Topic by mabry

  • Avatar
  • mabry
  • Member
  • Offline
  • Registered: 25-02-2010
  • Posts: 118

Topic: A 2-string acoustic piano VS A 3-sample virtual piano

The two string acoustic piano has middle C and the E above. Both of the hammers only move at one speed so all notes sound at the same volume. It did not come equipped with any pedals You can get three sounds with this piano:

1) C alone
2) E alone
3) C and E together

The virtual piano was created by sampling the 2 string acoustic instrument. Two samples were taken:

1) C alone
2) E alone

The two samples were taken with the finest recording equipment known to man and everyone who has heard them individually swears they cannot distinguish them from the individual notes on the acoustic.

Question: when the two notes on the virtual piano are played together would they sound the same as the C and E played together from the acoustic? It seems to me that they would not. Would not the C and E vibrating together cause the piano soundboard and other parts of the instrument to vibrate in ways that would impart different characteristics to the resultant sound that would not be duplicated with the virtual instrument? Especially if the virtual instrument's two notes were, after all, at least somewhat distinguishable from the acoustic's?

If the path upon which I now tread is heading in a right direction, then this helps explain to me some of the inherent difficulties in creating a truly realistic virtual piano based only on samples. The imaginary virtual piano would only need one additional sample to be as near-perfect as it could be but a near-perfect one with a full complement of 88 keys would need so many more...and way too many to ever be taken, especially since they would need to be created at multiple volume levels in order to create a truly functional virtual instrument.

And so, instead of a full complement of samples, we get a full - and getting fuller! - complement of compromises: Alicia's Keys, Galaxy Pianos, Garritan Authorized Steinway, Ivory, True Keys...and yes, even the algorithmic Pianoteq.

Quiz Time: assuming no pianist alive can play more than 10 simultaneous notes at any given time, how many samples would a single-volume level, 88 key virtual piano need? In other words, how many chords are there, from 1 to 10 notes, on an 88 key piano? I suspect that a mathematical equation for determining this number would contain an 88, a 10 and perhaps a factorial symbol in addition to other mathematically things but I have no idea what its final form would be. Anyone?

Reply by EvilDragon

  • Avatar
  • EvilDragon
  • Member
  • Offline
  • From: Croatia
  • Registered: 27-06-2008
  • Posts: 3,994

Re: A 2-string acoustic piano VS A 3-sample virtual piano

mabry wrote:

Question: when the two notes on the virtual piano are played together would they sound the same as the C and E played together from the acoustic? It seems to me that they would not.

Of course they would not, there's no sympathetic resonance when you just play back two samples. This is where Pianoteq excels at.

mabry wrote:

Quiz Time: assuming no pianist alive can play more than 10 simultaneous notes at any given time

Very bad assumption - one can play more than one key with a thumb, and there's sustain pedal that can create a lot of sustaining notes. The simple answer is - this is not feasible to do at all with samples.

Hard work and guts!

Reply by mabry

  • Avatar
  • mabry
  • Member
  • Offline
  • Registered: 25-02-2010
  • Posts: 118

Re: A 2-string acoustic piano VS A 3-sample virtual piano

mabry wrote:

Very bad assumption - one can play more than one key with a thumb, and there's sustain pedal that can create a lot of sustaining notes. The simple answer is - this is not feasible to do at all with samples.

Yes, certainly not a valid assumption, I am just trying to get an idea of the extreme number of samples that would be required in order to even approach something that is realistic. I started thinking about this after I saw the True Keys website proudly claiming that

"The sympathetic resonances within a real piano are what bind the sound together to make the sound rich and full. Unlike other sampled and digital pianos, True Keys uses real samples for both sympathetic and sustain pedal resonance."

Good luck with that, guys. You did a good job on your latest release - I have the True Keys Steinway - but I remain convinced that an algorithmic approach to resonance is where the best solutions will be found even in sample based instruments. So go, Modartt!

Reply by custral

  • Avatar
  • custral
  • Member
  • Offline
  • Registered: 22-08-2010
  • Posts: 288

Re: A 2-string acoustic piano VS A 3-sample virtual piano

Well, take Modartt's 3 levels of Hammer Hardness visible (and it's probably the applied total) as being one level of simplification over reality, and contrast with what looks to be the Truekeys equivalent (the Tone control, top right) as being a 3 times coarser simplify, you have a one-string measure of the applied simplifies; and in Modartt's case this probably affects only the 10 overtones it exposes controls for (per-string in the case of Pro, across all strings elsewhere), versus zero control (or one timbre solely, or one truth-table of timbres solely, applying across the keyboard for Truekeys). Two different simplifies, Truekeys's lots coarser, compared to Pro.

But I take it you are interested in two-string interactions, like with one key held down and the other struck, or both struck. There's got to be truth tables which apply for all notes as to which neighbor strings, above, or below, are sympathetic, and these have to be limited in number by their *appreciable* effect, and carry amount-measures as entries. PTQ has one control for these, and whether it's applied to two tables only across the keyboard (seems less likely), or (seems more likely) applied in a per-truth-table-pair table-of-entries up the octaves of the keyboard, is hidden inside PTQ. Why those "seems likely"s exist is, something must provide for PTQ's individuation between instruments, and this area (cough) seems quite likely - consider the Historic Instruments, and how small the files are to carry their info, even while the effects vary widely per-instrument.

Contrast this with what we can see for an equivalent in Truekeys (the Harmonic Polyphony truth table limiter, and its associated Amount knob in dB, bottom left). Presumably as you make the Polyphony less-limited, and the bang-per-buck achieved by further increase doesn't justify the CPU cycles, Truekeys silently stops applying any truth-table entries it even records (the mechanism would be, sound any paired-note table-entry at the corresponding Amount-modded level). I don't see why this shouldn't be as complex as PTQ's second "more likely" approach, though of course Historic Instruments become impossible.

Summarising, the *appreciable*-effect limits the array-size of the needed truth tables involved, so (though a truth table of samples has to be byte-heavier than one of note numbers, while a sample-trigger has to be time-cheaper than a note generator), I can't, just now, see why one approach beats another, as to complexity.

And hence richness of output. Interesting question. Different from the overtones of a single string.

ADDED: I guess, if we can hear so fast, and finely, a race goes on between make-a-sample sound, and generate-a-sound. If one wins, we have the opportunity to hear the result as more immediate, more clear, articulate. But then it's up to our ears, and with advancing tech, any diffs will become less important over time.

Last edited by custral (04-09-2013 06:00)