A theoretical limit on sampled piano dynamics

[kensuguro]

Since the v-piano from Roland is going to cost bazillion dollars, and pianoteq won't be coming out with a new version yet, I was thinking how would one make a good sampled piano. Not that I can make one, but how would it be done.

I think the primary problem with most pianos is that they boast extreme dynamic range. I think that's going for too much. Every keyboard controller has a hard limit in terms of the maximum velocity it can take. If a specific velocity registers as 127, that's it. That same velocity, played on an acoustic piano, is the loudest the sample library can go. If it goes beyond that, the range is too wide for the velocity range, and the result is that there will be no happy medium between the MIDI velocity and the sample loudness because the relationship is not 1:1. Meaning, if the velocity range of the sample lib is greater than the keyboard controller's range, then the keyboard controller is too responsive... which seems to be the case most of the time.

What I feel is that many sample libraries go way far beyond a typical midi controller's velocity range. For most libs, the 128 sample is as if some one elbowed the piano. Although that's possible to do on a piano, and some music may call for it, I think it's safe to not include it in the lib.

Well, then what's the problem? The problem is that although MIDI is a standard for file format and communications, it's not a hardware standard. So, a velocity value of 56 is just that, a value. The velocity isn't directly tied to the physical velocity of the key (I think). It would have been nice if MIDI velocity was directly related to a function of the key strike start time and key strike end time (bottom out) since the delta is how machines derive velocity anyway. But that would pose all sorts of hardware issues since essentially, all keybeds need to react in the same way, having similar resistance, and end up feeling similar. (impossible with synth keys and weighted, semi-weighted, etc.. even acoustic pianos aren't consistent) I'm guessing the delta times on a synth key versus is a weighted key is significantly different, the weighted key delta being larger... seems to make sense with regards to physics..

But by limiting the velocity range of the sampling session, to match the actual velocity range of typical controllers, sample libs can be more "focused" to a particular range. If the maximum velocity is generated from a 300g object dropped from 40cm above key top, with a 2cmx2cm contact point, then the lib would be 20 velocity layers within that range. That saves resolution since the range is not exaggerated.

The hard work would be to go through the most popular piano controllers from various manufactures, and to collect all velocity data. Then all you need to do is to find the one with the highest and use that as a reference. I wonder what the difference between the highest and lowest velocity range would be. (if it was just full weighted controllers taken into account)

Uhmm.. so that's my business pitch. Anyone care to pick it up? lol. The idea's gotta worth at least $1.5mil.

[Neutron]

ok give me 1.5 million and ill do it for you
soon the piano libraries are going to be so big that you will need hard drives that take up more space than an actual piano.

[dawman]

Well as much as developers keep trying they might have it by 2025.
I still think my 80 USD Library and my 300 USD contoller are as good as it's going to get for a while.
Every year I go and demo the new hardware digitals that cost as much as a real Piano and they still suck too.
I even buy a new Library every year praying it will sound as good or better than the Sample Tekk, and some have a slightly better tone up top or below, but have their flaws as well. The only way to trigger the different layers is by using a sequencer and spending time editing and drawing unnatural curves, etc. By that time I could have recorded my Acoustic and still have a much better sound. That's why after seeing the best on offer at NAMM, I still walked away shaking my head. But hey, after 25 years of developement they are finally adding the una corde and sostenuto capabilities...
Pianists do not use these libraries no matter how many Jazz or Classical performers they bribe and sponsor. Even the hardware digitals are only used for the sake of conveinence.
I watched Herbie Hancock on a Fazolli w/ on the Elvin Bishop show, and you'll never see him performing on the " Authorized Steinway " library, or any of these other ploys that plague us.
I know you are probably living in cramped quarters, for a Grand Piano that is. But someday you will play one, and I know you have played many. But one day you will probably see one stuffed away at a theater and you know after 30 seconds of playing..........that this is the one !!
There are only a few great Pianos in one's life, but once you play it, you gotta have it.
Then you'll soon forget all of this virtual bull shit that we unfortunately have to make due with.
Having that Piano in your house is the most precious thing in your life, besides your family and your talent.
But they do not need tuning maybe once every ten years after a real pro gives it a working over.
I found that Piano 30 years ago and still have it. It's a cheapo Upright, and the other one is the Grand Piano my Grandmother left me which is at my parents house. Whenever I visit I play for hours on end.
That's when you know you got a good one.
I wouldn't lose any sleep over this stuff if I were you.
Nobody is interested in doing a quality library or a quality controller that can come close to a real Piano. The technology is available, but the competition would cause it's price to be too high, and since these companies do not care about anything other than how much they can profit, we won't see one for several years.
That's why I keep checking in case, but always go home to the Baldwin or the 80 dollar Black Grand w/ no pedals and five different mic positions....

[Neutron]

another bitter vegas whore
next on CSI. musician would not toe the line with the plugin developers, so he got plugged!

[kensuguro]

hehe, ya, the grand that won't get out of my head is my teacher's 7' yamaha, and a smaller conservatory Kawai at my grandma's place in Taiwan. The kawai's a vintage one built before a bunch of technicians got brought to Yamaha. Cool thing was that I could play it lid open without getting yelled out, throw pencils in it, tie ribbons on the strings, do whatever I want. (to everyone else it was just a big piece of furniture) I think it was one of my favorite toys growing up. Every few months I'd come up with something new to do with it. I once even had the tuner tune it to werkmeister I.

Now I realize how lucky I was, having a grand in the first place, and also having a living room to go with it.

[chriskorff]

I know what you mean Ken,

We didn't have a grand in a massive room, admittedly, but I have some extremely fond memories of the upright I had in my family home. Unfortunately (and I see this as a mistake), I decided, after a couple of years on the keys to learn guitar instead. Which I succeeded in, by the way, but who wants to hear shredding and tapping these days? NO ONE!

BTW, and back on topic (oh shit... this IS the off topic section!)

I'm pretty sure that this dude didn't spend 1.5mil on making this free, multi-sampled Ldwig drum VSTi
http://www.bluenoise.no/mydrumset.html


Though at the other end of the scale, I shudder to think how much this one cost to make:
http://www.soundonsound.com/sos/jan08/a ... saurus.htm

Cheers!

Chris

[braincell]

That was thoughtful Kensuguro. It will will get better. MIDI is decades old and needs to be totally updated with more resolution and to include audio and wireless connectivity. If people are worried that their current midi gear will not connect, too bad, that is the way technology goes some times. I mentioned this before in another thread and some moron said "No, it's fine the way it is".

I like computer control of a piano. I like spending hours editing the midi notes to get the exact velocity. Some people don't edit their velocities and that is why their music sounds like crap. I have fooled professional classical musicians into thinking I play a real piano. Listeners can not tell the difference and that is what counts.

Everything has a purpose. If Yamaha comes out with a similar piano to the V-Piano, the price will drop dramatically within 5 years. Since Yamaha owns Steinberg, they could have a better interface to it. Yamaha has started working more closely with Steinberg in Cubase 5. At this point I have no reason to buy a piano if I could afford one. I don't want a tuner to come to my house. There is also all the expensive equipment it takes to record a piano the right way not to mention room acoustics. You are talking a hell of a lot of money to do it right.

Acoustic grand pianos sound and feel nice but that is no reason to trash talk technology (by the way different brands and models sound totally different from each other). Progress is a good thing. No matter how advanced the fake ones get, there will always be some asshole who says it is not the same as an acoustic. That is not relevant. It's a musical instrument, and a nice one. To call it "virtual bullshit" shows a lack of imagination. I really don't know what your problem is Jimmy. I predict Kensuguro will play the V-Piano or something akin to it and love it.

[kensuguro]

got a point there brain. acoustics are acoustics, at this point in sampling technology, they're very much separate things. I think both Jimmy and I (and the rest of the world) are aching to play a physmod piano that makes us go "hey, that sounded real". Maybe with vpiano, maybe not. Everyone wants the same thing though, a good sounding physmod and an awesome full weighted controller. But after these things pass the "real" test, I think there is another threshold that these things have to pass before we get into the "sound good" criteria. That's where things get interesting.

The thing that you say about a listener not being able to tell if a track is real or not, is one theory of sampling I think. As in, a library is a set recordings you can patch together, and if the results sounds good, it's all good. That's the way most sample libs are built I think. It's essentially a bunch of recordings you can choose from, and the keyboard controller is just your switch. That is for production. But for performance, the theory is totally different. Obviously everything needs to be final, and the connect between the key input and sound output needs to be learnable, preferably very close in line with a real acoustic so it doesn't mess up your touch. You may want to edit a mis-executed velocity, but you can't once you've played it. With performance, the main thing isn't accessing a huge pool of recordings through a mysterious interface. Everything needs to be crystal clear, simple, and straight forward. You think a note, play it, and get something close to that note. The value of the instrument can be judged on just those three steps as far as performance is concerned. On the other hand for production, it's just the end result that really matters. So the criteria is different. I also think that many libs can't get out of production use even though the libs were created to be performance orientated, simply because the can't get it to feel right.


I just think sampling 5000 layers with truck loads of HDD, like the new Bosendorfer Imperial seems to be energy directed in the wrong way. I swear by the time people are done sampling, we'll be running them at remote datacenters with mainframes and stuff. So, why not cut out all the velocities you won't ever reach on your keyboard, focus all the sampling efforts there, and be done with it. The way I see it, if you lock down the sample mapping method, but vary the sampling technique (micing, etc) the consistency in "feel" would make switching between different libs a lot easier.

[astroman]

well, I don't see a problem here, or a contradiction...
if you want to enjoy the full 110 db dynamic of the real instrument in all it's nuances - use the real thing
embedded into some music (tracks) or live on stage transmission and processing will restrict the instrument's dynamic anyway - so you'll get along with midi resolution just fine (imho)
I don't see it from a dogma point, but just practical...

You can improve on the keyboard mechanics, though - Kawai uses a system to measure the real impact of the hammer and I've found their stage pianos to be outstanding in response and well tuned, even with realtively small sample sets.
I've mentioned it a couple of times that huge samples of long decay notes are a pure waste.
They are only 'realistic' in case you play the exact same note with the exact same intensity - in all other cases it's the envelope generator's duty anyway.
If you carefully pick the proper sample sections from a good recording and manage to setup the 'control system' (envelope, filter, additional tones) properly, you can make a realistic grand from 256 MB or so.

as mentioned, it starts with the mechanic
Kawai is probably not the only company using this hammer-impact technology.
With 'synth-velocity-measuring' you've already lost before you started.
Even with plain Midi there's no restriction to have 1024 (or more) velocity steps driving your custom controller.
The point there are no such devices is that they sell to a market, not to snobs
Noone will spend 5 times the price for his stage instrument if you cannot even hear the difference ina rock 'n roll band or a jazz club or whatever - just due to plain background noise...

It's not a problem at all to build such an instrument, but it's a problem to sell it

cheers, Tom

[JoeKa]

well Ken, if you need the very large range of physical velocity spread to be happy, get a keyboard that offers the right feeling. I find my old GEM equinox pro 88 does have a rather large range, even the softest velocity curve invites to strike it pretty hard to get the max value out of it. the rest would be your own problem, adjusting the velocity curves to your liking so the different virtual instruments react in a similar way. It would, of course, be good if the sampleplayer of choice would offer such a unified velocity curve adaption individual to each library, rather than having to re-tweak all the samples one by one.

Tom&topic: I agree with most of what you said that, in a band context and a sensible limit to the sustain length of notes, a lib of <256mb can easily be good enough for very nice results indeed. But: the matter is different with what today's user has taken to wish for: a lib that will sound very nice not just in a given, rather limited musical setting, but equally fine on its own and fulfilling the high demands of classical solo quality in one. I'd also go as far as to say, a lib tuned to fit the need for such quality demands does not have to be several gigabytes in size either, but maybe 512mb for satisfactory results. It will be questionable in the end though, if the classical lib would make the right impression when used in the band context, which I doubt.
But as people have come to think, that if both venues can use the same real instrument, a virtual instrument has to match this and sound as "universal" as possible, making it necessary to cover the whole range of what the instrument can sound like and put it into single, gigantic libs that are outgrowing one another every other year. Imho it is only due to such overinflated libs that the question for more dynamic controllers ever surfaced. I don't feel a need for more than what we have now, in this matter: if there are already libs that are good enough to fool even trained classical players, that is good enough for me.

[astroman]

c'mon Joe - we're not in a bazzaar here, 256 or 512 ... what's that compared to several GBs ?
we might agree on 376, tho

anyway, my argumentation has a severe mistake
a pianomaker could drive it's (on instrument) sound generation with higher resolutions than 127 midi steps - while response to midi data is scaled down.
Scope synths do the very same, and as you can experience yourself, some clever internal programming can overcome this limitation quite easily.
I never understood this bigger-is-better approach, but it still seems to sell things.
Of course not restricted to musical instruments at all

I have not enough experience with real world pianos myself to judge what part of the mechanics and resonance body has which kind of influence on the tone of the instrument.
But I wouldn't estimate it less complex than guitars and basses, acoustic and electric.
Sixstring is a very good example that a perfectly emulated string doesn't sound anywhere like an 'instrument'.

cheers, Tom

[garyb]

buy a #$%^ piano.

[JoeKa]

I just referred to these typical sizes reminiscent of the last generation of serious hardware samplers and what current rompler pianos are likely to be equipped with, too. In computer based sound reproduction, these values are not of importance but merely to be regarded as a reference. if you know what you need, you won't need all at the same time. If you feel you do, however, cough up for what you want. that's how it has always been.

True about the higher internal resolution, haven't thought about that myself either. Then again, I've had my mind set on midi only as we spoke about controllers. I'm not sure a USB connected keyboard would be able just now bypass that snag of midi, as what the appz do record is bound be the limiting factors of the current midi specs, too...

[Tau]

A funny quote from the Mixosaurus review linked by Chris:

As the manual states, “in a normal drum groove, Mixosaurus will continuously play 100-200 stereo voices”

O...K...

[braincell]

I haven't used many sampled pianos but I can tell you that so far, the larger they are, the better they sound so astro's comment that size doesn't matter is rubbish in my personal experience. Dynamics and piano action needs improvement. The comment "buy a piano" is also a bullshit statement. It reminds me of the redneck bumper sticker you see "America, Love it Or Leave It". I hate that attitude. They are not perfect but they will continue to improve. That should be obvious to anyone. Physical modeling probably is the future or some sort of hybrid. I listened to synful orchestra recently and was not impressed. It does do articulations well but it does not sound as nice generally.

[garyb]

awww c'mon man! my humor isn't bull pucky....

besides, it IS funny when people obsess over an artificial something when the real thing is a reasonable option...

[braincell]

I am sorry Gary.

If you just want to just play piano by yourself, that is fine but there is a cost to recording; an enormous cost if you want to do it right. We are not just musicians in this group. I assume everyone wants to record music at home.

[astroman]

Braincell, you may have ignored the context in which I made the 'size doesn't matter' comment
'Your' libraries lack it completely
they are a brute force attempt requiring minimal mental effort, highly automatable
... with the tremendous advantage that quantity sells easily and the media itself acting as a kind of copy protection

afaik the Kawai 9500 was sampled to (say) 1 GB raw (some 10 years ago)
they picked representative parts, made the programming, tuning etc and then burned it into a ROM
guessing from the time of release it's size must have been something between 32 to 128 MB

did you ever play a 9500 or the 2nd version of the GEM Realpiano ?

I don't question that your samples are recorded brilliantly (probably better than the ones I quoted)
But if either Kawai or GEM would use a tiny part of that sampleset and apply their technology to it, then it would be you who couldn't tell the difference.

The important point is that tuning a sampleset is a sh*tload of work and it requires an expert.
A huge library (once the raw data is taped) can be done by an automated process attended by a student or part time worker.

The seller can brag with the biggest and best record ever done in the universe
but in the end they sell a piece of poo and I really mean poo, because if it was properly crafted, you wouldn't be bothered with disks, streaming, memory and all that stuff

these people create a product, because they make a livin by the revenues of a product
they don't serve you, or us or anyone
I won't repeat myself, but doesn't the pattern read familiar ?

[braincell]

I haven't heard the Kawai 9500 so I will not comment on that. I know their acoustic pianos totally suck, at least the ones I have played.

[LHong]

Other Solution?
Roland's Expansion Board SRX-01 (Drum) & SRX-11 (Piano) - $249.00 sounds pretty good!
It has 4-stage velocity switching employed of total 704 samples – 64/128/256 MB. Listen to the Demo that what we are expected!
http://www.rolandus.com/Multimedia/Flash/srx/index.html
It works conjunction with the Fantom-XR Sound-Module - $1,495.00 or a Fantom-X Keyboard.

Long