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

Hi all,

I have a request. Could someone send me a .wav file of a DP that uses sampled grand data sets that includes a 10-15 second held C1 (lowest C on piano) with no use of damper pedal, followed by 10 seconds of silence followed by another 10-15 second held C1, this time with the damper pedal held down before the note was played. Here's why.

On my acoustic, when I play C1 with the damper pedal pressed, I can see that every note on the piano vibrates, not just the sympathetic resonances one would expect. This effect doesn't show up on a Pianoteq model I tried. I am interested to see if this effect is modelled into actual recorded sample sets.

Here are graphs of my acoustic C1-no sustain, followed by C1-with sustain, followed by Pianoteq C1- no sustain, followed by Pianoteq C1-with sustain. As you can see in the acoustic C1-with sustain every note shows energy. This can be affirmed by counting the number of spikes between the second and fourth tall spikes (around 63Hz and 125Hz) They represent the second and fourth partial of C1 which correspond to C2 and C3 on the piano. If you count the second partial as 1, you will count 13 spikes when you stop at the fourth partial. Those are the 13 notes on the piano from C2 to C3. While the amplitudes of the notes is low, it has an overall effect on the total sound.

Thanks.

What do you mean by a .wav file of a sampled grand? When recording an acoustic piano, grand or upright, people need microphones, or something like a Zoom H4. The companies sampling grand pianos to release as software pianos, use high quality microphones, usually in pairs to get a stereo effect. The average pianist may not have such equipment, and a Zoom H4 is often the choice, so it may be hard to record an acoustic grand piano and get a sound like the commercial software piano manufacturers get. Some members here might be able to achieve that sound, though.

edit:Thanks for catching my confusing request. I meant to say to record a DP that uses actual acoustic grand samples rather than modelled data.

By .wav file I just mean to record your DP using the Line Out or Headphone jacks on any PC computer or H2N or whatever. I just don't want a lossy file like .mp3.

As you will notice, my grand (the acoustic in my post above) obviously had to be recorded using a mic. I used a mono reference omnidirectional mic through a Roland Quad-4 into a PC running Audacity from which I exported a .wav file to my FFT analysis software which produced the graph I posted. (I normally record using two cardioid vacuum tube mics and get a sound that rivals precisely what I hear when sitting at the piano bench. You can check my recordings on the Pianist Corner Member Recordings area.)

Someone else sent me a .wav file of his Pianteq modelled sound.

If someone here has a DP that uses a actual sampled data set, and looking at your byline, I guess you do, all you have to do is record your DP and send me the file.

Hope that explains it.

Thanks.

edit: I actually run a Yamaha KX-8 to my iPad and use iGrand with the S&S D sample set when my own piano is in use by others or I am on the road, but it doesn't represent the state of the art. Still nice to have though. I also own a CLP-300 for baroque pitch practice at 415Hz and 392Hz.

Sampled pianos do not capture sympathetic resonances.
When the damper pedal is pressed they simulate the effect by mixing harmonically related samples.
So the answer to your question is: no.

OK. So you are saying that an FFT analysis of a DP using a sampled acoustic piano will look the same whether or not the damper pedal was pressed when the DP player struck the note. This is not the case for a modelled piano as can be seen in the example above. While very, very weak, there is noise added to the signal in the modelled DP sound when the damper is pressed. There is no major excitation of the other strings as is clearly seen in the acoustic grand.

Thanks, but I would still like to see an example I can analyze. Given the amount of dis-information available on the 'net and here at PW, I prefer to make my own analysis and show reproducible results. This not a criticism of your response - I appreciate your candour, but as a scientist, I am skeptical until proven wrong or right.

FWIW, have you had your wife sing while you hold down the damper pedal (on your acoustic) and don't play any notes? You will hear a BUNCH of strings vibrate, but the most distinct will be the fundamental pitch she was singing. It's pretty cool.

And I believe that digital pianos attempt to emulate this - this is what I understand sympathetic resonance to mean when described in DP manufacturers' descriptions. But I can also understand how this might be difficult to sample in such as way as to be able to modify it. So when I see DPs that have this feature that can be adjusted, I understand it to mean it's a simulation and not a part of the original sample.

This is definitely NOT what I'm saying ...
However, I did say:
-- "When the damper is off they simulate the effect by mixing harmonically related samples."

That's not what I meant. I've corrected that post to say:
-- "When the damper pedal is pressed they simulate the effect by mixing harmonically related samples."

Perhaps that error caused a misunderstanding?

I did not misunderstand you. If you look at the 3rd and 4th graphs I posted above, which represent a Pianoteq C1 struck with the remaining strings damped and a Pianoteq C1 with all strings undamped, you will see essentially no difference except for a bit of added low level noise. This is very clear. Yes, the model may have added, hopefully, inharmonicically related samples, since no piano string produces harmonics, but that is another discussion - I get your point.

All I am asking is for someone to send me three samples from a DP that uses a data set from an actual piano to create its output - 10 seconds of background line out noise, 10 seconds of damped strings with C1 struck, and 10 seconds of undamped strings with C1 struck. I know for a fact, having played some state of the art DPs, that one can hold down, silently, C1, strike any related partially related note - C4, G4, and C5, for example, and C1 will ring with sympathetic resonance at those pitches, so there is some effect added. I just want to determine precisely what it is.

I'm puzzled at what it is you're trying to do. But maybe you can clarify. This must surely be a typo?

PMFJI --

I was the source of the Pianoteq file. I was surprised that they didn't do a better job of modelling the "sonic wash" that comes off a few hundred undamped strings.

Here's the thing:

As Prout has shown, the sound signature of an acoustic piano -- with one note struck -- is substantially different, with "pedal up" vs "pedal down".

I would expect, that when someone _samples_ an acoustic piano, to develop a _sample-based_ software piano, they would sample each note two ways:

a) with the pedal up (dampers down), and

b) with the pedal down (dampers up).

That way, their "software piano" would have the same tonal difference as the acoustic it was based on. If the pedal were up when a MIDI signal came in, the "pedal up" sample would play; if the pedal were down, the "pedal down" sample would play.

[I don't know how they'd handle transitions from "pedal down" to "pedal up", when notes were sounding -- maybe that's where serious audio engineering skills are useful.]


So, if someone with a _sample-based_ software piano would be so kind as to turn off the reverb, and play it as Prout requested, and get a WAV file to him, both of us would appreciate it.

I have the MIDI file I generated, playing Pianoteq. If anyone wants it, for rendering, I'll send off a copy.

Thanks --

That's depends : EWQL and VSL use damper pedal pressed recorded sounds.

The result is more accurate, but when you release the damper pedal, the virtual piano doesn't switch back to a normal sound when continuating to play already played notes. I suppose it would be hard to have a good enough switch.

Looks like the sonic wash on the Pianoteq is simulated, but at much lower amplitude compared to the extra peaks on the acoustic. Also, the PT "sonic wash" appears to arbitrarily jut up from the original contour, unlike the acoustic. Would say they've got work to do on their algorithms.

Would be interesting to compare the V-piano or RD2000/LX17 "sonic wash" (like the jargon btw) with PT and acoustic grand's.

No typo. I am merely being pedantic. Bear with me a moment.

If you bow a violin string, it produces a series of harmonics, exactly related to the fundamental pitch (correctly called the first partial) by integer values, that is for example 100, 200, 300, 400, 500Hz...

If you pluck a violin string, it produces an inharmonic series of non-integer partials - 100, 201, 301.5, 404, 507Hz...

A struck piano string does the same thing as a plucked violin string. It produces inharmonics, and that is what makes a piano sound like a piano. It is this inharmonicity that requires that the tuner stretch sharp the treble and the stretch flat the bass in order for the octaves to be more or less in tune, since every octave on the piano is a little larger than the 2:1 ratio we were taught.

What I am trying to do is simple. Every one of the roughly 240+ strings on a acoustic piano vibrates a little when a low bass note is played AND the damper pedal is pressed. Does a DP do this as well?

Absolutely. It is an effect we sometimes use when performing modern music where the sympathetic resonance of the piano is part of the sound of the piece. S. Barber uses this effect (though it is not noted in the score) in some of his art songs. It does help the audience to hear it if the lid of the piano is on full stick for that piece.

Tests conducted:

Mason & Hamlin BB

Pianoteq Grotrian Concert Royal

Your DP here?

Does my CN35 qualify? Kawai PHI uses a sampled attack and then loops, but also has modeled sympathetic resonances. I could turn the resonances down with Virtual Technician.

There are actually DP recordings linked to in the DPBSD thread, that should have all notes with and without sustain.

Seems to me your ears will tell you the answer and a bit o' thought will tell you why.
No digital piano sounds like an acoustic, because as stated, all strings as well as other things resonate.
The samples are cleaned up, and only those resonances deemed sonically acceptable to the great buying public will remain, and even those will be processed in some form to suit the overall sound.
Some models can be adjusted to personal taste within that which is supplied.
You are getting white bread with bits in it.
I would suggest the same raison d'etre applies to modelled stuff irrespective of how it's put together.
Some of us like it that way.
Having said all this, some acoustics have fairly low resonances. MY old straight strung upright was one.

Garritan CFX shows the number of "layers" being played at any given time, which I assume are the voices stitched together to produce one or more keys played. With damper off, any given key pressed and held varies between 1-3 voices. With damper pressed, the player also shows 1-3 voices for any given key, suggesting it's not mixing the regular samples from other keys to achieve a damper resonance effect. IIRC it's the same case if you silently hold down keys to trigger string resonance and then play a single note (only 1-3 layers are triggered).

Example here

I happen to have a synth/DP which does both - sampling and modeling.

00:00 : sampled piano + modeled string resonance, C damper pedal up
00:09 : sampled piano + modeled string resonance, C damper pedal down
00:17 : striking chords, which "transfer energy" to the held C because of modeled string resonance

00:38 : sampled piano only, string samples
00:41 : sampled piano only, resonance samples
00:47 : striking chords, no "transfer of energy" to the held C because no modeled string resonance

For example, the Yamaha Montage uses only samples, so no "transfer of energy" possible to emulate that aspect of a real piano. And of course, DPs and modeled pianos can not currently emulate a real piano resonance, as it's way too complex. The modeled string resonance algorithm would need to use the gazillion of different combinations possible. In other words, a triad played on a real piano does not create the sum of the three resonating piano keys sampled individually. It creates an unique string resonance sound for this particular triad only, which is also different if played with or without damper pedal, etc. Also, depending of how a piano is tuned (partial ratio chosen for octaves, etc.) all these unique string resonance possibilities change. A few software pianos pretend to emulate this, but it's quite weak compared to the real thing.

Here are Bosendorff's C1 samples, the first with no sustain and the second with sustain. The samples were too short to really capture the low level effects of sustain, and you can see very little energy at the fundamental (first partial at about 32Hz) but I think (this may be subjectively biased) I see hints of other notes being present in the spaces between the 2nd and 3rd partials (C2 and G2) in the sustained C1.