 |
Welcome to the Piano World Piano Forums Over 3 million posts about pianos, digital pianos, and all types of keyboard instruments. Over 100,000 members from around the world.
Join the World's Largest Community of Piano Lovers
(it's free)
It's Fun to Play the Piano ... Please Pass It On!
|
|
51 members (brennbaer, Damien PG, CianistAndPomposer, Beowulf, A. Lucato, clothearednincompo, Boboulus, 12 invisible),
427
guests, and
416
robots. |
Key:
Admin,
Global Mod,
Mod
|
|
|
Joined: Nov 2016
Posts: 266
Full Member
|
Full Member
Joined: Nov 2016
Posts: 266 |
8 bits is not good enough for classical piano in terms of either level or timing. I believe Yamaha use a much higher resolution for their ePiano Disklavier competitions for good reason. You can prove it yourself. Play a piece with plenty of rubato and record the audio out while recording midi data into a sequencer. Now run the midi data back to the piano and record the output. Compare the audio outputs they should be the same? They won't be. And one will sound weird and the other won't, just like every classical midi file.
This is aside from the discussion on dynamic levels, but the timing thing is not true at all - MIDI files have more than enough timing resolution to capture a natural performance and have done since the 80s. Its resolution will vary depending on settings, but it will be down to the milliseconds level, more granular than any human would be able to discern anyway. Rubato is irrelevant, it will just be capturing the exact time you played whatever notes you played - I've recorded a DP as MP3, and simultaneously as a midi file so I can layer other sounds beneath it, and if I play the midi file back through the digital piano it sounds exactly the same.
|
|
|
|
Joined: Sep 2009
Posts: 13,370
Yikes! 10000 Post Club Member
|
Yikes! 10000 Post Club Member
Joined: Sep 2009
Posts: 13,370 |
Two things: 1. Toddy: You are correct. Every MIDI velocity level can produce a different loudness from the piano, and each of the MIDI values is assigned to one of the available sampling levels. There are up 127 of the former, and as many as the manufacturer sees fit for the latter ... with 127 being the upper limit, but it's much smaller in practice.
The number of velocity layers for samplers ranges from 1 for the el-cheapo pianos to about 20 for the better VST libraries (though I may be out of date with that figure because I don't own any of the newest mega-libraries).
2. How is any of this even relevant? Your ears cannot hear the difference once the number of sampling layers reaches some number. I doubt you'd ever benefit from having more than ten, especially since the transition between on layer and the next is blended in software.
If you have ten competently recorded sampling layers that number ceases to be a relevant deficiency in the piano. There are other deficiencies to be overcome. Ones that are much more important ... such as: - Proper pedaling (which seems to always be a deficiency despite the glowing claims made). - Proper resonances (likewise deficient). And those are just the audio aspects. Next comes the shortcomings of the action ....
Short story: I'm not concerned about the number of sampling layers anymore.
|
|
|
|
Joined: Jun 2014
Posts: 1,445
1000 Post Club Member
|
1000 Post Club Member
Joined: Jun 2014
Posts: 1,445 |
T 2. How is any of this even relevant? Your ears cannot hear the difference once the number of sampling layers reaches some number. I doubt you'd ever benefit from having more than ten, especially since the transition between on layer and the next is blended in software.
Short story: I'm not concerned about the number of sampling layers anymore. Ivory uses interpolation between layers, but also has adjustable velocity layers anything from 4 to 20 on Ivory ACD. The differences can be noted. It may not be obvious immediately in a blind test, but when you play for a while you will notice every note at some velocity can have characteristics unique to that piano adding character to the sound. If you really go down in velocity layers that goes away and the sound becomes more homogeneous with less character IMO. I like to think of it as a pint of fine organic milk with all the froth, that is lets say 20 or more layer. 4 - 8 layers is more like homogenised sterilised milk, like my Ca78 for example, which I would guess has absolutely no more than 8 layers, so that character does go up with layers, you get more nuances and details that would be missed out on with less layers. Where it stops to the ear I don't know, I never tested. To my ears, suffice it to say I always found the Vienna imperial by VSL one of the most convincing sounding timbres, it indeed has a 100 layers, and interpolation will likely interfere with the natural sound it was recorded from at layer boundaries that are audible if you compare it to the original sound it came from ( I guess), VSL Imperial is not interpolated btw at all AFAIK
Last edited by Alexander Borro; 10/19/18 04:06 PM.
Selftaught since June 2014. Books: Barratt classic piano course bk 1,2,3. Humphries Piano handbook, various... Kawai CA78, Casio AP450 & software pianos. ![[Linked Image]](http://www.pianoworld.com/ABF_Medals/12xmedals.jpg) 12x ABF recitals. My struggles: https://soundcloud.com/alexander-borro
|
|
|
|
Joined: Oct 2010
Posts: 368
Full Member
|
Full Member
Joined: Oct 2010
Posts: 368 |
8 bits is not good enough for classical piano in terms of either level or timing. I believe Yamaha use a much higher resolution for their ePiano Disklavier competitions for good reason. You can prove it yourself. Play a piece with plenty of rubato and record the audio out while recording midi data into a sequencer. Now run the midi data back to the piano and record the output. Compare the audio outputs they should be the same? They won't be. And one will sound weird and the other won't, just like every classical midi file.
This is aside from the discussion on dynamic levels, but the timing thing is not true at all - MIDI files have more than enough timing resolution to capture a natural performance and have done since the 80s. Its resolution will vary depending on settings, but it will be down to the milliseconds level, more granular than any human would be able to discern anyway. Rubato is irrelevant, it will just be capturing the exact time you played whatever notes you played - I've recorded a DP as MP3, and simultaneously as a midi file so I can layer other sounds beneath it, and if I play the midi file back through the digital piano it sounds exactly the same. Depends what you play. For advanced classical pieces standard MIDI doesn't work. I can tell you that if this standard MIDI file is representative of the gentleman's performance, he couldn't win a toffee. http://www.piano-e-competition.com/MIDIFiles/2017/BuiJL08.MIDIt sounds quantised, granular, mechanical and dreadful like a robot at the keyboard. The contrast between that and any audio version of L'isle Joyeuse that you can find on Youtube is so blindingly obvious to anyone with an ounce of sensitivity to rhythm.
|
|
|
|
Joined: Nov 2016
Posts: 266
Full Member
|
Full Member
Joined: Nov 2016
Posts: 266 |
Depends what you play. For advanced classical pieces standard MIDI doesn't work.I can tell you that if this standard MIDI file is representative of the gentleman's performance, he couldn't win a toffee. http://www.piano-e-competition.com/MIDIFiles/2017/BuiJL08.MIDIt sounds quantised, granular, mechanical and dreadful like a robot at the keyboard. The contrast between that and any audio version of L'isle Joyeuse that you can find on Youtube is so blindingly obvious to anyone with an ounce of sensitivity to rhythm. BiB: nope, you're wrong. I have no idea how that file was generated so can't comment on it, and any midi file sounds awful on my computer's soundcard, but a midi file is plenty good enough to reproduce a digital piano performance of any piece played by any pianist - whatever you would record via audio during the live performance, that is what you would get if you replayed the midi. That is what midi was designed to do, and it's been pretty much unchanged since the 1980s.
|
|
|
|
Joined: Oct 2013
Posts: 3,169
3000 Post Club Member
|
3000 Post Club Member
Joined: Oct 2013
Posts: 3,169 |
@anotherscott: DPBSD is about digital (hardware) piano. I was thinking of virtual pianos. Perhaps blending is less useful with near 20 layers or more.
Last edited by Frédéric L; 10/19/18 06:02 PM.
Yamaha CLP150, Bechstein Digital Grand, Garritan CFX, Ivory II pianos, Galaxy pianos, EWQL Pianos, Native-Instrument The Definitive Piano Collection, Soniccouture Hammersmith, Truekeys, Pianoteq
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
@anotherscott: DPBSD is about digital (hardware) piano. I was thinking of virtual pianos. He tested both, actually. But as you probably know, there have been no updates in a while, so the VSTs evaluated there are not the most current.
|
|
|
|
Joined: Oct 2013
Posts: 3,169
3000 Post Club Member
|
3000 Post Club Member
Joined: Oct 2013
Posts: 3,169 |
Ok, then I read about the Galaxy Vienna :
Velocity switch @ vel=45, 52, 70, 80, 90, 102, 112.
About Galaxy Vintage D : some velocity switching is visible particularly when zoom up and midrange is nicely blended (perhaps with scripting since I haven’t seen Kontakt builtin blending)
Garritan Steinway : OK velocity switching (I don’t know really what it means)
Not all VST are analysed : sometimes, there is only the mp3 file.
Last edited by Frédéric L; 10/20/18 04:36 AM.
Yamaha CLP150, Bechstein Digital Grand, Garritan CFX, Ivory II pianos, Galaxy pianos, EWQL Pianos, Native-Instrument The Definitive Piano Collection, Soniccouture Hammersmith, Truekeys, Pianoteq
|
|
|
|
Joined: Aug 2016
Posts: 397
Full Member
|
Full Member
Joined: Aug 2016
Posts: 397 |
I don’t think blending samples from two layers is common. I think it is the exception rather than rule, But that's one of the things the DPBSD evals tried to suss out. Eh? I'd say it very much is the rule for hardware dps at least. I certainly cannot think of any recent model from the Yamaha, kawai, Roland and probably even Casio (if you exclude the toys) that shows sharp changes in tone. Yamaha does, rather shockingly, still sample stretch on eg the p115 but the velocity tone shift is smooth (though I'm sure I've seen some speculation that it's a single layer with the tone change entirely done by filters etc)
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
I don’t think blending samples from two layers is common. I think it is the exception rather than rule, But that's one of the things the DPBSD evals tried to suss out. Eh? I'd say it very much is the rule for hardware dps at least. I certainly cannot think of any recent model from the Yamaha, kawai, Roland and probably even Casio (if you exclude the toys) that shows sharp changes in tone. Roland SuperNatural (and V) use modeling, so they come at it from another angle. But even when there are multiple layers with no blending, it is not so easy to discern the presence or absence of blending in normal playing (though you can tell as the DPBSD tests do, by sending MIDI to all 127 values and putting the results on a scope). In normal playing, I think it's pretty hard to tell if there are sudden shifts between layers, outside of same-note crescendos. Especially as the number of layers increase (which means a lesser degree of difference between adjacent samples). Yamaha does, rather shockingly, still sample stretch on eg the p115 but the velocity tone shift is smooth (though I'm sure I've seen some speculation that it's a single layer with the tone change entirely done by filters etc) Not shocking, plenty of DPs don't sample all 88 keys, especially at lower price points. In fact, I'd say it's more common than not. But yes, to the other point, filters can also be used to give the illusion of layer blending (i.e. a softer strike can play the same sample as a harder strike but with less brightness by using a filter).
Last edited by anotherscott; 10/21/18 11:28 AM.
|
|
|
|
Joined: Aug 2016
Posts: 397
Full Member
|
Full Member
Joined: Aug 2016
Posts: 397 |
Well I was thinking of the DPBSD thread when I said most do not show sharp changes in tone. They are not necessarily blending the samples but they are applying a timbre change across each layer so that there are no sharp changes in tone like there were with older DPs (or all the really low quality extra instrument samples they throw in), this is shown fairly well in the DPBSD thread were even if layers show up in the phase display, they do not show up in the spectral display.
Similarly for 88 key sampling. As far as I'm aware, of the 'big 3', yahama is the only one currently selling DPs with stretched samples and this has been the case for quite some time.
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
As far as I'm aware, of the 'big 3', yahama is the only one currently selling DPs with stretched samples and this has been the case for quite some time. I don't know model by model, but up to $600, Yamaha's competitors are Casio and Korg, and I think those competitive models stretch as well.
|
|
|
|
Joined: Oct 2017
Posts: 1,181
1000 Post Club Member
|
1000 Post Club Member
Joined: Oct 2017
Posts: 1,181 |
As far as I'm aware, of the 'big 3', yahama is the only one currently selling DPs with stretched samples and this has been the case for quite some time. I don't know model by model, but up to $600, Yamaha's competitors are Casio and Korg, and I think those competitive models stretch as well. Unless the "big" manufacturers started changing their technical designs lately, stretched and looped samples are to be found on most DP models except for some top-tier models...
|
|
|
|
Joined: Oct 2015
Posts: 2,544
2000 Post Club Member
|
2000 Post Club Member
Joined: Oct 2015
Posts: 2,544 |
Unless the "big" manufacturers started changing their technical designs lately, stretched and looped samples are to be found on most DP models except for some top-tier models...
Kawai has 88 key sampling (=not stretched) already on its entry level model ES110. The same for the smaller cabinet models. Of course also for the higher tier models.
|
|
|
|
Joined: Oct 2017
Posts: 1,181
1000 Post Club Member
|
1000 Post Club Member
Joined: Oct 2017
Posts: 1,181 |
Unless the "big" manufacturers started changing their technical designs lately, stretched and looped samples are to be found on most DP models except for some top-tier models...
Kawai has 88 key sampling (=not stretched) already on its entry level model ES110. The same for the smaller cabinet models. Of course also for the higher tier models. Yep! Kawai is one of the few manufacturers that states that feature explicitly. The ES110 has 32MB worth of samples vs. 512 MB on the CS67/97 vs. 1GB on the CS68/98/NV10 . So, the lower-end models probably have 88 key sampling with a single velocity layer and then the number of layers gradually goes up in the higher-tier models. (cf. http://forum.pianoworld.com/ubbthreads.php/topics/2746707/how-much-sample-memory-is-in-there.html) The sample libraries used by DPs are really small when compared to average VST libraries. But the results that a DP can produce are quite impressive if we just consider the absolute numbers. I wonder what would be degree of improvement if a DP manufacturer supported a, say 100 GB, high-quality sample library combined with its know-how on sound processing...
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
The ES110 has 32MB worth of samples vs. 512 MB on the CS67/97 vs. 1GB on the CS68/98/NV10 . So, the lower-end models probably have 88 key sampling with a single velocity layer and then the number of layers gradually goes up in the higher-tier models.
The difference does not have to be velocity layers (or only that)... other things that affect size are how much of a sample there is before it is looped (or whether it is looped at all), and samples for additional instrument sounds (including possibly additional pianos).
|
|
|
|
Joined: Aug 2016
Posts: 397
Full Member
|
Full Member
Joined: Aug 2016
Posts: 397 |
And we know from the DPBSD thread that roland uses 88key sampling (when it does actually use samples) for, eg, the FP30, which leaves just yamaha, as I said 
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
I wasn't sure which you were considering the big 3!
But again, at up to $600 (for 88-key weighted action models), where Casio, Yamaha, and Korg have multiple models (and Roland and Kawai have none), I think all use stretching.
The FP30 uses their Supernatural sampling/modeling hybrid. It may well use 88 key sampling, though I haven't been able to determine that for a fact.
|
|
|
|
Joined: Oct 2017
Posts: 1,181
1000 Post Club Member
|
1000 Post Club Member
Joined: Oct 2017
Posts: 1,181 |
The ES110 has 32MB worth of samples vs. 512 MB on the CS67/97 vs. 1GB on the CS68/98/NV10 . So, the lower-end models probably have 88 key sampling with a single velocity layer and then the number of layers gradually goes up in the higher-tier models.
The difference does not have to be velocity layers (or only that)... other things that affect size are how much of a sample there is before it is looped (or whether it is looped at all), and samples for additional instrument sounds (including possibly additional pianos). You are right. The length and resolution of the samples play a significant role as well. But the large jumps in library size are often the result of a new "layer" (velocity, mic position, etc.). The big jump from the ES110 to the CA67/97 should be the combination of longer samples, higher resolution and more layers. The other instruments should not take much space since one can get at the same quality with small sound fonts. On the other hand the CA78/98/NV10 has twice the memory of the CA67/97. Half of it (512MB) should be the to store all of the "sound mode" instruments that came from the previous generation. The other 512MB (or part of it at least) should be the "pianist mode" SK-EX. I would say its 10 rendering characters are a result of processing the same set of samples with different modelling parameters and effects. But we would need to run these DPs through the DPBSD project to confirm this :-)
|
|
|
|
Joined: Feb 2010
Posts: 5,875
5000 Post Club Member
|
5000 Post Club Member
Joined: Feb 2010
Posts: 5,875 |
The other instruments should not take much space since one can get at the same quality with small sound fonts. Depends how fussy you are. Quality Rhodes, Wurli, Clav, can easily run into 100s of megabytes each.
|
|
|
Forums42
Topics206,370
Posts3,083,696
Members101,224
|
Most Online15,252 Mar 21st, 2010
|
|
|
|
|
|