🌏 Exclusive! Grab the NordVPN deal ➼ nordvpn.com/angethegreat and get extra subscription time. Try it risk-free now with a 30-day money-back guarantee ✌
If I remember correctly from the physics of musical instruments course I took in undergrad, the cylindrical bore closed-open configuration limits the harmonic content above the fundamental to the odd numbers partials only. A conical bore does not. (It has to do with nodes/anti node locations in the pressure wave at the closed end) this is why a saxophone and a clarinet sound so different. And I believe this is why your clarinet is a more convincing approximation than the trumpet.
Former professional tuba player here - you warned us about the intonation and the difficulty of tuning all of the lip & instrument parameters in real time, and then when the sound simulation started... it sounded *exactly* like someone struggling to get a good tone from a weirdly-built trumpet. This is IMPRESSIVE
exactly! it sounds like someone playing one of those really shitty plastic pocket trumpets! i was never pro, but i played brass and woodwind for many years. well i suppose i did get paid to play a few times so maybe i can claim to be professional lol. i've heard a lot of weird horns over the years though, and the sim sounds like a weird plastic trumpet in its mid range. in the low range it becomes more "cheap yamaha keyboard" to me.
@@Gunbudder It makes sense to say it sounds like those plastic trumpets I think, since the simulation does not take into account things like materials and how those materials vibrate through pressure fluxuations. Perhaps ATG might add something like that since there is a big reason why professinal instruments are brass. Though I have never played a instrument. The sim is really cool and I like where it is going.
@@nimits9437 yeah I don't think the material was brought into account for now. More of a big deal when talking about instruments instead of engines. Anyways reeeealy cool stuff!
In reality the geometry and material properties of the brass and it's connection's, as well the contained and surrounding air produce very complex resonance characteristics with very similar dominant harmonics but different timbre due to small deviations and imperfections.
@@wordsareuneccesary Its a society thing. Even if he doesn't care, others do. Employers do. Employers that could give more money. Its practical, especially if he's done the work and just needs to do some paperwork and formalities.
@@MrCreeper20keven discarding societal labels, PhD work is excellent place to get big sound corporation sponsorship for this amazing work. From corporate management perspective, it's helluva lot lucrative to get PhD work and university tie ups for investor confidence than let's say a youtuber. Number of patents and publication under the belt go a long way
@@MrCreeper20k Why should he care about whether employers do? I don't think this guy is making this trumpet for money either. I'm sorry mate, but following societal expectations is not the road to happiness for everyone
@@yesto9676to defend his point. Its not a road for sadness either. Nothing wrong with submitting it. He literally loses nothing(maybe time and other stuff). And the guy just gave a suggestion. And sorry mate but following the societal expectations is not inherently wrong. He can do it if he wants to. I dont know why you are so defensive.
As a physics student and as a hobbyist trumpet player I just want to add my name to the choir singing this video's praises. It's really impressive what you have managed to pull off here!
What is really good about it as opposed to a convolution model is that you can generate NEW instruments with this! and then make a convolution model out of them *at least*, discounting other possibilities of this running real-time
@@Beatsbasteln well as he says, the cpu / gpu demands are insane - plus an audio plugin has a lot of different demands. You’ve really got to be in that field to pull off that stuff well. But the technology and theory is very exciting. It would just take a lot of work to move it from a purely simulation based goal to a creative goal
@@NightMind0 exactly - it’s a step beyond the delay line based physical modelling that has come before. Full on fluid dynamic simulation would be an insane game changer from a technological point of view - but such an idea would live and die by its implementation and workflow.
I think the reason you need lots of air to get good sound is for a very simple reason, the bell will enhance acoustic coupling and also resonate, making the instrument louder and sending that resonant acoustic energy back into the trumpet.
Think also because the actual embouchure opening is much smaller than the instrument bore. Probably only a few millimeters compared to the trumpet bore at like .468”. Sure it takes air to play, but it’s more about compressing the air with your lips.
@@prkassel Yes I agree, its not like your lips are flopping around in the mouthpiece like what the simulation depicted. As well, when you play the trumpet, youre trying to pressurize the air from the bottom of your lungs to your lips using your diaphram. Creating a large area of high pressure before a tight embouchure. Hope it helps Ange, fantastic video.
The mouth isn't a brick wall either. Some amount of backpressure will cause some amount of air to slosh across the lips. I think this video shows the lips being modeled as a perfect pressure source. Making it somewhat softer might lead to a mouth resonance.
physical emulation of music instruments has been developed since a very long time in the pro audio world, and contrary to this "experiment" they are ressource efficient, and sound same as actual instruments (which is the purpose). A lot of engineers have worked on the topic since a very long time with actual results used everyday by audio engineers, producers and musicians. You just ignore everything of this world, but it doesnt mean it doesnt exist, and it doesnt mean what you discovered today is any kind of achievement or "revolution", far from that.
I've been playing the trumpet for about 8/9 years now, it's absolutely amazing how close this sounds to a beginner trumpet player. The coolest part to me was when you did the model of increasing airspeed, the way that the pitch increases, then decreases and distorts is EXACTLY how an actual trumpet plays when you do the same thing.
Trumpet player here. I’m surprised how good the trumpet in your model sounds. The characteristic sound definitely comes through. It sounds a lot better than many Synthesizers I heard.
Not _most_ synthesizers? If it _wasn't_ better than most wave-based synths, that would be disappointing. Is it better than the _good_ ones is the question, and comparing it to Samplemodeling and even Audio Modeling gives you the answer to said question.
I used to be a trumpet player. I am very impressed with this. Especially the over-blow bit. The simulation was able to pretty accurately predict what actually happens when you blat. The mouth simulation would need a lot of work to be accurate. There is much more to embouchure that would need to be considered. You'd probably have to model a whole head. Its all very impressive!
Agreed to at least the whole mouth, not sure about the whole head. Like, how essential is your skull when buzzing, y'know? I think simulating all the way to the back of the throat would make the most sense, because of how important the tongue is in shaping your tone. Still super impressive!
@@angelicbeast7793 i think will's point was once you have put in the effort to model the mouth, you might as well just simulate the synapses in the brain that decide which note to play :DDD
I know this is really not the goal of this project, but as a music producer, I have to say, that this would be the most awesome vst plugin for physical sound modeling… especially with the possibility to create completely new instruments and tweak real world properties, this would be an entirely new way to sound design!!! Sooo cool
The swam trumpet from audio modelling does that. I believe their approach to pgysical modelling synthesis is a even more sound oriented and it is super efficient and realistic
As an engineer I would 100% pay for classes taught by you, very clear and to the point. Demonstrating a high level of dominion of material and amazing pedagogy skills. Keep it up! 🚀
Unfortunately my virtual trumpet skills are just as bad as my real life trumpet skills. Having a semi-realistic simulation didn't help matters but kind of made things worse.
the meaning of "great" isnt so definitive, so maybe it kinda isnt? a baby finger painting for the first time is great, but so is winning the lottery. obligatory 🤓 here
@@genralty ackshually a baby's first finger painting isn't anything great, as it's not an art and even among baby's firsts it's not that important as walking or a handprint 🤓
In the overblown demo, I was amazed it recreated the "double buzz" effect, where a lower frequency is also heard. This usually happens when a tired player can no longer control their embouchure (lip stiffness and geometry) and unintended vibration modes are produced.
Also, I wonder (1) what would happen if you're fully simulating the valves and what happens when you only depress the valves half-way. Half-valving produces a very distinct tone and diminishes the volume. And (2) if you can slide the stiffness of the mouth to recreate a lip slur between harmonics.
Trumpet player here. Your simulation is uncanny. The tension in the lips being higher reduces air volume required and increases air Velocity. The resonance of the horn provides orders of magnitude more amplification than you would expect. The density of the metal plays a role too. Lots of sympathetic resonance in real instruments. Simulation inception would be necessary to capture all of these effects. The fluid simulation is but one part of it. It's probably the most important though which is why your results are so close. Love it man. You should share this tech with music companies like Roland, Yamaha, Nord, etc. They could really benefit from your achievement, and their effects and signal manipulation, user control, and expression technology could make this sound 100% better. Like a quantum leap in woodwind and brass synthesis. Patent it please. Open source it. Share it some how. We need this in synthesizers!
Share? Hell no, he spent a lot of time, effort and I bet fixing all the bugs decreased his lifespan by at least a few years. I looked up roland - 3k employees in 2013 and you say to share it with them? He will get 0 profit, credit or anything from that
@@shauas4224 if you open source it you can open source it a way such that it cannot be used to generate profit, there are licenses for that. Or he could just BSD type beat licensing and say fuck it. patenting it is admittedly pretty cringe, especially for home gamer stuff.
@@killingtimeitself yeah good luck finding out that company won't care a single bit about license and then good luck finding out that they used it since I highly doubt any of companies will open source their software as you say author should do. Don't get me wrong, I m also curious to dig in the code and see how this works but downside to that - companies will abuse and steal his creation
as a first chair trumpeter, i can say you may be getting the weird intonation from lack of resistance based on the valve position. no 1 valve generates sufficient resistance but there are a few combinations that do generate resistance at certain octaves due to the resonance. to someone without experience playing it can be hard to feel it, its subtle but very significant. all in all, very cool to see this done.
I think the biggest reason there's not much resistance is that the model doesn't have a mouthpiece yet. We've basically got a really narrow didjeridu so far. (I think the model is doing a great job giving us what it would actually sound like to play a tube that's shaped like that!) Once we get a mouthpiece that's the right shape, it's going to give us *way* more realistic resistance.
@@thomasscheevel5779 i feel really stupid and am kicking myself in the ass now. i totally neglected the fact that different mouthpieces have an effect as well. good spot thomas
@@puddlejumper6999 tbf, i probably wouldn't have recognized the sound if i wasn't the kind of person who's tried to play brass (a tuba, for me) with no mouthpiece before. so when he went in the video about how surprised he was about how much pressure it took to get a sound, it just clicked - yep, that's how it feels and sounds to play a brass instrument with just your face right on the leadpipe
As a musician by training, I can tell you that this trumpet simulation sounds precisely like a trumpet -- and before adding the reverb, it sounds precisely like a trumpet in a well-dampened practice room. Awesome job mate!!!
@@gownerjones There will be some reflections from the ground, but if you have a soft ground, like grass, it will be minimum. To make an echo you will need a hard surface for the sound to bounce off (off). Where I live it is much easier to find a place with no echo that one with. But if you live in a mountainous area it might be harder. But you can try the top of the highest mountain or the peak with the larges distance to the nearest mountain.😀 Also you need a low wind speed to reduce the bagground noise and new fallen snow can also help you reduce any reverbe or echo. glhf.
Truly unbelievable the fidelity you can get with a simulation, I cant beleive how you can get those note transitions without nuances of the valve shapes, and the sound quality with that simple lip simulation, amazing.
Valve horn player here. Great job on trying to capture this complex procedure! Couple things I know from playing a brass instrument: - The most important part is the lips that distincts between a good player and beginner. I can make my instrument sound like your simulation, just by adjusting my lips. - Air Pressure and Lipstiffness increases with higher notes. Playing loud and deep is as hard as playing quiet and high. But one can compensate this by having a more accurate and more stiff/loose lip - The way I let my instrument sound different with the same pitch is by increasing or decreasing the volume of my mouth. Bigger Volume results in a more round tone that is often nice to hear. But sometimes in certain musical genres you want this duck like quacking and therefore you decrease this volume (by adjusting the tounge mostly). This also is dependent on the tone - higher tones are generally played with less volume while lower tones require more volume for the same quality of sound. Do with that information what you will, just some insights from my end.
Thanks for sharing your knowledge! It would be cool to refine the lip model using information from real trumpet players since most of the parameters are purely physical/scientific
As a trombone player, I agree with this comment. I think to really improve the sound you would need to simulate the mouth cavity (up to the whole body) as it acts as a resonance box. It would be fun to play around with a trombone sim as it is simpler than a trumpet (sim-wise, there are no valves) and it would illustrate how the pitch changes when the length of the tube changes. Awesome work, I really enjoy what you're doing !!
@@AngeTheGreat You may try to simulate buzzing just on the mouthpiece. Some consider it a bad practice because the embouchure for buzzing is not exactly the same as for playing the instrument but without the brass tube's resonance you can freely change the pitch so might be able to see better what lip configuration is in tune with each tone.
@@AngeTheGreatDr. David Wilken did his dissertation on brass embouchure, and he has some interesting videos showing brass players playing into transparent mouthpieces. One of the more interesting things I learned from that is that not only are the lips not centered vertically on the mouthpiece, but a brass player will adjust the mouthpiece position depending on range.
One reason I love these simulations is because they push the limits of computation. With web development dominating mainstream, it’s refreshing to see non-trivial C++ projects.
I mean, if someone says they do computational fluid dynamics, "engineering, acoustics, and coding" is basically what it says on the tin. CFD is vital to all sorts of engineering work, and coding is its own sort of engineering. Acoustics basically falls out of the equations that you use to model the behavior of air. And any field that has "computational" in the name is going to involve coding.
Im a trumpet player and not an engineer. The thing i like the most in this video is that, although the trumpet isn't tooned, the fist demonstration of the trumpet's notes is as accurate as it gets. Although im not a luthier, i can definitly say the chromatic scale is exactly as its portrayed: 000, 010, 100, 110, 011, 101, 111. Descending, not ascending.
I'm a trumpet player too, and loved that this video is a defiantly on point for how a trumpet produces sound. I think some of the parameters that are missing to be able to enhance the sound go something like this..... 1. lips, tension. There's several ways to achieve this, basically we tighten up the cheek muscles to make the lips more tense. In addition, we can change the pressure of the mouthpiece on the lips to achieve the same. 2. Instrument construction. The tone is effected by the mixtures of metals that make up the instrument. I'm talking about he zinc/copper ratio on the brass, which effects the "brightness" of the sound, also whether the instrument is silver plated, or lacquered. Again this finish can have an effect on the tone. 3. And here's a whole new can of worms, what's happening behind the lips. The tongue is important as it's placement will dramatically change the sound, as well how we open our airways in the throat whilst playing. Puffing out the cheeks. Also whether we are breathing with the upper part of our lungs or (as is better) the lower part using our diaphragm to apply the pressure that creates the volume. I believe that we tend to use higher pressure than you'd think to produce the vibration at the lips, but the opening is very small, so that volume of air passing is relatively low. It's common for a trumpet player to be able to sustain a note for about 60 seconds with one breath. This should be able to give you the idea of the amount of air that can be delivered with normal lung capacity.
@@Goliath83It’s counting in binary, except the digits are shuffled, because the pipes attached to the valve aren’t in ascending order of length. Instead of being worth 4, 2, and 1 semitones, they’re worth 2, 1, and 4 in order from closest to farthest.
As an audio engineer who sometimes dabbles in recreative game development, loves physics and used to play horn, this couldn't be more up my alley. I absolutely adore your recent projects, Ange. Much love and massive props for all the progress you have made in such a short time frame. You rock my dude :)
do you realize how much of a market there is for tools like this? sample libraries of physical instruments are costly to make and often inflexible to use. with a ui designed for musicipans, the ability to physically model wind instruments accurately and in real time would be a gamechanger!
It’s so crazy to me that like 8 or so years ago I had an idea that modeling instruments using physics would either happen or I would have to learn to do it myself. I would absolutely love to see a version of this project in vst format later down the road. As a producer this excites me so much!!! Keep up the great work!
Well, physical modelling is all the rage now, look at Plasmonic or SWAM. Swam is especially impressive at modelling the wind instruments, i wish it would be cheaper 😢
@@snesmocha That's what Samplemodeling is for. Listen to some good demos of it, i.e. Sam Paryss's Star Wars mockup using it or The Trumpet v.3's demo medley.
The sound breaking up at 18:48 is spot on for the state the simulation is in, I had that happen quite a few times when learning a decade ago and to hear a simulator replicate it almost perfectly is amazing
first thought on the first sound was "hm, straight cut trumpet sounds like Sax with a very, very wet mouthpiece", i guess it's more about the lip params and I wish someone will take it upon themselves to jump through the hoops making this the most versatile physical modeling VST-Instrument ever. Ange, you are breaking new ground in software development, with a perseverance and consistency that is nothing short of legendary.
I really like the particle visualization for fluid flow! It reminds me of… an infomercial for a vacuum cleaner. I think including the ability to create extruded 2D shapes in engine simulator would be a good feature, because that would allow simulating different muffler designs.
I want to say that its impressive how you send the information to the viewer. I dont know most of the things you say but somehow I understand everything.
This project is so cool! From what I’ve read about mouth pieces, they actually function as Heimholtz resonators, with another pressure boundary (or whatever its called, like the bell) at the base of the mouthpiece cup. This junction is the interior point at which the vibrating column of air reflects. In real trumpets, the size and taper of the backbore - as well as whatever gap there is between the end of the mouthpiece and the beginning of the leadpipe - has a large effect on tone and slotting (how easily the notes pop out when you play). And the size and shape of the cup also affects whats going on. I know thats a lot of extra stuff to model but would you consider it? Other fun aspects of brass instruments which I’ve wanted to see modeled for a very long time are curvature and bracing. The curves in the pipes of brass instruments seem to function as longer pipes than the straight sections. Brace placement/rigid points along the tube also affects slotting and intonation. I know you’re trying to keep things simple and only doing a straight tube right now but these are some things I would absolutely love to see. Amazing job!
Its hard for me to describe how captivating it is, to listen to a truly passionate and intelligent person explain something they've made. Its just SO nice.
Would you consider doing a similar simulation for the human vocal tract? That would be a great educational tool. There are no truly good online resources that explain how and where the different formants of speech are created, the explanations are quite rudimentary. E.g. visualize how the same formant (in the output spectrum) may originate from two alternative configurations of the tract, due to the tounge position.
I don't think this type of simulation is capable of coming close to simulating human voices. It's useful for systems that consist of connected cylinders of air, not something with complex geometry. I imagine it would be very hard to physically simulate a realistic / convincing sounding voice.
if this could be possible, then it might be possible to generate voices from it, I imagine it would be impossible to generate the voices that is the pressure waves and other mouth expressions manually to output voice, we may train a neural network to generate the needed air pressure and toung and mouth movement to generate a voice, This could be the most realistic TEXT to SPEECH generator of all time. have goosebumps just thinking about it.
having heard a good deal of trumpet-including music, both live and online, i have got to say that this sounds pretty damn good for a fluid simulation. there were some small parts where the audio became noticably "rough" in a way that i only really know from computer-generated audio, but during some other parts it sounded like an authentic trumpet being played by someone just getting used to playing the instrument. i've got to say, i am very, very impressed by your work and i'm certain that the quality of your other computer simulations can only go up from here.
I've played the trumpet for well over 10 years throughout middle and high school. I'm also self taught in C/C++ and 3D Graphics / Game Engine programming using DirectX, OpenGL and Vulkan. I'm also adept with mathematics and physics. To see a video of a physics simulation that uses C++ and Vulkan to project the fluid dynamics of a trumpet is really cool. This video is very informative, creative and quite awesome. Job well done!
This is so cool! It goes quickly from "huh, that's a neat simulation" to "wow, that visualization looks incredible!". I also appreciate the non-fearmongering VPN ad.
What an emotional ride! From the terror of partial differential equations to the hilarious idea of an engine with lips, this video has it all! As usual, I'm totally blown away by your success! Want a fun next project? How about soft-body tubes? Might not be possible at real time speeds, but it might be fun to try anyway sometime since you could potentially write a physics-based text to speech engine, where phenoms are actually caused by muscle movements! :O I always thaught that would be cool in general but I bet linguists would go nuts being able to physically model accents and speech impediments!
Use the system to simulate the call of the parasaurolophus. It has a large resonating chamber in its crest that drastically changed the sound they make.
This project is insane to me in the best way possible. As a former awful trumpet player, the off putting part about the first demo was how clean the sound was with a stable air supply and no reverb. Once some reverb was added, it just sounds like a trumpet with some tuning problems. Can't wait for this to release and see what people can do with it.
This is absolutely incredible from start to finish. The data visualization is a master class in visual representation, and the project itself is just mind boggling.
This could definitely be great for music production purposes as its honestly the most realistic trumpet simulation I've heard. Idk if it'd be possible but integration for use with DAWs would take this to another level.
make a realistic pipe organ, replete with all the stops etc. That would make not only a great steam game but also fill a market void of physically based music synthesis for musicians
Experienced brass musician here. It sounds like an early beginner trumpet player who is trying to find out the best way to make a sound! I'm mostly impressed at how accurately the valve movements distort the tones between presses, which sound *very* accurate. From the sounds of it overall, the tone of the trumpet could mostly be fixed by making adjustments to the player's embouchure (the mouth and its positioning on the mouthpiece). But I can say that I've heard these kinds of tones from the beginning trumpet player.
To echo what a lot of other trumpet players are saying here, it really speaks to the accuracy and sophistication of this model to say that you have accurately recreated a beginner trumpet player. You said yourself that the embouchure is difficult to model. Well, a beginner trumpet player has little concept of embouchure and therefore sounds almost exactly like your trumpet model. You have perfectly modeled the trumpet, but not the player (which is beyond the scope of this project). Anyone who has learned the trumpet or been in a middle/high school band can vouch for the accuracy of this model!
During your research did you use the book Computational Fluid Dynamics with Applications by John D. Anderson? It uses a finite difference approach which can be easier to program. The other very good book that maybe you used is Introduction to Computational Fluid Dynamics - The Finite Volume Method by Versteeg. This book really gets into the details of the solvers, PISO and SIMPLE.
The mathematics reminds me very much on my recently finished university project about finite difference methods for electrodynamics (Since they are also hyperbolic equations). You said something about further sources about the math behind your simulation is available in the video description. But I can’t find any link. Edit: Links are now available. Thanks!
I forgot to add them before, but they should be there now. I have some additional links that I'll post later as well, I'm just trying to collect them all lol
Played trumpet in junior high and high school, and I can say with confidence, trumpet is among the easier instruments to learn, and if you up and bought one, I'd highly encourage trying to learn it.
You could make it as a VST so it can be played with midi inside a DAW or other programs. There aren't many pysical trumpet simulation so it could be interesting
Because of the embouchure needing to be emulated, this isn't so easy to replicate or achieve using a MIDI keyboard, but I do agree that this would be great. Solo trumpet VSTs and sound fonts tend to sound less than ideal for a quality recording, so something like this would be incredible.
I played the trumpet for almost 10 years and I can say that this is actually amazing to me. Despite its flaws it is the most accurate sounding audio I've heard a trumpet make artificially. Everything I heard sounds very familiar to what I'm used to hearing. I could almost feel the different pitches as if I was playing it, definitely a wierd feeling 😂 if there are any other players here, you might know what I mean. Now that i think of it, the sounds the simulation makes reminds me of what it feels like to play with a earbud in one of your ears. You can feel those same vibrations through the audio of the video.
This is amazing - as a physics graduate working on particle simulations, and a trumpet player of 10 years, its amazing to see those two worlds collide. The fact that you can even get a roughly-correct chromatic scale and multiple harmonics working is extremely impressive. To me, it sounds exactly like a trumpet playing into a cushion, or a really closed mute. I think simulating the bell flare will definitely help (I've experienced many dented flares). For even better tuning, the third valve tube would need to be extended slightly in the lower harmonics.
Flute player here, sounds the same as any trumpet I’ve ever played with ;) A simulation of a flute would be very cool but certainly very difficult. There are a lot of weird things that can happen. Anyway, I’m very impress that you can get this to work at real time, I didn’t even think that could be possible. I’m a big fan of optimisation so I love watching your videos!
as a trumpet player, although it doesn't sound 100% right, this was absolutely amazing to hear. Especially the parts where the trumpet is mid way between pressing a valve. That sounded almost exactly like what you would hear in real life if you don't press down the valves as quick as possible. I know i kind of rambled, but I've watched your videos in the past and please keep on doing whatever you are doing cause this is absolutely amazing to watch!!!!! THANK YOU!
This video devoured my internet bandwidth like a snack... What truly blows me away about you, and this resonates with me (haha, resonates), is that you give your 110% in everything you do. In this instance, not only is the video content top-notch, but the production quality is off the charts. I mean, even your compression settings seem like you've mastered them to perfection just to deliver the most incredible results. Simply mind-blowing.
This is an incredible project! Definitely earned a sub from me :) I want to point out that the reason you get the woodwind sound in some cases is most likely because of the cylindrical bore that you pointed out. The shape of the bore has a lot of influence over the harmonic structure (timbre) of the instrument, so I'd love to see a follow-up once that change has been made.
Dude you’re such a role model for me. I’ve always had ambitions to write a ton of simulations myself, but you’re actually doing it, fast and in amazing detail. Do you take apprentices? 😁
The resonance of brass as a metal is vital to the reduced air flow necessary to produce sound because every inch of the tube is added shaking of the instrument which aids in the increased efficiency of kinetic energy’s transformation into sound.
Oh my god. Oh my god. This is AWESOME! First of all, this video showed up in my feed while i was in a practice room practicing trumpet. Second, when this is finished, PLEASE PLEASE PLEASE PLEASE PLEASE make a VST3, AU, and AAX plugin for this if you can. Also! As a trumpet player and an audio engineering student I have a few observations, questions, and suggestions which you may or may not have already taken into consideration: It sounds like a french horn, or other low brass instrument. Making it sound more like a trumpet (i.e. generally brighter) is going to come down to the resonation of the metal in the bell. I couldn't tell if you were simulating the actual metal itself resonating, but the temperature, thickness, material, and plating of the metal is going to be very important to the sound. Heres a few parameters that I think could improve the accuracy of the simulation (and are also interesting to tweak): metal materials (i.e brass, copper), plating (gold, silver, none), presence of lacquer, Bore size, Bell size, Bell thickness, Different mouthpieces(??), just to name a few. I noticed that your tuning slide wasn't adjusted at all. Unless it is VERY cold in your room, you are going to need to put it out a little to get proper intonation. Also, when playing middle D (Valves 1&3) the third slide will need to be extended slightly, and fully extended when playing middle C# (123) for proper intonation otherwise they will be very sharp. On the topic of the bore and the bell, these things have a significant effect on the brightness and tone. I play a Bach Stradivarius Model 25L, meaning it has a relatively small bell (brightens the tone and increases projection) and a large bore increases the richness but requires significantly more air. I've played quite a few other trumpets and the difference in intonation and playability is massive. Mouthpiece - not sure if you did research into this. Theres a bunch of different parameters for this (cup size, depth, shape, rim size, thickness, etc.) and it affects both intonation and playability. Each mouthpiece has its pros and cons in terms of range, tone, and playability. You can buy a screamer mouthpiece for playing lead and screech all day long but it wont be very rich and warm. Im not sure if any of that applies tho, as theoretically the simulated lips aren't subject to pain or fatigue. Again, if you aren't already simulating resonating metal id say do that first, as it has an enormous impact on tone. I know thats a lot of things to simulate, but I think it would make a significant improvement. Also, I think the most impressive thing here is what you've done already. It sounds like a brass instrument, and it does tuba pretty convincingly. You just need a few finishing touches (and high end). It would be super cool to see this released to the public, but regardless I'm dropping a sub on patreon right after I finish typing this out so I can start playing around. The ramifications for this in the production community are insane. Hyperrealistic brass instruments right in your daw???? Theoretically, this simulation could be used to recreate room reverbs, without the need for an Impulse Response and convolution reverb. I wonder if there is enough performance optimization available to make a sim that large work effectively in real time. Anyways, this is super awesome, and bravo for making something real that I've only ever dreamed about. I can't wait to see how you improve this, and whatever other projects you have in store.
In the past, they had to get all dressed up and go to a jazz bar. Now I can just get drunk at home while watching a trumpet simulation on youtube. What a time to be alive.
I find this project so inspiring, and I can't help but feel so much pride watching this! You've done so well! Edit: like seriously, as someone hugely interested in audio, music, physics, engineering, computer science, statistics, and many forms of visual art: this project meshes my biggest interests together in a way I'd never have expected. I cannot wait to see what's to come!
As someone who has played trumpet for 15+ years this is more accurate than you may realize. I know others have commented it as well but the trumpet is indeed a high pressure high volume instrument. I'm not sure exactly the scale of air volume you were having to simulate but you comment about probably not needing that much air made me chuckle as I thought back to all of my teachers saying "more air more air" as I was learning. The portion where you increased the volume of air over time and the pitch got lower does actually happen and many trumpet students struggle with "blowing themselves flat" because they don't properly tighten their lips to account for the additional volume of air. I would love to see this come to steam and be refined as its a really great visualization of how playing trumpet works! P.s. you should make some trumpet lessons one of your creative outlets for a bit just to get a sense of how close this actually is!
Oh my god, this is an amazing project. As a passive fluis sim enthusiast, you have just reignited my passion for this field thank you! 1 tip, i think 1 of the reasons why you needed so much pressure to make a noise is the mouth piece. Not sure if the modeled mouthpiece is the same as the simulated one, but the concave section is too small. From what I remember from playing a trumpet in middle school is that the domed section in the mouthpiece was much deeper, and the walls where thinner. Hope that makes sense haha, can't wait to see the next update of this!
With like two more layers of simulation inception this could get scary close. Aside from the bell, material type also plays a large factor in sound resonance. This video is truly awesome, you did a great job.
Professional trumpet player here, this is fantastic! Even without fully modeling the stroke of the valves, the sound of changing pitches is quite accurate. To answer why a seemingly-ridiculous amount of air volume seems to help the sound of your simulation despite it not being realistic: there is a great deal of air compression generated when playing trumpet, such that if the lips did not resist the flow of air, one would empty their air capacity in under three seconds. If the 'lips' in your simulation do not make a distinction between pressure and volume, then it makes sense that volume could stand in interchangeably for pressure. I have a few questions about the physics behind what I'm hearing: 1) I hear each pitch 'bloom,' growing louder over a small period before reaching its full resonance. What does your simulator look like as this happens? How many periods does this take for the sound to fully 'saturate?' 2) What parameters are you changing with the 'lip' so that it sounds like these different respective instruments? 3) In reality, the mouthpiece-side of the trumpet is semi-open as well, as behind the 'lips' would be one's body, a resonant chamber that directs a positive air pressure differential past the lips. As a result, the vibrations of the lips resonate not just into the trumpet, but also back into the body. The body also receives reflections from the standing wave of the trumpet, creating more interactions that influence the sound. Thus, in practice, the size of the oral cavity, constriction of our airflow, and corresponding air pressure play a pivotal role in our tone production, and explains a big portion of why people sound different to one another. How much of this resonating open tube behind the trumpet can be modeled with Tubular? What would happen to the sound of your simulation if you modeled it? What would the 'wave' of the body look like at different cross-sections?
@@Erin-ks4jp I didn't mean as-is, I meant with more work ofc! :D Considering this was made mostly for a fun quick video, and the video mentions many things that need to be fixed to improve it, I think the potential is there though, if it was focused on.
I've been a trumpet player for about 10 years, and indeed, the lip parameters are extremely important to hit the right tone. Trumpet players even anticipate their lip setting and sometimes also pull the instrument tighter to their lips to prepare for the next note, at least to minimise the air gap between each note. If not well prepared, it even occurs to players that only air comes out, completely missing the note. As found in the simulation, the amount of pressure lips have endure on a Trumpet is immense, professional players need multiple years of lip training and lip hardening to cover an acceptable tone range. It happened to me to have shattered lips close to bleeding after a concert. The simulation exactly sounds like someone with no experience in lip setting, which is the case here. Lip tension is even changing along the duration of the note, call it expressions, phrasing of the notes or modulation, like vibrato, staccato, pitch bend and more... Adding randomness in lip tension is an easy way to improve realism, but simulating expression with lip tention needs musical feeling, which is a learning process that probably only AI would be able to implement.
Came here to say something similar about the lips. The tweaking of the params to get the 'clarinet' sound definitely made that pop out. I Imagine more work on the flare will help, didn't hear any talk about the "material", i.e. certainly a wooden trumpet would sound much different than a brass one--I don't know how that's accounted for here... maybe covered in a different video, but yeah, so much is in the lips and I wonder if the original model is just inadequate for the needs here. Buzzing with even just your lips, you can quickly start to understand how much that contributes to the sound. I imagine he could also simplify things and just focus on perfecting buzzing with just a mouthpiece and then plug in the rest of the horn to a massive improvement. Hope he decides to take at least one trumpet lesson as I feel like that will be an epiphany on how important that part is.
This is truly amazing, not least as educational visualizations. If I may have suggestions: a trumpeter's lips do not close completely when playing. It's a common misconception that players buzz their lips, then add the mouthpiece on that, but there is a always a hole between the lips while actually playing the instrument. Second, it would be very interesting to see how "the gap" affects sound, impedance, and intonation in this simulation. The gap (aka annulus) is the subject of many heated discussions. It is the space between the end of the mouthpiece, and the start of the leadpipe. Some kind of venturi. You don't see it as it is covered by the mouthpiece receiver, but a number of respected trumpet makers swear that it has a very important influence on the instrument. Simulating it's effect would be so exciting!
I was thinking about how you could use a fluid simulation to simulate a subwoofer or smaller speakers in a range of different enclosures with different sizes and such, and how different length ports and transmission lines change the sound and group delay and such, it would be a cool video, I don't know how hardware intensive it would be but it would be cool.
Amateur brass player (and professional programmer) here. With regards to the amount of force needed to make a good sound, a brass player will generally use lip pressure to try to make an approximate of the output sound in the mouthpiece. You can see this by searching for 'buzz'-ing, where just the mouthpiece is used to try to train 'muscle memory' of the lips. With enough force, the chamber will resonate anyways, so likely your simulation is just 'powering through it'. One thing you could try is cutting off the simulated pipe before the first valve and tuning the lips to about the pitch you want, and then using that preset for your lip/air when attempting to 'play' that note. Note: This varies based on the note you're trying to hit, as you need different lip pressure as well as valve setting for different notes (well, people do). Another fun fact is notes get 'closer' and require less valve movement the higher the pitch, so eventually you can hit a full scale without touching the valves if you go high enough in pitch. This also requires more care in lip pressure, as you don't 'lock into' the right notes.
Incredibly fascinating demo and also a very clear visualization of what exactly is going on physically in a wind instrument and why! I think one of the hardest things about the learning curve of a wind instrument in general is understanding and controlling all those various parameters in real time while only being able to abstractly grasp what the physics behind the instrument actually are doing, so this is amazing to see! Also I definitely laughed out loud when you started playing the Jurassic Park theme haha
As a trumpet player I am extremely happy and impressed with this video. There are so many different parameters making this an impressive simulation. One interesting thing I noticed isn’t simulated here is there’s a cutoff after the low range of notes where it’s extremely difficult for notes to resonate. It is interesting to see different mouthpieces simulated, as you can technically put a Sax mouthpiece on a trumpet and get sounds.
As a former trumpet and horn player and current software developer, BRAVO sir. The way the note breaks apart when the embouchure is too flat for the instrument on that note is EXACTLY how it sounds in real life! And the way the pitch transitions between pitches on input changes was spot on too
Holy shit. That was impressive. We do blow a lot of air into the trumpet... and it takes more air flow to go higher in pitch. It also affects the "harshness" of the notes we play. Without understanding what you mean by "a lot of air" it's difficult to really say for sure what you are dealing with in this simulation, but I am still impressed at how it actually sounds like a trumpet. One thing about the way trumpets function is that the length of tubing used can be redundant with subtle differences in tuning. Some valve combinations are never used and some become entirely useless as we climb up in pitch. At lower pitches, we use OPEN, 2, 1, 1+2, 2+3, 1+3, and 1+2+3 to descend. However, one octave above that, we don't use the 1+3 or 1+2+3 because those are substituted by OPEN and 2. Being in a different harmonic series, you end up with less valve combinations before you just end up back at OPEN again. Tuning becomes an issue that can sometimes be overcome by using "alternate" valve combinations. 3 alone can act the same as 1+2. It won't be the same pitch, but that's where the slide triggers on the first and third valves help as we use our fingers to manually slide the length of tubing to adjust pitch. This is required for some notes under all circumstances as well. If we never lengthen those slides using the finger triggers, the notes would have to be pitched using our lips alone (which we do in fact have a little control over). There is also a "half valve" technique were we can make a sort of squeezed sound from pressing the valves down in between their fully open and closed states. Not sure if your model can produce that. And when we "overblow" you can actually get a harshness that sounds pretty terrible. Underblowing can still produce a pitch, but sounds weak and feeble. Anyways, hopefully that shared some important and/or interesting trumpet info from a trumpeter's perspective.
The detail in the attacks of those notes is insane. If someone works out a comfortable mapping for pitch+velocity, this would be an incredibly useful VST synthesizer on its own :D Keep it up!
@@YCbCr This has got to be one of the most generalized physically modelled synthesizers right? Is there even anything we can download right now that would be able to simulate arbitrary horns and intonation at this level?
Awesome work ange! One thing I didn't hear you mention was the resonance of the trumpet itself. It probably isn't that important since even plastic trumpets sound like trumpets, but you can definitely hear the difference between someone who clutches their trumpet very tight and someone who lets it hang off of their fingertips. The shape of the bell is designed to make this ringing as loud as possible. I know Le Redditors won't like using a convolution filter for this but I think that makes the most sense here. Anyways I love how your model shines when moving between notes. Those moments really give the model a personality. I hope you continue to learn to play your real and your virtual trumpet!
I was about to say, people would pay an arm and a leg for a VST of this. I think VSTs can't be multi-threaded though, so you'd need to run it as a seperate program and record and export the audio.
As someone who does simulation and modeling for a living, I cannot actually describe how difficult it is to get a simulation running consistently at 44KHz, that's CRAZY fast. I simulate airplanes and submarines and cars and you're lucky to get 20Hz sometimes, from even basic setups.
First of all, this is amazing and thank you for showing us your work. I understand this is a giant technical feat and worthy of praise and I agree with everything you said. Second of all, the moment the trumpet started playing, I burst out cry laughing and haven't stopped since
It's amazing how you manage to get that running on a normal PC. Every time I try to simulate something myself from scratch I use way too much memory lmao
@@calvinkohl6220Learning the math is one thing, but to optimize it to make the code run faster is a different can of worm, especially if aiming for accuracy.
![Real-time Trumpet Simulation [C++/Vulkan] [WARNING: Flashing Lights]](/img/logo.svg){kind=logo}
