Acoustic Instruments are Nearly Unstable

[ssashton]

I hope it’s okay to share a video link here—I thought it was really interesting as an audio nerd.

It’s a short interview with Manny Fernandes, whose job was to develop mathematical models that represent the acoustic physics of real instruments. He did this work for Yamaha when they launched the VL1, one of the first commercial synthesizers to use physical modeling instead of recorded samples.

at 5min ^

What I found fascinating was his description of how the models behave: most of the parameter space is unstable or produces no sound at all, but when everything is balanced just right—right on the edge of chaos—you get something that behaves like a real instrument. Skilled players are able to “dance” along that fine line, shaping timbre and expression.

It leaves me in wonder how people managed to invent so many different acoustic instruments, each with such a wide range of expression, where the physics is so carefully balanced.

 

[somebodyelse]

What I found fascinating was his description of how the models behave: most of the parameter space is unstable or produces no sound at all, but when everything is balanced just right—right on the edge of chaos—you get something that behaves like a real instrument. Skilled players are able to “dance” along that fine line, shaping timbre and expression.

It leaves me in wonder how people managed to invent so many different acoustic instruments, each with such a wide range of expression, where the physics is so carefully balanced.

It's going to depend a lot on the instrument, and which parameters you're playing with, which I think he alluded to. If you've heard anyone in the early days of learning to play some instruments, or learning some more advanced techniques, you'll have heard when it goes horribly wrong. The ones that spring to mind are wind instruments where breath pressure and details of your lips interacting with the mouthpiece make the difference between a note, a horrid screech, or nothing at all. Once you can reliably maintain the note you start learning how the variations within bounds can change the tone, etc. Bowed strings have a similar propensity to screech if you get it wrong. Many others are more constrained in how you get to affect them, so it's harder to take them outside operational limits in the physical world, and what your model variables are in the synthetic one.

I could liken it to learning to balance a bike, almost all of which are marginally stable to varying degrees depending on how you want them to behave. They all walk the line between being so unstable you can't balance them, and so stable you can't steer them, tending toward unstable for manoeuvrability and stability for relaxed riding. And they have been a bugger to model accurately because of the squishy human in the loop actively correcting things as they go.

 

[Barrelhouse Solly]

Wind instruments are the hardest. If you listen to classical flute or recorder music carefully you'll hear a lot of inaccurate pitch. This is true for the best players. Oh, opera singers, particularly coloratura sopranos have this problem as well. In either case it's not due to incompetence. It's the result of the sweet spot being very tiny. It's like hitting a baseball. The best pro hitters don't make solid contact every time.

 

[ssashton]

Yeah my experiance with clarinet was that I couldn't get any sound at all to begin and then I'd manage intermittently etc.. Bass guitar OTOH, got it right away, of course, plucking is easier.

So I was wondering if the instruments with more delicate physics like pitched wind (I.e. not just a pipe, but one that can be tuned) or bowed strings came later than simple plucked strings. Surprisingly not, here is what GPT had to say:

Wind with pitch (flute with holes):
Geissenklösterle bone/ivory flutes (Germany, ~40,000 years ago).

Plucked string with pitch:
Mesopotamian lyres (Ur, ~3000 BCE).

Reeded instrument:
Aulos (Ancient Greece, ~2700–2500 BCE).

Bowed string (with rosin or similar):
Rebab (Middle East, ~9th century CE).

The invention of the bow itself (Central Asia, ~8th–9th century CE, with the rebab) is tied directly to resin use — they almost certainly applied tree resins from the start.

 

[ssashton]

BTW the second episode is interesting too, talking to an acoustics / electronics professor. However I felt like, while he knew a lot about the subject, he didn't do a great job at explaining things - disappointing for a professor but typical of an academic, haha.

 

[TurtlePaul]

Wind instruments are the hardest. If you listen to classical flute or recorder music carefully you'll hear a lot of inaccurate pitch. This is true for the best players. Oh, opera singers, particularly coloratura sopranos have this problem as well. In either case it's not due to incompetence. It's the result of the sweet spot being very tiny. It's like hitting a baseball. The best pro hitters don't make solid contact every time.

I find it interesting how you here this is classical jazz recording and not heard in modern pieces. I guess when John Coltrane misses a note it becomes part of the piece and not just some mistake.

 

[Cbdb2]

Using samples (the instrument is called a sampler not a synth) came 10 years after the first synthesizer, the Moog in 1964. Listening to the video I was only impressed by the wind instruments with the breath controller (which have been around since the 70s) which is what it was primarily designed for. All the other sounds could have been coming out of any number of synth/samplers. Musicians dont care how a sound is made, oscillators, samples, FM synthesis, or physical modeling, only what it sounds like and you can get great sound from a number of different types of systems. Nobody was replacing there Synclavs or Fairlights with these when they came out. Might be good for solos. Useless for anything else with only 2 voice polyphony.

"The VL1 was created with reed instruments and other wind instruments in mind, and thus is two-note polyphonic."

 

[Palmer]

Using samples (the instrument is called a sampler not a synth) came 10 years after the first synthesizer, the Moog in 1964. Listening to the video I was only impressed by the wind instruments with the breath controller (which have been around since the 70s) which is what it was primarily designed for. All the other sounds could have been coming out of any number of synth/samplers. Musicians dont care how a sound is made, oscillators, samples, FM synthesis, or physical modeling, only what it sounds like and you can get great sound from a number of different types of systems. Nobody was replacing there Synclavs or Fairlights with these when they came out. Might be good for solos. Useless for anything else with only 2 voice polyphony.

"The VL1 was created with reed instruments and other wind instruments in mind, and thus is two-note polyphonic."

In my brief time on stage, my fellow artists always said: The only thing unstable around here is that guy...

 

[ssashton]

Using samples (the instrument is called a sampler not a synth) came 10 years after the first synthesizer, the Moog in 1964. Listening to the video I was only impressed by the wind instruments with the breath controller (which have been around since the 70s) which is what it was primarily designed for. All the other sounds could have been coming out of any number of synth/samplers. Musicians dont care how a sound is made, oscillators, samples, FM synthesis, or physical modeling, only what it sounds like and you can get great sound from a number of different types of systems. Nobody was replacing there Synclavs or Fairlights with these when they came out. Might be good for solos. Useless for anything else with only 2 voice polyphony.

"The VL1 was created with reed instruments and other wind instruments in mind, and thus is two-note polyphonic."

Personally I don't care how it sounds, I just find the way it works interesting!

I know it can be a really tricky job to sample some instruments like bowed strings and wind. Perhaps more advanced / fuller physical modelling will come back with AI helping to optimise. The progress in natural voice synthesis in the last few years is breath-taking and far outstrips anything you could do with samples, imagine the same for virtual instruments.

Edit: ahh well, apparently SOTA voice synthesis does not use physical modelling but rather a kind of deep learning diffusion model to generate spectrograms that are then converted to waveforms. I guess this will be the direction for instrument synthesis- I look forward to seeing something like Edirol Orchestral VSTi (showing my age) with this kind of tech.

 

[MRC01]

Wind instruments are the hardest. If you listen to classical flute or recorder music carefully you'll hear a lot of inaccurate pitch. This is true for the best players. Oh, opera singers, particularly coloratura sopranos have this problem as well. In either case it's not due to incompetence. It's the result of the sweet spot being very tiny. It's like hitting a baseball. The best pro hitters don't make solid contact every time.

Another factor with winds is the position of the tone holes. It is always a compromise affecting both pitch and timbre. Consequently, some notes tend flat, others sharp, some tend bright, others dark or stuffy. To play in tune with consistent timbre, the player must compensate with embouchure to bend each note slightly differently. This is especially evident with the higher pitched instruments, as the ear becomes more pitch sensitive at high frequencies - or put differently, the "same" amount off in pitch becomes more evident and dissonant.

Which brings to mind various musician piccolo jokes...

 

[Barrelhouse Solly]

I find it interesting how you here this is classical jazz recording and not heard in modern pieces. I guess when John Coltrane misses a note it becomes part of the piece and not just some mistake.

Never said that. I don't listen to much jazz flute. I don't think jazz players' intonation is any better than classical players.

 

[Palmer]

If you want to blow your horn, but not use acoustic instruments, why - here's the Roland Aerophone Brisa...
For a measly $1759.99 one can bring all sounds of , well, sounds to the party: The 100 onboard tones cover the full flute family - alongside saxophones, clarinets, trumpets, orchestral strings, world instruments such as shakuhachi and pan flute, plus synths and drums for contemporary styles... Should be stable, perfectly stable.

Roland introduce Aerophone Brisa

Roland have expanded the Aerophone family with the launch of the Aerophone Brisa, a new flute-style digital wind instrument.

www.soundonsound.com

 

[ssashton]

Well that's cool!

Here are the sounds

I found a portable one from China too:

 

[CJH]

Are there any electronic keyboards that sound analog and not "electronic"?
CJH

 

[DVDdoug]

Are there any electronic keyboards that sound analog and not "electronic"?
CJH

Yes.

I'm not a musician but I believe most MIDI samples are recorded from analog so it sounds exactly like the recorded analog. Better keyboards have velocity and pressure sensors so the sound varies just like a real instrument. And there are "humanize" plug-ins that add randomness so you don't get the exact same waveform and timing every time.

Most music you hear on TV and in movies is MIDI. There's no band or orchestra like in the days of Henry Mancini.

 

[nerdstrike]

Are there any electronic keyboards that sound analog and not "electronic"?

The Roland V-piano (one of many, but I'm most familiar) is trying very hard to simulate an acoustic instrument, with several layers of processing including dialling up and down cabinet and action contributions to the sound. It doesn't sound that good in 2-chanel, but the 4- and 6-channel noises are pretty persuasive. Good enough to mitigate my frustration with being unable to get a real piano up to a 3rd floor flat with it costing as much as the instrument.

The premium noises in software sequencers (and embeddable into high end synths) are sample-based. They spend disgusting amounts of time sampling an instrument in different conditions to generate a bank of realistic noises. The hard part is using a keyboard and array of knobs to articulate the sounds sensibly. In practice I think it turns into a marathon of tweaking after recording to create natural-ish phrasing and intonation.

On topic, as a one-time clarinetist I can tell you that it is a particularly unstable instrument, giving it a very wide pitch range in three registers. Pitch control in the two upper registers is extremely fragile and gives us the famous opening of Rhapsody in Blue. At my best I could gliss a fifth on embouchure alone, an octave with smearing of fingering and something like two lumpy octaves with half-holing. You've got at least two fingerings for the "throat notes" and you choose the one that gives you the best tonality vs difficulty trade-off. It's a gnarly instrument once you get beyond the fingerings.