Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Pianoteq 6!

Are you claiming that on your CFX you have 127 levels of amplitude?

I can't say with authority that Garritan have provided 127 levels, only that the software could easily cope with that. More to the point, the test I proposed for you would challenge you no less than a library that did effectively provide 127 levels of amplitude out of 20 layers.

I'm afraid I don't really know enough (do you?) about this interpolation business (especially when you're playing through a controller rather than a MIDI file)? Why claim only 20 velocity layers if in fact you can generate 127 volume levels? What in fact does it matter if you have 4 velocity layers or 90 velocity layers if you can interpolate and still get 127 volume levels (matching the MIDI values)?

The way you could measure is to program one note to play at all MIDI values, then play this through the CFX VST and use some kind of sound level meter to distinguish the decibel reading for each MIDI value played through the CFX VST. I don't have a sound level meter set up for that.

Anyways, to get the discussion moving further than this...
There are many other factors that make Modelled pianos more expressive to play (not just volume levels). I believe this explains why Roland moved from sampling to modelling rather than sticking to the hybrid approach.

I took the MIDI from whatever the top Google result for "classical piano midi" was.

My thoughts:

I don't think there is one "right" solution for the perfect piano representation. When you play a piano, you are hearing so many different things: the soundboard resonating with the combined vibrations of all the strings, the depression of the keys, the movements of the hammers, the damper moving up and down, the reflection of the sound off the walls, and even more, I'm sure.

To expect a perfect piano sound to be generated from a stereo setup, is, in my opinion, ridiculous. Think about how massive the "speaker" is on a piano. How can speakers even try to duplicate that, from a physics perspective?

I prefer playing on modeled pianos. Modeled pianos feel more responsive to control. The sympathetic resonance sounds better to me, as well. This is not to say at all, that it sounds like an acoustic piano. It doesn't at all. But, it's a lot of fun.

Whenever I play a sampled piano, I always feel there's something a little left to be desired. Maybe it's because no one records 128 velocity layers, or the sympathetic resonance of every possible combination of keys. It sounds real, but it doesn't feel real to play, like I'm controlling a puppet with strings or something. Some VSTs use "reverb" as sympathetic resonance as well, and it sounds terrible.

That said, I use each for different purposes. If I'm playing live, I'll use a modeled piano, since it's probably a cocktail party or a band gig. If it were a classical gig, they'd probably have a piano there already. If I'm recording solo piano, or for a track with other acoustic instruments, I'll record the MIDI data while playing a modeled piano, and use a sampled piano for the actual track. If it's for a rock / electronic track, then I'll use modeled as realism isn't as important, in my opinion. If I'm practicing at home, it'll be the modeled pianos on the RD-2000 + headphones or studio monitors.

I think there is also a lot to be said about tweaking the keybed to the software. I know we can calibrated them for all the decent VSTs, but it's not the same as having Roland tweak their own keybeds to their own software, or any other company doing the same. There needs to be a good connection between the hardware and the software/firmware, otherwise it'll result in a bad experience for the player.

Finally, you can do a lot with EQ/compression/reverb if you have basic sound engineering skills. Even a mediocre piano can be made to sound half decent (but this may only be possible in post-production for many of you).

There's no mistake I seriously dislike Pianoteq, and version 6 is yet another reason why. I was thrilled to try out the latest hoping that they really made it revolutionary, but it seems only slight improvement were made in the tonal department.

It still sounds like it has the same issues, and all the overhauled pianos sound like they've had their EQ adjusted, which of course is something you can do with version 3/4/5.

Where I really hear the difference is from version 3 to version 5. It seems that was the big leap.

Quite simply, we have only two ears and two eyes! So we hear in stereo, and see in 3D vision.

I believe I know enough, yes. Someone may correct me but I think the massive Vienna Imperial boasts 127 separate layers (or at least more than any other sample library.) AFAICT that in itself is a potential minefield of difficulty for realtime performance since the undoctored samples are unlikely to be as smooth in transition as those that are crossfaded with fewer layers. So layering matters to some more than others. Some would like the apparent attention to detail where every note has its own sample. Others will see the wisdom of compromise.
If you want more info on interpolation, there's extensive documentation on crossfading for the Kontakt engine. Check it out.

This point is made now and doesn't need repeating multiple times (if Pianoteq are reading posts here, they've probably picked up on this sentiment). Either you like the tone (or don't dislike it enough to stick with sampling) or find it initially ok then later get frustrated by the metallic nature, or you just don't dig the tone at all.



Yes, this is logical too because they've been adding more and more instruments to the library since version 3, meaning it's taking them longer to release updates and each update seems to be small evolutions rather than big steps. The same point has been made a few times in this thread.

Modartts business model is to bring out more models because they can make more money selling many instruments, and re-invest that into their R&D. It's growing their business faster to do it that way i.e., rather than to concentrate on improving the modelling for one piano model only.

Also, I suspect that piano modelling advances evolve with basic R&D, which moves at the academic output rate. Maybe faster advances will occur if someone makes a breakthrough or if computer technology suddenly takes a leap forward.

Thanks for the link. I'm going to read about it. Heard many people refer to the Kontakt engine, but never really read about it!

We do hear in stereo; however, there is a difference between the various sound reflections/absorptions and transmissions when comparing an acoustic grand to a stage piano with stereo speakers. I guess that's why Yamaha, Kawai and Roland have spent lots of time trying to place speakers, generate sound-boards etc to better mimic this.

Thanks for that post!
I remember Phil Best commenting on YouTube that Pianoteq 5 and the V-piano had different qualities but he like both. Do you feel that Pianoteq 6 competes well with your RD2000's (despite the keybed comment)?

I did think it was interesting that Phil Best had a Roland FP-90 that he was using in the Pianoteq 6 videos.

Think about it like this: If you were to shout "Get off my plane!" at 127 different volumes, the tone of it would change gradually as you got from softer to louder. At the quietest, it would be a sweet (maybe) sounding whisper. At the loudest, your mouth might be wide open and you may get some throat growl in there.

To reduce the amount of work, you could just record/sample a "quiet" one, a "medium" one, and a "loud" one. You could assign the "quiet" one to velocities 0 - 42, the "medium" one from 43 to 90, and the "loud" one from 91 - 127. Then, you would adjust the volume for each such that the higher the velocity, the louder the volume. While the volume changes from 0-127, you only hear three different tones. The same thing happens when the hammer strikes the keys at different velocities. It's not just a volume change that you hear on an acoustic, but a tonal one as well.

Piano modeling tries to find a mathematical equation (highly complex, I'm sure) that captures all of these characteristics across the velocity range, and how the strings interact with each other. So, while the problem of velocity layers is eliminated, there's a more complex one of the modeling being realistic enough for the human ear to recognize as a piano.

I'll let you know what I think of the RD-2000 / Pianoteq combination after I try it out a little more. Haven't had much of a chance yet.

Perhaps Phil likes the PHA-50 better than the PHAIII of the V-piano and the NW-GH of his CP4?
On his website it says he uses the FP-90 for live performance, so I guess he doesn't use Pianoteq much on stage.

Sample-based synthesizer hardware can morph velocity layers since decades. There is no modeling needed for smooth layer transitions from pp to ff. Software libraries took very long to get this right (if at all).


I'm comparing Pianoteq, Ivory II American Concert D (22 velocity layers) and the Kawai EX built into my digital piano over the course of three years and the dated sample inside my DP became my favorite instrument. It sounds great, especially in the forte-fortissimo range, unlike Synthogy's Steinway (there might be great sounding Steinway Ds, but the instrument Synthogy chose to record was not one of those) and is perfectly playable just as Pianoteq's model.


Rendering MIDI files has nothing to do with playing piano at all.

Ah, so the 20 volume levels on the CFX are actually 20 different tonal characteristics. That makes sense, thanks!

I felt my MP7 EX grand sounded great with the Chopin MIDI file.
Dire tonic and HSFlik have just explained about the smooth layer transitions thing: quite interesting!

You forget about the sample crossfading/morphing interpolation. Most (if not all) samplers gradually blend the samples, so it's not only the volume that changes, it's also a smooth and gradual timbre variation. You can see in the DPBSD test that almost all of the tested hardware and software pianos utilize such a technique.

Not saying you've been mislead or anything but I can easily see how a false assumption like that could lead many people to incorrectly believe it's only modeled pianos that give you smooth velocity and timbre gradation, hence they must be better than sampled.

Hi Doug,

thank you for your interesting suggestion. We have prepared a special
comparison page between Pianoteq 5 and Pianoteq 6.

Philippe

Thank you for the correction!

I think it's all personal preference, not necessarily one is better than the other, anyway.