How to heal Achilles' heel of Sampled Pianos

I think I could identify the very weak point of Sampled Pianos.

!Accoustic rules! by a large margin for me too, to be sure - but within DP experience I am a late convert to Pianoteq (from some very good sampled instruments like Galaxy Vintage D and Vienna Grand/Boesendorfer, Sampletekk and PMI instruments.) I was very intrigued by the contradicting experiences with these two types of instruments: Sampled pianos are much more appealing sound wise, it is more pleasing to make the sounds with them initially, but while playing them they are becoming more flat, narrow and lacking of something.

Finally I gave Pianoteq another chance, and this was the opposite experience: starting with an initially seemingly mediocre sound with some plasticly overtones in the middle, at best uninteresting sound-wise, but then as You begin playing it after a few minutes You will get more and more involved - like with an accoustic instrument.

I was really very intrugued indeed, and a little confused, seemed a challange to find an explaination. Now I think to have identified the problem.

There were many discussions with comparison of these technologies, e.g. this lengthy one raised about
Much discussed possible weak points with Sampled instruments so far:

1. Limited Number of layered sounds. (Limitation by one order of magnitude more sharp then by velocity layer! - but
I don't think it is very limiting either - think of the minor differencies between SW instruments and built in DP sound, 13-16 vs. 3-5 velocity Layers). Nevertheless, this is much more probably a real issue than

2. MIDI velocity levels of 127 limiting expressive playing of dynamic nuances (I don't regard it as serious problem because expressivity is even possible with harpsichords - with 1 velocity layer). By the way, this would be a limitation for modelled instruments equally as well.

3. Sound amplifying issues, to get the sound from very specific, narrowed direction of boxes/monitors.
Not very convincing argument, headphones should be void of such limitation.

4. I could read and partially reproduce some [b]regulation issues[/b] with the original sampled instruments: inconsistent mapping of velocity parameter, changing pitch due to changing environmental parameters as room temperature, humidity, barometric pressure and simply the longer time intervals since last instrument regulation or the need of re-tuning/re-regulating in between could lead to inconsitencies. I was able to detect some of such inconsistencies by pitch measurements in more than one sampled instrument, measured a few cents pitch differences with single notes in different sample compartments. (Una Corda, Pedal Up/Down, different velocity Layers.) I felt definitely the need of re-tuning the instruments to overcome some pitch inconsistencies.

And finally my (at least for me) new explanation of the main limitation intrinsic to the sampling process:

5. Imprinted sound space perspective and invariably hardcoded recording parameters within samples . Sampling is done by sound recording through microphones. The samples are reflecting all of the recording parameters, as microphone position, microphone characteristic, even microphone types, along with other recording parameters. All of these parameters can be constant or some of them could be changed during the sampling process. But after recording there comes with the sampled instrument a fixed set of parameters, which after some time you get somehow aware of. These are characteristic for the sampling process itself, not for the sampled instrument, and you will get sooner or later inevitably feel their presence.

One good example of this is the preferred close microphone perspective, having a near-string position. Sound source is the whole excited string, but from the more distant part of the string the sound comes with a slightly different phase, so the microphone position relative to the string will be captured percievably by the microphone. With a 0.4-3m long strings and in between 55-4000 Hz there will be captured very significant phase differences with the sample sound image.

I think our unconscious mind could become more and more uncomfortable and disoriented by such hidden effects of the recording process.

Microphones are the very narrow pipeline opening all sampled must sounds flow through - this could be the single major bottleneck I think.

How to overcome it? One simple arrangement to eliminate or reduce this limitation could be recording the samples faithfully from the strict perspective of the player.
You have to choose a specific ambience (room, music saloon, concert hall) and place the acoustic instrument to be sampled in a specific postion. Chose an opening for the lid and let it unchanged for the whole process.

Make a sphere with 2 mics on the left and right side of it modelling the head and ears of the performer, sit this capturing head unit on the same position as the player in front of the instrument and capture the single notes in the usual way.

Finally you have to apply binaural/stereo filtering to optimise output through headphones/boxes. Your dedicated instrument is now ready optimised for playability. And You have then make a

What You get is a dedicated player-perspective instrument with perhaps fewer flexibility towards free configurable perspectives, and ambient reverb.

But what caveats do we trade in?

Post processing could be limited. Close miced sampling is preferred for a better raw amterial for mixing, convolution reverb, such effects as sympathetic resonance, etc. I don't know how smoothly the more wet samples behave when chords are played, how natural additively activated ambient components would be.
(There comes old SampleTekk Piano Black Grand Medium Ambience in mind. However, it is not recorded from the player's perspective and there are practically no modern programmed features with it.)

If the concept could work, sampled instrument makers could make a prolonged living from such an approach: they could probably bring out different dedicated Performer/Editions with specific ambient recording.

Your thoughts? Attila

I basically agree with many of the things you have said about the limitations and strengths of sampled pianos. I'm not actually sure I agree that some people's perception of flatness of a sampled piano is due to fixed recording position and settings, though. The ability to tweak the settings in pianoteq and change things like the mic position doesn't mean that playing in the same position all the time doesn't lead to the same type of fatigue you get with a sampled piano. I kind of think it doesn't get noticed because PianoTeq's other warts prevent you from getting to that level. To each his own, though.

In fact I don't personally find what you have described about sampled pianos sounding flat or missing something. Comparing sampled pianos to PianoTeq, I personally find that high quality sampled pianos dominate or are equal in each respect (aside from the interface...PianoTeq is near perfectly designed for usability).

A number of VST's have taken a similar approach to what you describe: sampling from various positions and having different versions instead of using a single position and then post-processing to get whatever perturbations you want. However, I should point out that the pianos that did that also are not very popular on this forum, presumably because they are not as good as the ones sampled from a single perspective. I'm not saying it isn't a better approach, but it doesn't seem to have been as successful in the past.

Your binaural idea seems interesting, though it is not altogether clear to me that player perspective is really that high a priority. Most sampled pianos are miced basically where you would place the mics if you were making a CD of solo piano music in a studio. As such it sounds sort of perfect. Is the sound a pianist at the keyboard really better than what mics on the side of the piano pick up? I don't think so. I would be happy if, sitting at a grand piano, I could hear things as they come out the side, rather than as they hit me where I sit. Again, cool idea and maybe worth pursuing, but in my opinion, not the crux of the problem.

Unfortunately, I'm not sure what the crux of the problem really is (if we accept that there is one), so I can't get too critical.

I strongly agree here. It doesn't matter how dynamic or expressive PianoTeq is because it doesn't SOUND good enough. So the fatigue will be there, perhaps even more so because there's nothing you can do to make it sound real. I imagine all the tweaking becomes tiring and frustrating for that reason.

But sampled pianos do cause a kind of fatigue in some of us. I include myself and in my opinion it explains why I can't stay happy for longer than a few days (weeks at most). I think it is the never-changing character of sampled pianos.

Let me put it this way, if PianoTeq sounded right and offered a genuinely interesting palette of *real* sounding pianos, we'd all be users. The same goes for the V-Piano - perhaps even more so because it would be a one-box solution with great key action and no faffing about with wires and computers etc.

The day is getting nearer, I have no doubt. Modelling has got to be the future.

Perfect, or aggressive? Perhaps just to listen to it could be satisfactory because you get more detail and nuances with the music otherwise not reaching you as a listener, as a player however You have other intentions: to control music melodically,harmonically polyphony and that can be just this another approach be ruined by aggressively present sound engineering artifacts and characteritics percieved as "Raspy metallic sounds" complaints or alike reported here on the forum previosly.

Just an interesting example: I own a very good Beurmann Studio harsichord (the "Dutch"). They introduced Velocity Layers to the harpsichord (what a nice perversion!). But the intention was not to give an improved harpsichord with a piano-like dynamics, but to simulate the fine variability of the timbre. It is never possible to have the same sound out of a theoretically velocityless harsichord keystroke, but there are coming out slicghtly different wave patterns. They simulate this by giving 5 samples with slightly different tensions plucked. I than gave a MIDI velocity randomness in Kontakt to gain the intended "humanising" effect. (The "human being" being the Harpsi itself in this case not the player.)
Ubiquitos uniformity can be percieved as agressive exerting a fatigueing effect.

Some of the issues you cite with sampled pianos also apply to Pianoteq. i.e. Pianoteq is still limited to 127 MIDI levels from the keyboard and your "sound amplifying issues". As far as "the limited number of layered sounds", I don't think sampling at 13-20 layers is much of a timbre limitation when combined with 127 amplitude levels, but I won't argue the point if you think Pianoteq or other modeled pianos produce more useful (the key word) timbre variations.

There certainly are "regulation" issues with sampled pianos. But it can be argued that if those "regulation issues" are resident in the original acoustic piano, and that acoustic piano is considered "fit" or "concert ready" those minor regulation inconsistency issues may be increasing the realism of the sampled piano, and thus desirable by providing the actual piano's character as opposed to a sterile modeled approach. I'd say the Vintage D is a good example of a sampled piano with such natural and desirable character. On the other hand, if the original piano is not well regulated, or has other noticeable problems, then those problems are not desirable and can ruin an otherwise good sampled piano (I think I know of a perfect example of that). Finally, the process of recording samples of an acoustic piano is an imperfect process and samples sets are full of inconsistency issues for technical and human error reasons. (I've done a lot of technical analysis of sample sets using tools I have created). Sometimes those issues are minor and may be unnoticeable in actual use, may add a little random character that is actually desirable (vs more sterile modeling), or may be highly objectionable. (I know examples of the latter).

Your last point (sound space recording) is very interesting. Binaural recording of samples using a dummy head is certainly an optimum method (with headphones designed for binaural playback) of reproducing the acoustic piano samples from the player's perspective, and has long been known as a technique for hi-fi reproduction. But those samples are not desirable for reproduction using speakers in a room environment. The closely mic'd sampling techniques are much better for that purpose, where additional room modeling using convolution techniques, and other acoustic piano modeling processes can be applied to the samples. Other mic perspectives are present in some sample sets but I find them far less realistic when reproduced using speakers, which essentially creates a convolution of the reproducing room response with the recorded characteristic of the original room and mic placements.

Finally, I think one of your points was "after recording there comes with the sampled instrument a fixed set of parameters, which after some time you get somehow aware of. " I would say that is equally true of having an acoustic piano in your room. You aren't likely to be moving the piano around in your room from day to day to vary the acoustical interaction with the room. The voicing of the acoustic piano will certainly shift over a length of time, but it is easy to vary the voicing of the piano samples if you find some variation over time is desirable.

+100

Greg.

Macy, My point was, that the possible variations are might leave and thereby even possibly stress out some core characteristics of the sampling process (instead of the original) reveling thereby their intrusive nature.

Otherwise we have more consens than dissens - as I wrote, all of the issues listed 1-4 I didn't regard myself as crucial and not even as characteristic for sampled instruments (vs. modelled).

gvfarn, EssBrace:

I dont't know modelled pianos internal merits enough yet, therefore I cannot be an advocate for Pianoteq (not yet). On the contrary, I detested Pianoteq on just having some listening probes a'la PurgatoryCreek and it seemed so trivally inferior that I neglected it all the time (along with VPiano.) And even after the long awaited PT4 I felt promptly the same disappointment. And yes, now it seems I was wrong, my expectations were based on some elusive first impressions. It was a very interesting discovery - admittedly a very subjective one, but I share it now with all of you because it was a very enforcing paradigma change. It was not a one-day whimsycal change of mood either, some self-deception, which I often had with music in the past. I can tell it is a very puzzling and very real experience of the last some 3-4 weeks which calls for explanation.)

Would be interesting to know in detail what the internal limitations of Pianoteq or VPianos are, or Physis (and SN and GEM with theit partial modelling) what details is their models capable to cope with, where are the approximations just too rudimentary, what they can achieve. (I found Dewster's enthusiasm about PT a bit suspect, how could it perform so well with the DPBSD test at all.)

I just want to exhibit my paradoxical experiences and thoughts to all of you open-minded music experts/enthusiasts.

Greg, was my point too, I just listed the purported limitations.

Ok. Note that EWQLP has more than 13 layers, but it is very lumpy. So, the 13+ layers have to be done well. ;^)

Greg.

Yeah, and Dewster over in the DPBSD project found a police siren and an air conditioner in some of their samples, lol.

Yeah, PianoTeq 4 (just about) fixed the two biggest problems it had: an unnatural metallic ringing and heinous reverb. I still find it too artificial sounding and somehow the reverb isn't perfect, but the two big improvements take it to the place where I find it on par with my (admittedly old and not very desirable) onboard samples. I have actually been playing with it again lately, but I can't be satisfied. Basically it's in the uncanny valley of piano sounds at the moment.

as is a nice birds' twittering with PMI Emperor(a Boesendorfer) in a quite few samples - presumably from outside of the ambient room and not nested in the piano body. By the way a good example of a Piano with wet ambient sampling - but an older and not very consistent one.

The uncanny valley of piano sounds is my living room where my vintage acoustic is waiting for me to get be played on...