I just add this : modify layers volume may affect the realism. If the present curve is the robot’s curve, we probably can fine tune the layers curve, but not much (the curve will lower some samples that were recorded at a louder volume, for example)

(Last edit : it’s hard to say if it’s the robot and/or the instrument and/or a mapping cause(s). If the robot send a perfectly linear strength on each piano key over 128 levels, and if there is no velocity mapping post-production, well, the velocity we get here is just the way the piano react. And we have to adapt our play to it, like a real player has to.)

Gabriel Plalame

@Crystal said:

...(the curve will lower some samples that were recorded at a louder volume, for example)...

 

I don't think so. I think there is a opportunity to change the velocity value of a keystroke. For example: you hit a velocity value of 50 and it is changed in value 30. When you make a tool with such a curve, you don't change the sample volume. 
With this tool you can make a curve, in which e.g. with very high "key hit velocity values" you lower the attenuation in velocity change.

For example:
"key hit velocity" 110 changes in velocity 95,
"key hit velocity" 118 changes in velocity 110,
"key hit velocity" 127 stays 127.
In this way you can use the full dynamic range of the piano. I think there is also a disadvantage in this, that has to do with the resolution in certain area's of the velocity curve (sometimes 2 or 3 "hit key values", e.g. values 40-42, can change in this solution in the same velocity value, e.g. 20). But I don't know if this is very audible.

Hi MMKA,

@MMKA said:

I think there is a opportunity to change the velocity value of a keystroke. For example: you hit a velocity value of 50 and it is changed in value 30. When you make a tool with such a curve, you don't change the sample volume. 
With this tool you can make a curve, in which e.g. with very high "key hit velocity values" you lower the attenuation in velocity change.

For example:
"key hit velocity" 110 changes in velocity 95,
"key hit velocity" 118 changes in velocity 110,
"key hit velocity" 127 stays 127.
In this way you can use the full dynamic range of the piano. I think there is also a disadvantage in this, that has to do with the resolution in certain area's of the velocity curve (sometimes 2 or 3 "hit key values", e.g. values 40-42, can change in this solution in the same velocity value, e.g. 20). But I don't know if this is very audible.

That’s what I call a ‘midi to midi velocity curve’ principle. I use a tool like this between my keyboard and my sequencer, called Velocity Curve Changer : http://www.trombettworks.com/velocity.php

Here, a midi level is translate in another midi level, so then yes, as the example you give, in that way, original samples volumes levels aren't affected.

But, in that case, and particularly if it’s a (wonderful) sample library with 128 samples layers per key, you necessary lose samples (/layers), and velocity sensitivity (when several midi levels are pointed to a single midi level (so then to a single sample file/layer), as you describe too).

That’s why I suggest to build what I would call a ‘midi to layer velocity curve’. But the sample player has to handles a separate volume scale from the ‘zero to 127 regular midi scale’, and it has to be much more precise. Then the midi scale can be convert in another 128 scale/layers with different (and independent) volume values, but inside, let’s say, a 512 volume level scale for example.

In another words, if we want the midi scale being targeted to a curve with 128 layers, the ‘target scale’ has to handles more levels. Then we can built a curve without losing any layers. It stays a 128 to 128 ratio with the ability to fine tune the volume layers curve without any layer loss.

However, in that system, the original volume of samples are affected, and maybe the realism with a piano library.

In one case we lose samples and sensitivity, in another case original samples volumes are affected.

Ah. Hard to consider.

I think the second solution (midi to layers velocity curve) may be acceptable with a reasonable curve.

But it’s not easy to do. You need a sophisticated engine with much more than 128 volume levels for layers.

I don’t know if I make it clear, sorry for my strange language.

(I think the best solution would be to play like Guy Bacos, but unfortunately, that patch doesn’t exist. It’s a scandal, no ?)

Gabriel Plalame

Hi Gabriel,

To begin with: I understand by what you write about it, that writing English is difficult for you. But I think, that you made yourself very clear in what you wrote 😊. I'm not a native English speaker also. And I know the difficulty to write something complex in another language.

So I understand from what you wrote, that in your solution you need much more layers than 128 and then you can make a choice which layers you use. So then there is a lot of work to do for the team of VSL I think.

I for myself I think I would perhaps never use the most loud layers of a Steinway. But I'm very aware that that is a cause of personal idea for the ideal sound. 

I had last week the opportunity to play in a church service on a very beautiful Steinway, a Steinway B (2,11 m), special hired for that occasion (it was with the funeral of my father, who died last week. He was 95 years old, and the piano had always had a very special place in his life. He had a Steinway O 1,80 m and I played the last years often for him. He couldn't play anymore). But on that Steinway B I felt it as a pleasure, that I didn't have to do so much to get a loud tone, almost without heavy arm weight. Of course I used a little, but I got a loud tone very easily. And also playing very soft was easy. The micro dynamics went so natural, in my feeling the piano invited me to play very sensitive and beautiful. For me playing became much easier! I'm sure that that Steinway B could sound much louder than I used, but I didn't need that at all, and I really could play a fortissimo. I think some pianists use a louder fortissimo as I do, but I know also from certain concert pianists who have the same sound ideal as I have. So for me it is not a problem if I miss the loudest tones, because I can express myself very well without them, just as on that real Steinway (in the same week I bought the VSL Steinway and in the sound were really many simularities with the real one. But the keys of the real Steinway played definitely much more easy! I have the MP10 of Kawai). Sorry, a whole story.

But I understand that you need the loudest tones also, and in that case you search for opportunities to use that part of the instrument . I sincerely hope that there will be a solution that works great for you!

How about a real piano?

Sorry, it is just that someone demanding samples beyond 128 velocity levels is just silly.  There used to be ONE.  Then it was 4 and that was impressive.  But now somebody is upset with fewer than 128.  I can't help but find that hilarious.  People today have no concept of just how much they are constantly demanding, and unsatisfied with.

If MIDI is totally inadequate for your awesome musical demands -- (and why do you have those?  Are you the new Rubenstein??  I guess so, you must be powerful and brilliant beyond anything that MIDI can ever accomplish! )  --  then you might want to consider investing in a real piano and play 5000 levels of velocity.  Or whatever it translates into.  Maybe 10,000.  Would that be sufficient for your utterly exquistite level of dynamic control?

In the meantime, doing a recording of MIDI piano, I am still using the Bosendorfer and think it is great.  But of course, I am nowhere near this level of artistry.  

Hi William.

It’s a misunderstanding.

I don’t ask more than 128 layers. (I don’t ask to record anything more anyway.)

The idea is to move layers volumes without losing them.

If the sample library use 128 layers, it needs more than 128 volumes levels to do so, I guess.

It’s hard to explain, (I don’t know how to post a picture on the forum, it’s impossible ?) let’s try an example with a 512 volumes layers scale (not 512 layers !):

Midi volume                Layers numbers           Layers volumes (512 scale in the exemple)

127                            127                              511                    Gap :

126                            126                              503                    8

125                            125                              495                    8

124                            124                              488                    7

123                            123                              481                    7

122                            122                              475                    6

121                            121                              469                    6

etc.

Here you get a velocity curve with 128 Layers.

Regards

Gabriel Plalame

Gabriel,

O.K., sorry I was not clear on what you were saying.  That makes sense.

@karvala said:

it's the timbral mapping that's heavily skewed in favour of fff.

I would like to 2nd this observation... and think to a small extent, the CFX could use a tweak fixing the bias towards the fortissimo and fortississimo as well.  It is a timbral issue, agreed.

For the harshest of tone, only the very top velocities should be reserved for it.  (Sort of like "harsh" shorts on the Synchron strings)  For the almost-top velocities, I feel there should be a gradual timbral curve.  But as of now, if I'm in the upper 1/3, it's all pretty "loud" timbrally.

@stephen limbaugh said:

I would like to 2nd this observation... and think to a small extent, the CFX could use a tweak fixing the bias towards the fortissimo and fortississimo as well.  It is a timbral issue, agreed.

For the harshest of tone, only the very top velocities should be reserved for it.  (Sort of like "harsh" shorts on the Synchron strings)  For the almost-top velocities, I feel there should be a gradual timbral curve.  But as of now, if I'm in the upper 1/3, it's all pretty "loud" timbrally.

One other thing.... I sent a file over to VSL about the keynoise on the Steinway, which to my ears is really loud, causing a whole lot of noise/mud in the 75-184hz range.  If this is adjustable in the sample, then perhaps that would alleviate some of the preceived timbral "bias" towards FFF.  Maybe? 

@stephen limbaugh said:

I would like to 2nd this observation... and think to a small extent, the CFX could use a tweak fixing the bias towards the fortissimo and fortississimo as well.  It is a timbral issue, agreed.

For the harshest of tone, only the very top velocities should be reserved for it.  (Sort of like "harsh" shorts on the Synchron strings)  For the almost-top velocities, I feel there should be a gradual timbral curve.  But as of now, if I'm in the upper 1/3, it's all pretty "loud" timbrally.

I sent a file over to VSL about the keynoise on the Steinway, which to my ears is really loud, causing a whole lot of noise/mud in the 75-184hz range.  

I have the same issue (annoying key noise on some release notes).The noise is more audible when playing softly (final chord of a soft piece, for example). It reminds me the noise of a silent piano.

The Steinway is beautiful but I agree that hammers are too hard (at least, compared to then my own 205 cm grand piano). It is not easy to produce a «muffled» sound with that software intrument.

Best,

C.

Given complaints about issues similar to those of the Steinway going back to the CFX release, it makes me wonder how likely it is that VSL will correct these issues in a reasonable time. Can anyone offer some optimism? Have there been any previous updates from VSL for any of their pianos?

@C.B. said:

I have the same issue (annoying key noise on some release notes).The noise is more audible when playing softly (final chord of a soft piece, for example). It reminds me the noise of a silent piano.

The Steinway is beautiful but I agree that hammers are too hard (at least, compared to then my own 205 cm grand piano). It is not easy to produce a «muffled» sound with that software intrument.

Best,

C.

Here is my solution thus far that may be acceptable in crafting a better sound...

Turn up Dynamic to around 120 (maybe more?) and turn down MIDI Sensitivity to around -20.

Also, go to the edit page and take the upper half of the keyboard and put an EQ shelf below 200hz... (or get the bell curves so you can keep a little bit of the piano "body" below 70hz) that helps reduce all that noise, yet on the lower half of the keyboad you can preserve the low end.

Hi all,

Thanks for the replies.  In the past few days I did a direct comparison with the VSL Vinenna Imperial and it does not have the same problem.  The Imperial runs fine with a linear curve off the Kawai NV10, much like Garritan and my other VSTs.  There is something definitely different with the VSL Steinway D.

I do not beleive there is anything wrong with the sample set, the problem would be remedied by adding a 1-D velocity curve editor.  The only case in which a 1-D velocity curve editor could not address this is if there was (a) a sparse sample set and (b) strong non-linear characteristic to how the samples were originally captured.  For example if the sample set had 8 layers and most layers were recorded near ff-sfz..not much can be done there.

However with so many sample points on the velocity axis, just fixing the curve would fine even if the 'robot' velocity curve for layers was non-linear.

Someone commented on the VSL video and how the pianist did not notice this, this is a good observation. I would note that that this was a specific weighted MIDI controller which has some implemented velocity mapping curve.  It could be that the VST was designed to work very well with that in-house controller, but that doesn't mean it would work with others, which is why having a 1-D velocity curve editor is standard on all other VSTs.

I agree that if the velocty curve issue is fixed as far as I can tell this could be the definitive sampled Steinway D.  What great sound, and ambience, absolutely mesmerizing.

I can't help wondering what is done with all this perfection of velocity levels etc. by the people demanding it.  

There are lots of people on this Forum who write all this complicated stuff about how this or that parameter is just not right and not acceptable, etc. etc. - and yet what do they do with all this technique?  

Not trying to be obnoxious!  Maybe all the complaints are correct, sure.  I don't know.  But I wonder what is the actual musical use of all of this?  If the instrument is just perfect ----  do the people who are complaining then proceed to create masterpiece after masterpiece?    Somehow I never hear any music by them.

@William said:

Somehow I never hear any music by them.

Allow me to flex. 😛

If the suggestions in this were implemented, I would say they would have the following direct musical benefits:

• Better "out of the box" sound requiring less EQ to fix the noise.
• Better playability requiring fewer adjustments to the MIDI after a performance.
• For film scoring, piano parts tend to air on the soft side (the opposite of the current curve)... could be more inspirational for composers!

Granted, this is what *I* would do.  If none of these are implemented, then I certainly have workarounds and handy "save presets" options once I get it dialed in how I like it. 😃  I have a feeling though VSL will introduce a curve... they're really good about listening to their customers when they have reasonable requests.

@opus64 said:

Hi all,

Thanks for the replies.  In the past few days I did a direct comparison with the VSL Vinenna Imperial and it does not have the same problem.  The Imperial runs fine with a linear curve off the Kawai NV10, much like Garritan and my other VSTs.  There is something definitely different with the VSL Steinway D.

I do not beleive there is anything wrong with the sample set, the problem would be remedied by adding a 1-D velocity curve editor.  The only case in which a 1-D velocity curve editor could not address this is if there was (a) a sparse sample set and (b) strong non-linear characteristic to how the samples were originally captured.  For example if the sample set had 8 layers and most layers were recorded near ff-sfz..not much can be done there.

However with so many sample points on the velocity axis, just fixing the curve would fine even if the 'robot' velocity curve for layers was non-linear.

Someone commented on the VSL video and how the pianist did not notice this, this is a good observation. I would note that that this was a specific weighted MIDI controller which has some implemented velocity mapping curve.  It could be that the VST was designed to work very well with that in-house controller, but that doesn't mean it would work with others, which is why having a 1-D velocity curve editor is standard on all other VSTs.

I agree that if the velocty curve issue is fixed as far as I can tell this could be the definitive sampled Steinway D.  What great sound, and ambience, absolutely mesmerizing.

Unfortunately it's not as simple as that; if it were there would be no problem because there are countless free velocity curve editors already out there which can easily be used with Synchron Pianos already.

The problem is that the perceived loudness of the piano comes from two things: objective amplitude (sound pressure level) and timbre.  If I record a piano really loud, and then greatly reduce the volume on playback, it doesn't sound like a soft piano, it sounds like a loud piano being played back quietly.

A velocity curve editor enables the user to change the mapping between the input and the amplitude-timbre combination (because the two are fixed together in the sample).  However, if the amplitude-timbre mapping is wrong, which in this it clearly is, that cannot be fixed by changing the velocity curve.  I could, for example, edit my velocity curve to require me to play much harder in order to achieve higher velocity layers, which would help to fix the input-timbre mapping.  However, that would have the effect of making it much harder to achieve higher amplitudes, i.e. it would distort the input-velocity mapping.  Or instead, I could edit the velocity curve to ensure that the input-velocity mapping is okay (which it more or less already is in this case), but that would then create (or leave/not fix) the problem with the input-timbre mapping.

In other words, you cannot, by adjusting the mapping of the input to different samples, fix a problem which is caused by internal inconsistency in the samples in terms of the amplitude and the timbre.  The amplitude-timbre mapping of the samples needs to be adjusted and that can only be done by VSL.

Steven, yes actually I realized I had heard something of yours after writing that. Excellent work though I'm sure you don't care what I think.  

Opus64 thanks for that - very interesting and I'm certain the feedback is valuable to VSL.

Hi Karvala,

Interesting point, but i'm not sure I understand something if you don't mind clarifying:

My understanding of how the piano was sampled is that multiple mics are placed in the room with fixed settings(preamp gain, etc).

Considering only a single pedal/etc condition layer, the robot plays a key at multiple velocities and the microphones capture the resulting sound.

Unless VSL is changing the mic settings for velocity(perhaps to accomodate dynamic range?) or post-adjusting the amplitude of the samples during playback, it seems to me the amplitude and timbre are linked at the recording point and if the microphones are linear, should represent what the instrument sounds like.

If these two are truly linked at sampling time, then it seems the midi->velocity map should allow to adjust for the problem, which in my view is that for a relatively soft keypress velocity the resulting sound is more ff or even sfz.

I'm probably missing something....

Thanks.