need for larger velocity scales, 0-128 to small

[synthjoe]

...the ability to express such small level differences in dynamic instrument playing is probably not terribly significant.

Let me wholeheartedly disagree. This is what determines expression, expressiveness. To some it makes a difference, to some others not quite, but with a mediocre key sensing mechanism and improperly fitted velocity curve with the whole limited to 127 and less discrete values expression is down the drain. Not much of an issue for Van Halen's 'Jump' to me, but renders most digital pianos useless when I'm trying to play Erik Satie's 'Rêverie du Pauvre'.

[Zeroesque]

I disagree. The advantage of a continuous analog device over digital (in this case, the motion of the hammer mechanism) is that there are no discrete steps. There are, in effect, an infinite number of possible velocities that a hammer can be thrown, in smaller and smaller fractions, between it's at-rest position and its maximum velocity. Just as there are an infinite number of frequencies between A=440 and A=441, and an infinite number of decibel levels between 50 db and 51 dB. It doesn't mean we can *hear* them all as being different, of course.

I agree with Scott on this one.

That's unfortunate. Planck's Constant says otherwise.

And I think that improvements in the actions themselves and possibly the velocity curves will probably produce more benefits than adding more gradations of velocity tracking. But, that's just my opinion.

And this one too. But having more than 127 after all this has been improved would probably be yet another improvement. Maybe not. I'm keen on the betting offered by Zeroesque.

Yeah, it's on, baby! You should probably first try a controller that is verified to produce all 127 velocities and also a piano that has a nice curve adjustment, such as TruePianos. Then you might not be so confident, but I'm definitely up for the challenge, either way!

[ozy]

[quote="synthjoe I'm trying to play Erik Satie's 'Rêverie du Pauvre'.[/quote]

and you play THAT on a casio stage piano?

on the m3's rh3?

using what samples: the factory ones?

you can't like that very much, can you?

but are the keys the problem?

[Kevin Nolan]

No, you can't equate MIDI velocity steps to dB.

Actually I think you can because the volume level increases with velocity relate directly to loudness of a piano. It may be crude at just 7 bit resolution, but I do belive that the linear increase in MIDI velocity may link to the logarithmic increaee in loudness we hear from any instrument.

[synthjoe]

I'm trying to play Erik Satie's 'Rêverie du Pauvre'.

and you play THAT on a casio stage piano?

on the m3's rh3?

using what samples: the factory ones?

you can't like that very much, can you?

but are the keys the problem?

I used to perform that on real pianos (and practiced on a Bosendorfer). I was unable to produce the same number of 'shades' of p ever since on any of the digital pianos or synths I have tried. That includes Yamaha's prouct range (Clavinovas, P-series, etc.), Roland, Kurzweil, Kawai. I cannot recall all I have tried, but I remember not finding one true to life, for sure. My problem is that velocity steps in the quiet range are too far between, or I don't have the full dynamics for loud passages if they aren't.

[synthjoe]

That's unfortunate. Planck's Constant says otherwise.

Hmmm... I think you'll need to elaborate on that, maybe just for me.

Yeah, it's on, baby! You should probably first try a controller that is verified to produce all 127 velocities and also a piano that has a nice curve adjustment, such as TruePianos. Then you might not be so confident, but I'm definitely up for the challenge, either way!

I'm sure I already played controllers/keyboards that can produce all 127 values. I'm not saying I have verified them all and do not pretend to have verified all related velocity curves. All I'm stating is that I have not yet played a digitlal keyboard that satisfied me as I wrote in the post above. Given that it is subjective, you have to admit it'll be difficult to argue. I like the idea you have proposed, I hope I'll have a chance to try it soon.

[synthjoe]

Just for Zeroesque's reading pleasure, I have checked on my Clavinova CVP-92 right now and I can produce all velocity values between 1 and 127, though it is very difficult to produce consistent consecutive velocity values in the lower ranges (say, below 32) other than 1 (2, 4 and 6 for example, 3, 5 and 8 seems much easier to achieve, 7 is almost impossible).

Also created a MIDI file with 1-127 velocity values and the CVP's engine responds too sensitively in the top - say - 20 values range, meaning that between a change of 1 in velocity there is too much loudness increase. So my verdict is that in the soft range the keybed, while in the hard (loud) range the engine might be culprit of me being unsatisfied with this particular instrument (apart from the engine as a whole, which I think was not particularly good even at that time) - and between the two lies somewhere the velocity curve.

I admit it is an old keyboard, but this is what I felt to be the best at the time out of all I have tried (bought the exact one I have auditioned in the store). I have not found a significantly better implementation up until today making me sell the Clavinova and buy another - though I usually try them 'all' every year at the Musikmesse. Maybe the combination proposed by Zeroesque will be THE breakthrough.

[ozy]

velocity steps in the quiet range are too far between, or I don't have the full dynamics for loud passages if they aren't.

Your experiment seems to confirm my opinion:

1) definition is there. 127 steps are there. Tou can (0 the keyboard can) and CAN (= your fingers are able to) get 127 STEPS, which means several thousand combinations of them (velocity+duration, velocity+release velocity, relative velocity in trills, etc)

2) the CURVE is the first problem (not all velocities are indeed USED and output by the midi interface. Some of them are SUMMED. It is the MIDI INTERFACE, not the keyboard scanner, whcih takes veloicities 5to15 and outputs them all as 10. that's the CURVE problem),

but it affects only the lowest range.

Please be advised that low definition at lower velocities is often willfully PROGRAMMED by synth producers (check any manual, look what the curves look like when they are well defined).

The purpose is avoiding "smears" (unwanted note by less-than-good players who drag their fingers).

Bad players are 90% of synth and digital piano buyers, and they'll complain that the keyboard "has a bad feeeling" when they never learned playing staccato -

- then they can't avoid smears.

But of course, "the problem is in my tools", non "in my fingers".

Also, synth producers tend to "sum up" high velocities, giving a 125/127 output to anything above 110 (or 100).

Because computer-addicted, flat-chested, sushi-eating players have no muscles in their fingers.

This is also why aftertouch is often badly programmed: people use it as a "trigger" for vibrato or filter, and need it to be "easy" because otherwise they'd have to jump on the keys to activate it.

Aftertouch should be on the contary CONTINUOUS with velocity, "Aftertouch value 1" should be almost equivalent to "velocity 127+1".

That would help a lot in emulating piano btw.

On the contary, aftertouch is usually: "play any velocity then get stiff and press + nothing for a while + 50/70 if you get you wrist stiff and hammer hard" or "play at any velocity, press, bum! 127 to lfo!"

this is why somebody earlier said "I can't use aftertouch on rapid riffs". Absurd. Think emulating a violin and using the aftertouch for pitch bending. You DO need it in rapid riffs. but it must be of the "velocity morphing into aftertouch" kind.

[btw: vl1 has the best keyboard and curves under this profile. smooth transition. I couldn't test the kurzweil midiboard but I've been said it's good as well]

This is also why aftertouch should be BY DEFINITION polyphonic: if it's contiguous to velocity, that's obvious.

3) the second and biggest issue is: the synth engine doesn't CONSISTENTLY reflect the whole curve.

It is a fairly common experience (correct me if I am wrong):

a) you listen to a piano or electric piano emulation on youtube, and it sound nice (a sequencer is in charge, nobody makes professional demos live)

so you like the engine.

b) you go into the shop, and fiddle with the keys on the turned-off keyboard while waiting for the sales attendant.

You like the keys.

c) you turn the darn thing on, and what the fingers say to your brain seems disconnected from what the EARS tell him.

Progress has to be made in sound generation, with reference to correct modeling of response to keys.

It's not about the 127 steps,

it's about key --> curve --> synthesis --> sound

Always look for the real weakest link before wasting time on the APPARENT problem.

[ozy]

oops. sorry. double post. here's the removal line.

[Scott]

It is the MIDI INTERFACE, not the keyboard scanner, whcih takes veloicities 5to15 and outputs them all as 10. that's the CURVE problem)

I'm not sure where the calculation takes place, but the process begins not with a "velocity" (i.e. 5 to 15), but with a measurement of time... the time it takes for a key to trigger past one sensor and reach a second. Somewhere, that number of microseconds is converted to some number between 1 and 127 to arrive at a MIDI velocity figure. My understanding is that it can be done in a linear fashion, or in a curve... i.e. they can generate a curve before the calculated velocities are ever turned into MIDI velocity values. The method you describe, of generating a curve after the velocities have been turned into MIDI values, would be far more limiting. It could decrease resolution in some areas (as in your example) but it could not increase resolution in others.

[ozy]

The method you describe, of generating a curve after the velocities have been turned into MIDI values, would be far more limiting. It could decrease resolution in some areas (as in your example) but it could not increase resolution in others.

Totally agree.

I don't say i LIKE it. I say that I'm afraid that's how it's done in current keyboards.

In NO commercial keyboard reference guide did I find any reference to "stretched" keyboard curves (i.e.: increased resolution [better than 1/127] at, say, low and mid levels, lower resolution [lower than 1/127] at higher levels).

which is something that could make sense.

All reference guides talk about "here's how the linear response is converted into output values to the midi out: curve a) lower velocity [already measured, i.e. already quantized] results in higher values, curve b) viceversa".

I am afraid that from this kind of languae we could deduct that conversion happens within the midi (quantized) domain.

Doen't this make sense to you?

That could explain why curves are so bad.

Repeat: I don't say i LIKE it, I say that's how it currently happens.

Any documented refutation?

Any example of synths or master keyboard documentation providing a "velocity [in non-quantized terms"] to midi" conversion table?

Everywhere I find bland charts with 1/127 linear chart on one side and various 1/127 curves on the other. Stop.

I'd love to put my hands on the rh3 or pc3 keyboard specifications...

Now that you mention it: I could do it by taking voltages out of my doepferized fatar.

anche cross-check voltages with midi. It has both midi out and control voltage

Anyboady can help me with the practical details?

Charting a midi out I can. Charting in parallel a cv, I have no idea how.

[synthjoe]

...127 STEPS, which means several thousand combinations of them (velocity+duration, velocity+release velocity, relative velocity in trills, etc)

I respectfully disagree, simply based on the mathematical fact that the function x*y, where x is a positive integer between 1 and 127 (repesenting the velocity value) will always give you 127 distinct values. This is regardless from y being a single number or a complex expression (e.g. x*(y+i/j)). Hence, x is limiting the resolution of this 'curve'. I dropped the idea of highlighting that in your earlier post, but now I have to point that out.

I agree on the general disconnect on digital keyboards (mainly for simulated actual physical instruments) between what you do and what you hear, that's also my problem. Also agree that the lack of more than 127 is not the most important culprit in this issue.

I don't think velocity curves are calculated as you describe, however - I believe it happens in the quantized domain but not in the MIDI domain. I don't have the facts for actual manufacturers' product but I have the experience of attempting to build a velocity sensitive keyboard.

The time difference between altering the two contacts is likely measured linearly with a counter, it would be difficult to do otherwise (another - better - alternative would be to charge and discharge a capacitor, but that requires intricate electronics combined with fast and accurate A/D conversion, which is expensive).

A counter requires a clock pulse fast enough to yield a different value for the very loudest and the one notch softer keypress - the interval difference might be as small as a few or a few ten microseconds (us), I guess. Let's assume that the softest touch takes 100 ms (miliseconds) between contacts and we want 10 us resolution (e.g. the loudest key action takes 5 ms or 5000 us and the next less loud value you want to distinguish is 5,01 ms or 5010 us). In this case the counter must be able to count hundreds of ms with 10 us accuracy, that means a magnitude of 10,000 clock pulses. The difference of 10 us is just sufficient to differentiate between loud keypresses, whereas the resolution is wasted on the 100 ms end of the measurement. Therefore you need some transform (velocity curve) to convert this exponential(ish) function into a 'linearish' 1-127. Conclusion: a gross truncation or rounding has to take place at the large value (soft touch) end of the scale

I'm only speculating, but this is what two physical contacts in keaboards let me conclude. I'd be interested in a different (e.g. piezo impact, inductive or optical continuous) velocity sensing mechanism - or seeing facts about my speculation (pro or contra), just like you.

[synthjoe]

anche cross-check voltages with midi. It has both midi out and control voltage

Anyboady can help me with the practical details?[/b]

Charting a midi out I can. Charting in parallel a cv, I have no idea how.

Maybe applying CV to frequency? This way you have a voltage to frequency converter and you can record both audio and MIDI in some DAW software. Don't ask me about analyzing the results, though. For that we'd need to know for sure the Hz/V ratio curve of the oscillator.

[ozy]

I don't think velocity curves are calculated as you describe, however - I believe it happens in the quantized domain but not in the MIDI domain.

your hypothesis is sensible, and I am far better at playing (and at knowing how fingers work ), and at programming synths, than I am at electronics (where I am definitely an end user with no time spent on even a soldering iron, much less on a blackboard).

I'd be happy to be certain of what you guess.

Some research among manufracturers' documents would be undoubtedly interesting.

If what you say is true, then any hypothesis of "better definition" based on current technology (as in "they should give us better definition in midi, they are robbing us", which is the starting point of this thread) is wishful thinking.

New technologies are another issue. They'll be always welcome.

But that's a function of 1) capitals invested and 2) minds being applied to materials.

Maybe applying CV to frequency? This way you have a voltage to frequency converter and you can record both audio and MIDI in some DAW software. Don't ask me about analyzing the results, though. For that we'd need to know for sure the Hz/V ratio curve of the oscillator.

that is a known parameter, on a V/oct osc: double frequency means 1 V.

that would create a range from say 55hz to 1760hz, equivalent to 0/5 v output from the keyboard sensors which could be used to test velocities from "just stand by and watch at the keys" to "let my fat ass fall on the keyboard".

You try to do the math (is that range enough for extracting meaningful fractional data?). After a day spent at classifying ball rotation speeds on excel, I've lost any taste for further mathematics.

But IMO what you said above somehow devoids such test of much of his meaning.

[X-Trade]

Ozy, I enjoyed reading your bit about aftertouch.

I've noticed this too, although hadn't really put much thought to it before. My old Kawai K1mkII had a really deep aftertouch - you could actually lift yourself off the floor on your fingers to get 127, whereas my KARMA and formerly TR seem to produce 127 at just a slight push.
I'm told the Kawai was 'faulty' in that the aftertouch degrades in sensitivity after time, and I admit it required some pretty brutal force to get full on, but a sensible range would be somewhere in between. Somewhere where you can at least achieve a range of values progressively.

Similarly, I think 127 steps is plenty (for velocity, or aftertouch. Not necessarily for controllers), but that their response range is often poorly calibrated. Every player is different, and every real piano is different, so you can't cater to everyone's needs. Rather than velocity curves, keyboards need more personal calibration features