors there's a similar effect to them in real life.
I can add a bit of complexity you don't take into account; resonant frequency - most notably in metallic components or parts. The friction between the brake pad with low resonance (usually made from a compressed mineral like asbestoses), against a steal wheel which has a prominent resonant frequency based on it's mass, shape and size, and material hardness (steel is more resonant than copper). All this results in break "squeal". As a matter of fact every metallic component that experiences friction during operation has a resonant frequency that varies in amplitude (not frequency) with both the velocity and pressure of the friction. The resonance of each part is determined, like the wheel, by it's size, shape, mass and material hardness. Find a video of a B-17 engine start up. You will notice that at low RPM the sound produced has a heavy metallic, multiple frequency "clanking" sound. These metallic sounds disappear as engine RPM increases because the friction speeds increase resonant frequency harmonics beyond audible ranges and/or become overpowered by the exhaust drone. This specifically is a project I've been working on in VCV Rack modular synthesis program for a 3D animation of a B-17 from startup to cruising flight.
Incredible update as usual. The transformation is night and day, and it's extraordinarily satisfying to see application of physics principles translating to physical realism. Regarding popping/clicking issues, perhaps you could add an additional layer of latency between the simulation and the UI. If the UI was live, but the simulation ran even up to a few 100ms behind, the user would be unlikely to notice this difference at all, but popping issues would be greatly improved. This could just be provided as an option for users suffering these problems.
There is already latency applied and it can be set by the user in the game's options menu. Faster systems can handle very little latency but the game by default has the latency set to around 150 ms which is enough buffer for most slow downs. Thanks for the suggestion!
Every mixer in an audio-workstation taking multiple asynchronuous sources could make use of that "make zeroes go away by predicting some fake audio" hack. imho. Some great techology in here, bravo!
Wow! When you put all the elements together, I was taken right back to that time I stood next to a pumping engine many years ago! A sound I had forgotten, brought back - that's how you know you got it at least mostly right! The hiss of the valve you actuated in particular was uncanny!
Ange this is incredible, I've spent most of my life around Steam Engines, I could close my eyes and it would be like im right next to it with how realistic this sounds, you've done seriously well!
I am a train nut myself and this is insane. One thing that would be fun is to add different size and chime/note whistles. Which in the real world changes the sound and pitch. Just a fun idea of course, but i would love to see it done because the whistle simulation sounds outstanding. Cheers!
Bro, I'm more on Software/API side of the programming world and the simulation engine you're building is insane on all side. You need such deep knowledge of your programming language plus insane mathematics skill that I don't even know where did you get all this knowledge from. Incredible work.
That is absolutely amazing. I was away from my screen when you played the full demo, I could only hear it. Sounds like a massive steam train full of heavy parts.
I'm going to use this to teach my child more about engines this is amazing in every possible way i can't thank you enough. amazing work. as soon as I have the bandwidth I'm going to become a patron.
The incredible difference in realism in the side-by-side comparison is jaw dropping! You are doing great work, and you really are hugely inspiring :) Making me want to go and continue learning more math and physics
Amazing work as always! I'm very excited to see this attention to audio. The autoregressive model audio still sounded like clipping to me, just less pronounced. Maybe it's not in phase or something?
There are limits to how much the audio can be corrected when there is literally no audio present. It really doesn't sound like clipping or clicking and you can confirm this using a spectrogram and comparing it to actual clicking. There are no discontinuities in the signal whatsoever. Every other approach I've tried results in much more obvious artifacts when audio is dropped. In regular play, most errors with audio prediction enabled are reduced to just subtle blips or frequency detail loss. You can probably analyze the signal from this video yourself in a spectrogram and you'll see what I mean.
I love these videos - not because I can pretend to understand any of the magic you are performing in code, but because you lay out so well how you identify and approach the problems you are solving. It's a masterclass in how to do iterative design.
As I've seen, you have the best mecanical engine simulator engine, it would be very nice to see a "train steam engine simulator" with all that real time response of mecanisms and sound. Keep the good work!
i think i said this on an older video but god imagine the power if sound physics was applied to making mechanical sirens. with a rotor and stator and everything. not just hand cranked but like full on electric. electromechanical. hearing a giant steam train plowing through the tracks is already an insane feat and it's crazy to see where these sound physics have gone in the past year or two
Honestly it is INSANE. I cannot even begin to fathom how would I approach coding this. If i was blindfolded and listened to isolated sounds of real engine and this simulation I probably wouldnt be able to tell the difference. Wow
oh man that full demo comparison really makes the difference in realism noticeable. gotta love steam engines. during my vacation I went to a technology museum and they had these humongous steam locomotives there. truly magnificent machines
Love the screw reverser and how the simulation actual shortens the exhaust beat. It would be so cool to see your simulations be used in train sims for accurate steam noise and whistle quilling. It’d be cool to hear some brake squeal and added rod clank when the reverser is a bit more centered or off throttle completely as well… unless there is and I didn’t notice
I looove your content and have been tracking Steam Engine Simulator since launch. Seeing it progress into a full, super realistic simulator is so awesome. Keep it up!
It's so interesting seeing the entire process of this, I remember when it first started you had to explain to people that you were doing your best, and that what you have put together (at the time) sounds great already. But now fast forward to the present day, the leap was huge. And the people complaining did actually have a point.
I don't believe the people complaining actually had much of a point. They stated obvious things, contributed zero insight and ultimately changed nothing. Their actions are equivalent to me going back to 1980 and complaining that the CPUs are too slow. Sure it might be true and maybe if you fast-forward to 2024 and look back on my complaints, you could argue that I had a point, but I was ultimately just stating the obvious: that could always be better. I don't believe that's a very meaningful contribution to the discussion.
@@AngeTheGreat you definitely have a point, I'm here watching on the side lines, I always thought this was a great project, critics on here are way too harsh for no reason. And I do remember you explaining that you knew that there were issues and its not perfect, I appreciate your total honesty throughout this as well.
I checked your channel 2 days ago to see if there was any update and realised the original Engine Simulator video was 2 years old (Happy Birthday!). I remember when I first saw it I thought it would change everything, it could be a revolution for racing games, animation and probably a whole lot of other fields I didn't knew about. 2 years later I still believe Engine Simulator is one of the best project I saw on UA-cam and I'm really thankful for the time and effort you put into this project and I can't wait to see what's in the future. The second half of this video made me time travel to the industrial revolution 10/10 would steam again
Ange, I haven't watched this video, and I don't always get a real kick out of your videos, but there's a very simple reason I click on almost every one of them: you put a lot of effort into these projects and the videos, you want to make everything right, and it's all a very difficult challenge but you come through with results every time. I respect your fight, and I'm sure this stuff will be adopted into video games and will give them that next level of realism. I hope you're getting a solid piece of the cake from your industry partners, and that the industry will know your name. Keep it going and don't listen to the haters. Lastly, congrats to a very well deserved 200k. Edit: the audio improvement is insane! If the steam engine was built into a locomotive, you could also implement doppler effect for bystanders.
Wow, SES really went from a tiny steam engine running in an anechoic chamber, to a steam train chugging along the tracks in the city, and all in one video.. impressive.
Sounding great. It is very normal to band limit the audio with a very steep filter. E.g. 60th order IIR lowpass cascade just below nyquist is quite normal in hifi products. Although probably more typical is an FIR filter. Also, you could apply some dither during the creation/conversion of the audio. That would remove quantisation noise. The lowpass filter you use for distance might sound better if you used a first order lowpass filter, IMO it's a little too steep. Finally, if you plan to modulate the filters input parameters quickly then don't use biquads, use a state variable filter (I think sometimes called a "zero delay feedback filter")
The high-pitched sounds at 9:22 are what I thought it should have a bit of for that imperfect realism. It reminds me of a Semi truck's brakes, or just a misaligned motor. It's so cool that literal inaccuracy caused a realistic simulation of noise from inaccurate parts lol.
I've seen quite a few of your videos now and your work is absolutely breathtaking! looking forward to watching this video a little later on this evening.
It would be really cool to have some resonance in the system. Even if it only works for the brakes or for the main bearing when it's worn out something. This entire project is absolutely awe inspiring you are one of THE inspirations!
It does sound amazing but what papers would there be to write? He didn't really invent anything new just optimized existing techniques to work for his use case.
@@theairaccumulator7144 Firstly, most research papers are not particularly groundbreaking. Research papers vary all the way from being summaries of existing research, optimizations to current techniques, all the way to groundbreaking discoveries. It should be noted that in the world of computer graphics and audio, a lot of the research really is just the application of known physics to software-based problems. Secondly, I covered the updates in a very simplistic way in this video so I have no idea how anyone could determine whether or not the results are worth publishing. In any case, I'm not really interested in writing a paper on any of this and for the most part it is mostly applying known physics to a software-based problem. Based on my experience in actual software engineering research, a paper of some kind could definitely be written about this though.
@@AngeTheGreat : Having been on a paper review committee for a computer science conference, I can say that the very simplistic overview you've done in this video is already probably at the "publishable" threshold, so I'm sure the full work absolutely is. :) More generally, I'd point out that one shouldn't underestimate "optimizing existing techniques to work for a new use-case." In many cases, the optimization is the difference between whether or not it's possible to make the techniques work for the new use-case at all, and showing that it's possible is useful. It's likely that a lot of people who are doing audio work for video games are unaware of the details of how to do a physical simulation, and likely that they're even less aware of what is possible or not-possible to do with them as far as producing useful game audio (because very few people have even tried it), so a paper on that subject will tell them several new and useful things. Interestingly, it's also likely that most people who are doing detailed engineering simulations of sound generation also wouldn't have a good idea what's possible or not-possible as far as generating game audio. In engineering simulations, precision and accuracy are really necessary, because you're trying to figure out things like "will this jet engine exceed this airport noise limit?" and real-time computations aren't important. For game audio, it mostly just needs to "sound right", and it's fine to sacrifice a lot of precision to get the computations to happen in real time. An example of this is Ange's fix for computation lag -- that sort of thing would never be used in an engineering simulation, both because it wouldn't be needed and because it wouldn't be accurate. And so writing a paper about this is telling the audio-simulation community something new too.
Don't worry everyone, everything I talk about in this video is 100% applicable to Engine Simulator and needed to get done at some point. I figured it was best to do this in Steam Engine Simulator since it will soon have a major DLC release and has an active player base on Steam. Hopefully you all enjoyed the video!
Thanks again to today's sponsor, Brilliant! To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/AngeTheGreat . You’ll also get 20% off an annual premium subscription.
Speaking about sound generation. Are you familiar with this book: "Designing Sound" aspress.co.uk/ds/about_book.html ? It has a few examples of mechanical sound generations metal clicks, bell sound - that kind of stuff. (aspress.co.uk/sd/). It has even an engine sound generator .
D: I didn’t get notified from Patrion.
@@alobowithadhd6191 Discord might be a better place to get announcements. Patreon doesn't have the best interface unfortunately.
Have you tried talking to hyce about his simulator?
@@AngeTheGreat I’m assuming the dlc will be paid right?
insane how it went from sounding like a leaking compressor inside a tin can to something that really seems like a huge steam engine lol
He took the engine and made it sound like a fudging 4-8-4 Nothern type locomotive. Maybe we did get an official train for the dlc after all. (Even if this isnt the dlc and is just an update)
@@manitoba-op4jx Yeah the model does not fit the audio haha
I mean a steam engine is kind of a compressor in a can, just compresses a different gas and is in a different can
@@thepoglin8479 lol
its unclear what size this engine is meant to be
"The weird clunking sound update"
Next update:
"The oil pressure light update"
Lol
German engineering in a nutshell
Can't wait for the dead cylinder update.
@@yocapo32 Missfire in cylinder 5
@@last8exileECU cooked (lexus moment)
certified *check engine light* hood classic
finally, chugga chugga to go with my choo choo :D
Cool
And chi chungggg chi chungggg
This is the worst joke you’ve ever said in your life
Adds a full reverser and linkage system
Makes it sound great
Refuses to elaborate
:D
At this point having a full steam train sim is unavoidable!
@@quentin7763a cross over between Steam train simulator and engine simulator where you use a train to push a DeLorean up to 88mph
@@quentin7763 should do a collab with Derail Valley
@@quentin7763 nah, just steam engine sim, maybe N3V will buy that to use on the Train Sim World saga
he elaborated pretty well on the sound part lol
That is a massive difference in realism, it's so cool.
What's crazy, is *_both_* sound good, even when compared back to back!
Or at least I thought so... 😊
But yea, the new sound is definitely a marked improvement.
With the new sound effects, the engine sounds _absolutely massive._ Like an old locomotive!Now we just need a second cylinder.
I never realized just how difficult simulating the sound of a steam engine would be until these devlogs and the devlogs for Century of Steam.
In that regard, pretty incredible that both games have come to simulate them so well.
I think it could be worthwhile to remember what the components of the steam engine are made of in the real world because the material will have an effect on the sound produced. Scraping metal doesn't really make a white noise sound, it has weird harmonics and other vibrations which give it a distinct character. I'm a musician, so maybe I'm just hyper-aware of this stuff, but the low-passed white noise sounds very weird and not like anything you'd hear in real life.
I've built engines and have spun them over by hand while they're off. It very much just sounds like a normal sliding sound. There are definitely other forms of mechanical noise and the exact frequency spectrum of that noise might differ from exactly what I'm using here, but this is just a first approximation. Thus far I've been unable to find any research about the exact frequency distribution of sliding noises for different materials.
@@AngeTheGreatby looking at the new visuals of the brake, it seems it uses rubber, if that's the case, it would make a very different sound when braking. Also, maybe taking the materials properties into consideration to generate the friction sound, is way more difficult
@@AngeTheGreat That’s very interesting, I suppose my experience has been the resonances of various metals, resonances of the whole system, and the nonspecific distortion of other mechanical noise.
If you’re interested in emulating the internal resonances of metal components, I believe it can be calculated with the general rules for acoustic analysis, but I am not sure exactly where to find specific resonance tendencies of the atomic structures of different metals.
@@AngeTheGreat If you still have easy access to those engines and and a piezo accelerometer you could get some direct data to use.
Yeah, usually metals have a quite annoying and high-pitched whining or squeaking sound. Perhaps this is caused by some resonance mode and the stick-slip effect, a bit similar to how a violin works?
Bro casually unravelling a bunch of black magic to get awesome physically based sounds, it's like when physically based rendering came into full force in game engines. Congrats. I hope you might be able to work with BeamNG to get some form of this tech in their game eventually, assuming you both appreciate the partnership. Would be an absolutely masterful tech demo.
I have to assume beamng being one of his main supporters means they're pretty keen to get his tech into their game
Its probably been recommended already, but you should look into the techniques used by convolution reverb audio plugins. They can handle minute-long impulse response without a bit CPU hit, such that you can run several instances (though they still are not the cheapest thing to run). I believe the technique is call Overlap-Add (or Overlap-Save) FFT Convolution, where both the input audio and the impulse response are processed in chunks and in parallel, and the algorithm complexity is reduced from O(n^3) to O(n^2) by distributing the work over time.
The autoregressive solution to combat audio glitches is very interesting, and i think the audio world (and at the very least I myself lol) would love a more detailed explanation of the process. I can see game engines and maybe music applications benefiting from this, depending in how applicable the solution is to music.
Yeah, there are a few options, but using a frequency delay line is a pretty simple. In overlap-add you segment up your output into blocks, calculate their FFT, convolve in the spectral domain with a multiplication, and calculate the inverse FFT, then perform the overlap-add with the last block. An FDL is just the observation that the signal and the IR are interchangeable in the algorithm, and you can in fact chunk BOTH the signal and the IR. So as the blocks of sound come in you FFT those and put them in a circular buffer that you convolve with all the IR chunks. Now the cost to process the reverb is on small FFT/iFFT and a bunch of multiplies once per block instead of per sample. My crap home baked implementation of this was ~50-100 lines of single threaded, scalar C code, and handled a multi-second reverb on a Pi 4 with low single digit CPU usage! Neat!
If you want to go _really_ crazy with performance, you can make a hierarchy of FDLs of increasing block sizes.
ericbattenberg.com/school/partconvDAFx2011.pdf
Could be cool if you also added ringing to the friction noise to make different parts have that 'metallic' clang to them. Also it could be fun to add backlash noises when forces change direction in linkages and stuff. No idea how computationally expensive it'd be, but it shouldn't be bad on the surface of it.
Was gonna mention the ringing thing, brake rotors and drums tend to ring at lower pressure levels.
Considering he said at the beginning that friction was only the easiest, I imagine the next video could be about impacts like this
arguably using a different convolution reverb for every part is actually pretty realistic, given that every part is going to be in a slightly different environment and have its own internal resonances, which would involve taking apart an actual engine and hitting every part with a stick to capture its resonant impulse. this could be aproximated with possibly less computation by using an inharmonic FM synth to make metalic "ting" tones and then use those to create filter functions by aproximating their spectra.
also the brake pad could use some heimholtz resonance to create the sqeualing sound accossiated with real brakes.
In some sense, yes it is more realistic, however, I'm not referring to minor tweaks to each impulse response. I'm talking about certain parts sounding completely wrong without the impulse response applied which isn't right. Fine tuning the base impulse response for individual parts makes sense but that can't be done until the base signal is at least reasonably correct.
edit: it looks like helmholtz resonance is a different thing than "the resonance caused by helmholtz motion" which is what i actually meant
But that's what magic numbers are for... :P
I’d assume the squealing sounds are a bit similar to how a violin is played, through slip-stick effect and resonant modes?
This almost feels more like a 300-level college lecture than a devlog
But that's awesome!
It's crazy how much work and thought and science you're applying to this project.
I love it. That’s the kind of thing that gets me engaged in content.
Yeah the main part of his work is related to understanding concepts of engineering, physics and math. Programming ist just a tool to implement these ideas.
There's an old audio VST developer called xoxos, he made a few physical modeling inspired synthesizers (at least), one of which is devoted to stick-slip friction models I believe just called "friction". They have some great presets for various mechanical objects like doors creaking or a chair sliding against a floor. You couple that with some resonance models and you can get a huge amount of depth beyond just filtered white noise. I'd highly recommend you check it out!
Great suggestion, I'll check it out!
I love the new sound, it just sounds a lot more high quality than old SES! Good job Ange!
Funny it's just like improving graphics over time, when it first came out i thought it sounded epic. but now it sounds even better, i can't believe I thought the old sound was so great haha.
THAT WHISTLE IS AMAZING!!!
yess
This feels like the same kind of leap in tech that real time ray traced lighting was. Really awesome to see how you’ve worked through each issue methodically.
A cool feature would be something like a listening tube, you can drag it around and hear what the machine sounds like at particular positions instead of as a whole. Could be useful for debugging too!
This is an awesome idea
Maybe like zooming out making everything quieter, zooming in could "focus" on the visible parts, attenuating sounds from parts out of frame
The steam engine sounds terrifying now! Can you simulate the sound of the boiler exploding?
And simulate the vocal tract of terrified onlookers?
@@Mutantcy1992 You jest, but Vocal chord simulation will be very cool to see. Maybe a new way for AI voice sounds
@@TaylorfromPapaLouie charade you are, who said I jest? I of course look forward to the future where all sounds may be faithfully recreated via physical means
@@TaylorfromPapaLouie There's already a pretty rudimentary vocal tract simulation called Pink Trombone you can find online, it's far from capable of speech, but it's fun to play around with.
san antonios shopmen's locomotive explosion of 1912 would be crazy
Sweet! It's amazing the stuff you're working on. For too long, simulations have ignored how important sound design is, and how everything shouldn't just be sampled. Incredible stuff.
that last one is HUGE for crappy PCs. the clicking was nearly ALL I could hear on one of my computers. Thanks so much!
Reminds me of my days playing Quake on a 486DX/33. It ran so poorly that not only was the game a slideshow, but the sound broke up as well. If you want to try it for yourself on a modern machine, set DOSBox to run at a fixed 10-15k cycles. Around the 20k mark, the emulated PC is fast enough that the sound stops breaking up, though the graphics are still a slideshow.
Dayum, you don't really realise the improvement until you hear a side by side but when you do it is NIGHT AND DAY, insane work as usual!
i didn't notice how much different it was until you played the old one again, the new one is so much better holy shit
Those high pitched noises were a *very* accurate recreation of shoes rubbing on the floor of the school gym
I could see this entire simulation and its audio be usable as a means to generate the sounds of, say, a steam loco cab simulator in a museum or the like. If the control inputs were adedd via audrino sensors, a person pulling the whistle cord of a loco in said museum would have realtime audio of the whistle in the sim played throgh speakers in the cab.
As someone working with steam trains, I think your simulation is very well-made and sounds very good! 👍🚂
Every time I watch your videos I am in awe and, quite frankly, jealous of how near-flawless your end results are.
This step forward will revolutionise the way developers or modders make the engines sounds, for BeamNG, for example Great job.
Yeah, I can't wait for this kind of tech to be practical under the hood of basically any kind of game
I've said this months and months ago on one of your videos!
People don't realize that a big part that's lacking in the realism is every other sound that *isn't* coming out of the exhaust :D
Absolutely incredible work! Several times throughout I was thinking "yeah, but what about X...", and almost every time you already thought about it. Regarding convolutions for every part, a thing I felt was missing was the whine of f.ex a brake. The noise is good, and the low-pass is exactly what I was thinking (although, is the filter too steep? It sounds a bit dark to me IDK), but the "individual convolutions" would be the parts themselves resonating, no? Calculating resonating frequencies of metal parts seems doable.
Love the thinking in the distance parameter! A thing that's also pretty important to how something sounds up close vs from afar is panorama. Up close, you would be able to hear that the piston and the wheel are in different places in your field of hearing, but from afar the entire machine becomes monophonic.
Man this is so impressive, it really sounds dynamic and has a real feel to it.
If I got it right, the individual convolutions are because the different outlets' sound went through different parts, so the convolution for each should be different. You can aggregate for the path they took and have a single one for each outlet.
That is, ignoring the vibration of the parts themselves.
I cant imagine how good this would sound with the resonance of the objects
like the brakes resonating with the wheel when it comes to a stop or the joints squeaking when it starts to spool up, etc
Insane job
9:36 those squeaks sound so real. Not the popping, but as the engine winds down the squeaking.
When I was playing around with Stirling engines I heard them all the time, and no amount of oil made them go away. They're kind of annoying so I'm not sure you want them there, but even if they are artifact errors there's a similar effect to them in real life.
I can add a bit of complexity you don't take into account; resonant frequency - most notably in metallic components or parts. The friction between the brake pad with low resonance (usually made from a compressed mineral like asbestoses), against a steal wheel which has a prominent resonant frequency based on it's mass, shape and size, and material hardness (steel is more resonant than copper). All this results in break "squeal". As a matter of fact every metallic component that experiences friction during operation has a resonant frequency that varies in amplitude (not frequency) with both the velocity and pressure of the friction. The resonance of each part is determined, like the wheel, by it's size, shape, mass and material hardness.
Find a video of a B-17 engine start up. You will notice that at low RPM the sound produced has a heavy metallic, multiple frequency "clanking" sound. These metallic sounds disappear as engine RPM increases because the friction speeds increase resonant frequency harmonics beyond audible ranges and/or become overpowered by the exhaust drone. This specifically is a project I've been working on in VCV Rack modular synthesis program for a 3D animation of a B-17 from startup to cruising flight.
Incredible update as usual. The transformation is night and day, and it's extraordinarily satisfying to see application of physics principles translating to physical realism.
Regarding popping/clicking issues, perhaps you could add an additional layer of latency between the simulation and the UI. If the UI was live, but the simulation ran even up to a few 100ms behind, the user would be unlikely to notice this difference at all, but popping issues would be greatly improved. This could just be provided as an option for users suffering these problems.
There is already latency applied and it can be set by the user in the game's options menu. Faster systems can handle very little latency but the game by default has the latency set to around 150 ms which is enough buffer for most slow downs. Thanks for the suggestion!
@@AngeTheGreat Way ahead of me already then! I can't wait to see the mechanical component applied to your other engine models.
If you blindfolded me, and told me it’s a real steam engine, I’d believe you.
This is astronomical upgrade in sound quality and realism.
The "bad" sounds around 9:20 sounds like somthing Aphex Twin would remix
says a lot about aphex twin
haha yeah! steam licker.
It may sound bad but at least it still sounds mechanically accurate.
Every mixer in an audio-workstation taking multiple asynchronuous sources could make use of that "make zeroes go away by predicting some fake audio" hack. imho.
Some great techology in here, bravo!
Wow!
When you put all the elements together, I was taken right back to that time I stood next to a pumping engine many years ago! A sound I had forgotten, brought back - that's how you know you got it at least mostly right! The hiss of the valve you actuated in particular was uncanny!
Ange this is incredible, I've spent most of my life around Steam Engines, I could close my eyes and it would be like im right next to it with how realistic this sounds, you've done seriously well!
I love the new whistle! sounds like a nathan 6 chime. Have you considered adding a feature that lets the user edit the notes it produces?
Bad times have a scientific value. These are occasions a good learner would not miss.
I am a train nut myself and this is insane. One thing that would be fun is to add different size and chime/note whistles. Which in the real world changes the sound and pitch. Just a fun idea of course, but i would love to see it done because the whistle simulation sounds outstanding. Cheers!
Wait a second? BEAMNG!? As in, the car simulator? They support you?
They have supported him for months now
beam is goated
If you look at Ange's racing game video, he says they've supported him since basically the start of the project
They've always been a sponsor. I mean, it's clearly in their interests so it makes sense.
Bro, I'm more on Software/API side of the programming world and the simulation engine you're building is insane on all side.
You need such deep knowledge of your programming language plus insane mathematics skill that I don't even know where did you get all this knowledge from.
Incredible work.
That is absolutely amazing. I was away from my screen when you played the full demo, I could only hear it. Sounds like a massive steam train full of heavy parts.
The before and after is crazy here, good work!
Impressive as always, really believable and so cool knowing it is all 'made', nothing prerecorded! Ultra cool.
It really is the little details that make the biggest difference. As always, this is incredibly high quality work.
I'm going to use this to teach my child more about engines this is amazing in every possible way i can't thank you enough. amazing work. as soon as I have the bandwidth I'm going to become a patron.
just buy the game
The incredible difference in realism in the side-by-side comparison is jaw dropping! You are doing great work, and you really are hugely inspiring :) Making me want to go and continue learning more math and physics
You are absolutely wilding my dude. The progress is insane! Cant wait to see all this fancy shit somehow implemented to 3D Engine Simulator!
Imagine how cool this would be in Derail Valley. The S282 and S060 will sound AMAZING
Amazing work as always! I'm very excited to see this attention to audio.
The autoregressive model audio still sounded like clipping to me, just less pronounced. Maybe it's not in phase or something?
There are limits to how much the audio can be corrected when there is literally no audio present. It really doesn't sound like clipping or clicking and you can confirm this using a spectrogram and comparing it to actual clicking. There are no discontinuities in the signal whatsoever. Every other approach I've tried results in much more obvious artifacts when audio is dropped. In regular play, most errors with audio prediction enabled are reduced to just subtle blips or frequency detail loss. You can probably analyze the signal from this video yourself in a spectrogram and you'll see what I mean.
This is *amazing!*
Seriously, this could properly redefine how sound design is done, especially in vehicle simulation games.
12:20 Had me grinning like a fool when all the sound components came together. Awesome
I love these videos - not because I can pretend to understand any of the magic you are performing in code, but because you lay out so well how you identify and approach the problems you are solving. It's a masterclass in how to do iterative design.
As I've seen, you have the best mecanical engine simulator engine, it would be very nice to see a "train steam engine simulator" with all that real time response of mecanisms and sound. Keep the good work!
you are insane Ange! everything you do is above and beyond our expectation!
OH MAN! The best time is when Ange uploads
Im glad the algo brought you back to my feed. Seeing how much you have developed since then is absolutely awesome.
modulating the filter with the lateral piston force is so big brain i love it
i think i said this on an older video but god imagine the power if sound physics was applied to making mechanical sirens. with a rotor and stator and everything. not just hand cranked but like full on electric. electromechanical. hearing a giant steam train plowing through the tracks is already an insane feat and it's crazy to see where these sound physics have gone in the past year or two
Brother can write an entire research paper on this topic. This makes me feel so stupid yet amazed. Really informative content keep up the good work :)
Honestly it is INSANE. I cannot even begin to fathom how would I approach coding this. If i was blindfolded and listened to isolated sounds of real engine and this simulation I probably wouldnt be able to tell the difference. Wow
oh man that full demo comparison really makes the difference in realism noticeable. gotta love steam engines. during my vacation I went to a technology museum and they had these humongous steam locomotives there. truly magnificent machines
Love the screw reverser and how the simulation actual shortens the exhaust beat. It would be so cool to see your simulations be used in train sims for accurate steam noise and whistle quilling. It’d be cool to hear some brake squeal and added rod clank when the reverser is a bit more centered or off throttle completely as well… unless there is and I didn’t notice
I looove your content and have been tracking Steam Engine Simulator since launch. Seeing it progress into a full, super realistic simulator is so awesome. Keep it up!
Holy crap, and I thought the sim already sounded amazing! You have done such amazing work as always!
7:02 that whistle actually sounds very nice and realistic imo
It's so interesting seeing the entire process of this, I remember when it first started you had to explain to people that you were doing your best, and that what you have put together (at the time) sounds great already. But now fast forward to the present day, the leap was huge. And the people complaining did actually have a point.
I don't believe the people complaining actually had much of a point. They stated obvious things, contributed zero insight and ultimately changed nothing. Their actions are equivalent to me going back to 1980 and complaining that the CPUs are too slow. Sure it might be true and maybe if you fast-forward to 2024 and look back on my complaints, you could argue that I had a point, but I was ultimately just stating the obvious: that could always be better. I don't believe that's a very meaningful contribution to the discussion.
@@AngeTheGreat you definitely have a point, I'm here watching on the side lines, I always thought this was a great project, critics on here are way too harsh for no reason. And I do remember you explaining that you knew that there were issues and its not perfect, I appreciate your total honesty throughout this as well.
Whenever you have eliminated the impossible, whatever remains, however improbable, must be the truth.
Actually fucking insane. I wish the average game dev had this much dedication to detail.
just incredible. the lengths you've went to for accuracy in something that started off as a kind of as a joke is awesome.
Wow the engine sounds so much better now
I checked your channel 2 days ago to see if there was any update and realised the original Engine Simulator video was 2 years old (Happy Birthday!). I remember when I first saw it I thought it would change everything, it could be a revolution for racing games, animation and probably a whole lot of other fields I didn't knew about.
2 years later I still believe Engine Simulator is one of the best project I saw on UA-cam and I'm really thankful for the time and effort you put into this project and I can't wait to see what's in the future.
The second half of this video made me time travel to the industrial revolution 10/10 would steam again
Ange, I haven't watched this video, and I don't always get a real kick out of your videos, but there's a very simple reason I click on almost every one of them: you put a lot of effort into these projects and the videos, you want to make everything right, and it's all a very difficult challenge but you come through with results every time. I respect your fight, and I'm sure this stuff will be adopted into video games and will give them that next level of realism. I hope you're getting a solid piece of the cake from your industry partners, and that the industry will know your name. Keep it going and don't listen to the haters. Lastly, congrats to a very well deserved 200k.
Edit: the audio improvement is insane! If the steam engine was built into a locomotive, you could also implement doppler effect for bystanders.
He wasn't bitter that she had moved on but from the radish.
That train horn gets me every time. So fn cool.
I can genuinely close my eyes and picture tripod-mounted VCR footage of a rotary snowplough chugging through a snow-blanketed Swiss mountain pass.
different materials will sound different when sliding tho, like imagine the high puitch sound of a train breaking
The more I watch your devlogs, the more I understand you are truly AngeTheGreat. Maybe even AngeTheGOAT
I enjoy how simple small things add up to one complex thing.
the subtle sounds of the mechanical components adds so much more realism!
That high pitched noise is pretty accurate to real life
Wow, SES really went from a tiny steam engine running in an anechoic chamber, to a steam train chugging along the tracks in the city, and all in one video.. impressive.
In comparison the old engine sounds like a steam powered washing machine. I love the work! keep going you are making something amazing
Frequency dependent damping of sound made me smile
It sounds SO good. I need a steam train simulator with this simulation powering the audio.
I'm thrilled about the DLC and hope that there were no significant troubles in its creation.
You as Dev is what Simulators need.
Sounding great.
It is very normal to band limit the audio with a very steep filter. E.g. 60th order IIR lowpass cascade just below nyquist is quite normal in hifi products. Although probably more typical is an FIR filter. Also, you could apply some dither during the creation/conversion of the audio. That would remove quantisation noise. The lowpass filter you use for distance might sound better if you used a first order lowpass filter, IMO it's a little too steep. Finally, if you plan to modulate the filters input parameters quickly then don't use biquads, use a state variable filter (I think sometimes called a "zero delay feedback filter")
The high-pitched sounds at 9:22 are what I thought it should have a bit of for that imperfect realism. It reminds me of a Semi truck's brakes, or just a misaligned motor. It's so cool that literal inaccuracy caused a realistic simulation of noise from inaccurate parts lol.
I've seen quite a few of your videos now and your work is absolutely breathtaking! looking forward to watching this video a little later on this evening.
The improvement and approach using informed design is amazing, these videos are worth a lot
Im truly obsessed with these updates. Your thought process pleases my tism
It would be really cool to have some resonance in the system. Even if it only works for the brakes or for the main bearing when it's worn out something. This entire project is absolutely awe inspiring you are one of THE inspirations!
Thats probably the most amazing exemple of mechanical sound simulation that i know of. Have considered writing any cientific papers on it ?
It does sound amazing but what papers would there be to write? He didn't really invent anything new just optimized existing techniques to work for his use case.
@@theairaccumulator7144 Firstly, most research papers are not particularly groundbreaking. Research papers vary all the way from being summaries of existing research, optimizations to current techniques, all the way to groundbreaking discoveries. It should be noted that in the world of computer graphics and audio, a lot of the research really is just the application of known physics to software-based problems. Secondly, I covered the updates in a very simplistic way in this video so I have no idea how anyone could determine whether or not the results are worth publishing. In any case, I'm not really interested in writing a paper on any of this and for the most part it is mostly applying known physics to a software-based problem. Based on my experience in actual software engineering research, a paper of some kind could definitely be written about this though.
@@AngeTheGreat : Having been on a paper review committee for a computer science conference, I can say that the very simplistic overview you've done in this video is already probably at the "publishable" threshold, so I'm sure the full work absolutely is. :)
More generally, I'd point out that one shouldn't underestimate "optimizing existing techniques to work for a new use-case." In many cases, the optimization is the difference between whether or not it's possible to make the techniques work for the new use-case at all, and showing that it's possible is useful. It's likely that a lot of people who are doing audio work for video games are unaware of the details of how to do a physical simulation, and likely that they're even less aware of what is possible or not-possible to do with them as far as producing useful game audio (because very few people have even tried it), so a paper on that subject will tell them several new and useful things.
Interestingly, it's also likely that most people who are doing detailed engineering simulations of sound generation also wouldn't have a good idea what's possible or not-possible as far as generating game audio. In engineering simulations, precision and accuracy are really necessary, because you're trying to figure out things like "will this jet engine exceed this airport noise limit?" and real-time computations aren't important. For game audio, it mostly just needs to "sound right", and it's fine to sacrifice a lot of precision to get the computations to happen in real time. An example of this is Ange's fix for computation lag -- that sort of thing would never be used in an engineering simulation, both because it wouldn't be needed and because it wouldn't be accurate. And so writing a paper about this is telling the audio-simulation community something new too.
A lot of people give up just before theyre about to make it. You know you never know when that next obstacle is going to be the last one.
Man it truly is a treat to watch these videos. You are very bright, creative, and thorough in your work. I hope it takes you far.