Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Sampled VS Modelled Pianos ???

You are right ... LOL

I didn't think I had it right, but I just had to throw that in.

Don't ask me why. It just seemed necessary.

If you read what I said, it doesn't quote you as saying one thing or another. I don't dispute that they can't play with robotic precision, it's the rounding of the MIDI values that forces them to play like that. And they can sense it.



This is what misses the point. You're putting it in terms of an engineer not an artist. An artist couldn't care less about whole steps or half steps, they want the continuum of expression that means they are (as near as makes no odds) one with their instrument. When you listen to the master performers you can hear and sense that degree of unity. And at this point in time, digitals can't provide these artists with a tool that is at that level of refinement. I am told that a Hammond organ, with its nine drawbars can generate 253,000,000 variations of tone. The human body is infinitely more expressive than that - a ballet dancer limited to just 127 muscle positions would have a very hard time performing Swan Lake. Maybe that's not the best analogy, but it will suffice as I try to suggest that the more digital technology can free itself from coarse limitations, the more it will be capable of assisting the artist.

How is this relevant?If an artist plays 5 tones per second he'd have to play continuously for 38 hours to play them all! But how is 253,000,00 relevant, anyway?

I doubt this:No one can want, have, or use such a continuum. Our senses won't detect anything smaller than some minimally-noticeable increment. Just how small is that increment? It depends on the mode of expression. Regardless, anything smaller cannot be sensed, and cannot be wanted.

I also doubt this:Nothing human is infinite.

This seems reasonable:But that only means that 127 increments is not enough for ballet. Perhaps it requires 1,000? Or 10,000? I don't know how many, but there's a limit beyond which no further gain is meaningful.

As I sit here at my desk, I might choose to walk to the kitchen for some coffee. Doing so, my body mass will successively occupy a large (infinite) range of points in space between here and the kitchen. (Rest assured that my body movements will not be the envy of that ballerina. ) But do I sense all of those many points? Half of them? One-millionth of them? Would the ballerina?

Anyway, what relationship is there between what a ballerina needs and what a pianist needs?

Finally ... if your point is that you need finer resolution than what you can get from a digital piano, I have no quibble. (I don't need it, but perhaps you do.)

But you're unable to say just what level of granularity is needed. And you cannot support the notion that such granularity must be infinite (or infinitesimal).

Which is it?

It's not! It was just a link between 127 and the infinity of muscle movement. Don't spend the 38 hours, it's not worth it!

I believe that great artists do indeed want a level of expression that is, to all intents and purposes, unlimited. It is unachievable, but it is part of the quest.


They both need to be able to coax the maximum of expression from their respective instruments - one is a body, and the other is a piano, but they are both extensions of the artist and their desire to create/perform.


I was referring to people way beyond my performance level (up in the stratosphere by comparison). I can sense the limited response of digital pianos, and I prefer to play an acoustic when available (the audiences I play to usually prefer the results, too), but this is performance at a very low level compared to the great classical and jazz performers.


No way was I asking for infinite granularity. My argument is that the current level of granularity is too coarse to allow great performers to use digitals (of any kind) for performing serious works. Maybe 1000 levels would be sufficient, but I really don't know. You presumably would be happy for things to stay as they are. But I don't think you'll convert any major classical performers to your way of thinking.

It's BOTH. I can occupy an infinite number of positions. Physics allows that.

But I cannot be aware of every one of those positions. So if things were different -- if there were granularity (rather than continuum), I would not know it or sense it.

And that's the point. There exists much more than I (or anyone) can sense. Anything smaller than what can be sensed can be treated as non-existent.

Earlier on I posted that the point is really moot (with respect to pianos). If you need the finer granularity, and yet no piano can offer it, then that's the end of it.

You'd be very aware of it if there were only 127 positions between you and the kitchen. You could enter a robot dance contest!

Zeno made his name with this conundrum.

Not to go too far afield, but ... M-Theory suggests that the number of positions MIGHT not be infinite, and that there may be a minimum non-zero distance (using alternate ten- or eleven-dimensional geometries of the Calabi-Yau form). But this minimum distance is on the order of 10 to the -33 cm. I don't think even voxpops see that a problem or limitation.

I have been following this thread but after 5 pages it only now starts to get interesting.

So Mac*3....which is it?

Ballet or the robot dance?
And of course, a vid or it didn't happen.

Maybe yes, maybe no. You might not be aware of just how close I am to the kitchen!

That describes me perfectly, first thing in the morning when I've not yet had my first coffee. Under those circumstances, I couldn't tell the difference between 2 steps or 127 steps or 1000 steps.

I think this one got carried away a little into argument for argument's sake. The original (well, kind of) question is whether 127 is too few levels for playing.

I think it's been accepted that 127 can be too few for recording and studio work. Also that more than 127 won't hurt anything.

Is it too few for piano playing? We can't really test without a piano that can do 127 and more than 127. I think both sides accept that it might be or it might not. Whether your gut tells you it might be enough or whether it says it might not, it's really just a gut feeling because we can't distinguish this limitation from other limitations in our pianos or software.

Unfortunately the available hardware and software doesn't allow the hypothesis to be examined any more closely. I'd love to know, but I don't see it happening. The rest is just arguments about semantics, isn't it?

That's actually an interesting theory, which would suggest that the world is more akin to a digital representation than we might like to think. As with the "spin" of a quantum particle it is either one way or another, but never anything in between - rather like the zeros and ones of computing. And that would also suggest that the driving force is mathematics. Trouble is if you try to imagine what's between the minimum spaces you get tied in mental knots.

(this relates to the posts prior to the last two!)

Then we do agree to at least some extent, that "the value of a velocity point's existence is not predicated on the player's ability to produce that velocity at will." But I am not differentiating between being able to hit velocity x on demand, and being able to hit the same velocity twice in a row, which I think is the distinction you are making. The way I see it, either way, we're talking about the value of providing a velocity response point that is beyond the player's motor ability to reliably reproduce.

I guess my question, then, is, whether or not you think there is value in providing velocities that are within the "margin of noise." That is, if your attempts to repeat a note at the same velocity can regularly result in velocities that are 9 or 10 units off from one another, is it or is it not worthwhile to have all those in-between velocities available in the first place? I think you and MMM have suggested that, if they are beyond your ability to exercise control over (i.e. your attempt to generate one will as likely generate one of the others anyway), their existence is moot; whereas I am suggesting that they are still valuable, and would clearly make their presence known as you crescendo'd through the range.

Someone could play a VAX77 and Pianoteq with its 14,000 point resolution, or with 127 point resolution, and see if they could reliably tell which one they were playing, that might at least provide a clue...

As for unprovable gut feelings, my own is that the reduced expressivity of digital pianos compared to acoustics is at least as much a result of imperfection in translating our finger motions to the 127 points as it is to the limitation of there being only 127 points.

As an aside, in my experience, some acoustic pianos exhibit noticeably more timbral change at different velocities than others do. Though even pianos with lesser dynamic tonal variation still undeniably feel like real, acoustic instruments!

It is very easy to test with with less than 127, though. Just quantise the MIDI signal to whatever degree you want. Just perhaps, we may not even notice a difference with a 6 bit quantisation (63 steps). If we reach a point where we do notice a difference, introducing dithering to that signal may restore the overall experience to approximately what it was at a higher number of steps. I.e - there may be a point where the quantisation of our (human) random noise becomes audible, yet, we are in fact capturing the intended velocity with enough precision. Dithering allows the intermediate steps to be randomly invoked, hopefully in a fashion as if the velocity were being captured with higher precision at the outset.

If the goal is to reproduce any given performance very accurately, then that changes everything. In this case, the velocity "noise" may not be noise at all, if it causes audible differences. It doesn't matter if the performer can't reliably reproduce that same performance at will - we want to capture that performance as accurately as possible.

Greg.

Yes, I agree that we are talking about the value of providing velocity quantization that I think is beyond a players motor ability to reliably produce. For the purposes of my model, it is only important that the player can't hit the same set of velocities every time the play the same piece of music, so I'm not concerned whether there is any distinction between that and hitting a specified target velocity.


First, let me say thank you for taking the time to understand what I was saying and then considering the merits and issues of that model. I'm reminded of a college professor that once told me that unless you can understand the other person's argument so well that you could switch places and present their argument (without having to believe its merits), you have no business arguing with them in the first place. Hence the first step before arguing with someone is to fully understand what they saying. Unless of course, you are arguing with your wife, which is futile anyway. But I digress ...

To return to your question - yes, I think it is important to provide velocities that are "within the player's margin of noise". But the question ultimately becomes how many velocities within the player's noise margin (i.e. how small should the quantization step size be)? The reasoning is easily understood by considering the MIDI velocity quantization as a second velocity noise source. The player's noise and the quantization noise combine (but not simply additively, we need a little math to combine them) to produce the total velocity noise while playing, i.e. the total uncertainty and randomness of the velocities actually produced. The more MIDI velocity values provided within the player's noise margin (i.e. the smaller the quantization step size) the less quantization "noise" and therefore the less total velocity noise. However, there are non-linear diminishing returns. The player's velocity noise will begin to dominate at some point and then making the quantization step size smaller (more levels) doesn't help much more.

An equivalent way of looking at the quantization noise is to simply realize that the player is producing an analog key velocity with analog velocity noise (due to motor abilities) and the sum of those elements is being digitized with a quantization error set by the MIDI quantization step size. So we want to reduce the quantization step size until their are multiple quantization steps within the player's noise and the player's noise is dominant. We want to set a digital noise floor that is below the player's analog noise floor, but as the quantization becomes smaller the digital noise floor will be essentially hidden below the analog velocity noise. We have a wealth of analytical (mathematical) tools that we can apply to understanding the digitization characteristics of analog velocity "signals" just as we would apply those tools to designing digitization for analog audio or video signals.

So that's a long way of saying - No you can't eliminate all MIDI values that fall within the player's noise "margin".

We certainly can eliminate all those intermediate values, if those intermediate values produce neglible change in sound in the instrument. If they produce a noticable change in sound, the there are two ways of solving that problem: a) increasing the resolution with which we sample the velocity (i.e - sample the human noise as well as the human velocity proper), or b) introducing dithering, so that those intermediate steps are invoked in a natural way, as if we had sampled the human noise to begin with.

Greg.

You beat me to it.

I was going to suggest taking a MIDI performance file that people agree (ah there's the rub) produces a good result with a particular software piano, and then run that file through a MIDI filter that re-quantizes it to 6-bits instead of 7-bits. We could then record the piano audio from the two examples and post 4 unlabeled audio files, 2 identical files with the 7-bit MIDI and 2 identical files with the 6-bit MIDI, and invite people to a) identify which files are the matching pairs, and b) which files they prefer for piano realism. If people can't identify the matching pairs with statistical significance, we can say that 127 MIDI levels produces no different performance than 63 MIDI levels. But if the group can identify the matching pairs, then we can take the votes of the people that got the matches correct and see whether they thought the 6-bit or 7-bit pairs produced the more realistic piano sound. i.e. we would like to not only know if the pairs were different, but did fewer MIDI levels actually degrade the sound quality.

Maybe someone has time to setup such an experiment. I'm afraid I don't.

Of course, if the group can't statistically identify the matching pairs, we may be in for audiophile A/B testing-like nightmares courtesy of folks that still have some green ink on the rims of their old CDs.

@Macy
Yes, good idea. We could start with a standard Pianoteq preset, and if no difference was detected, try increasing the amplitude and/or timbral dynamic range, and re-run the test etc etc. This test won't determine the precision of human velocity though - it will merely determine the precision required to produce undetectable (or unimportant) differences in the output sound, and only for the particular sounds we test with.

Greg.