Synthetic piano

The CP33 of the last 20 years has only 32 voices & no sympathetic resonance.  That's low hanging fruit for a synthetic replacement.  The minimum improvement would just capture all of its samples at all loudnesses & play them back through a full 88 voices with simulated resonance. 

Sympathetic resonance is actually pretty hard.  How loud to make it, how many notes to make it, are all unknown without an instrumented instrument.

Improving the damper sounds from pianoteq would be quite difficult.  It needs prior information to use samples.  A synthetic damper could generate a certain loudness for a certain rate of pedal movement.  A sampled damper needs to know the full extent of the pedal movement to pick the right sample.

In the CP33's defense, dampers aren't audible in most music.  They're manely audible in extremely quiet passages.

Nice photo of a CFX.  You can count the number of single strung notes.  The stock soft pedal sound would work for the bottom 8 notes.  The next 5 are double strung.  The upper 75 have 3 strings.

Testing the Pianoteq binary, it actually has a headless mode & supports arm64.  It'll run in a container, export a .mid file to a 24 bit 48khz wav file.

It uses maraca samples to synthesize the damper noises.  Rapid pedal movements work & sound reasonable.  Gradual or half pedal is a failure.  It replays the same sample every time the damper value changes, like a shaking maraca.

Una corda just reduces the velocity codes the way an upright moves the hammers closer to the strings.  It doesn't perform any lowpass filtering or gain reduction like the CP33 & it's certainly not muting any of the strings.  The damper noise & una corda were basically the only reasons for trying Pianoteq.

The fact that no-one demos una corda or half pedaling doesn't bode well for any other windows tools.  They all demo the same jazz sounds.

There's still a way to reuse the pianoteq damper sound elsewhere.

As lions pivoted away from ever owning a concert grand, notes grew for how to construct the best possible synthetic recreation of a concert grand.  There should be some attempt at it before spending the money on a real one.  The Yamaha CP sound has been good enough for the last 20 years to never get around to it.

The 1st problem is creating a synthetic una corda sound.  It's not really possible with any kind of post processing.  Additional sounds no-one else sythesizes are the dampers rubbing, sympathetic resonance.  Damper sounds would be quite difficult to generate with samples.  They interact with the string vibration.  A clean sample of a damper with no string vibration wouldn't match the sound of  damping a vibrating string.

Robert Estrin connected real steinway D actions, probably made of spare parts from scrapped pianos, to https://www.modartt.com/ Modartt pianoteq.  That is an algorithmic synthesizer instead of a sampler.  He never mentioned what sensor went into the action.

The leading sampler is now pianoverse.  It used to be synthology Ivory.  They're all VST plugins which require a GUI.  Modartt is the only one with a dedicated Linux VST plugin.

They're all binary only.  The problem with binary only solutions is they don't work after a few years, even if they're perpetual.  The manetainers disappear.  Past versions of Linux are no longer being archived, hardware is being forced into obsolescence, so you can't perpetually have the 1 system that worked with the binary.

There are ways to just capture all the possible sounds & use them in a source compiled sampler.  The cost of obtaining the original samples could be easier to justify if their destination is permanent ownership in another program.  At minimum, it's 88 notes x 64 velocities x 2 una corda positions.  It would probably compress the samples in a neural network. 

There are ways to hook a Windows VST plugin into Linux through wine,

https://github.com/osxmidi/LinVst

The fastest way to get started would be to hook a CP33 as a keyboard controller into Modartt on a Linux box just for that, with its own monitor, keyboard & mouse.  Lions are in the same situation they were in 30 years ago, with the only good solutions being expensive binaries.   30 years ago, mp3 encoding required a $200 encoder & photo editing was $300.

-----------------------------------------------------------------------------

Lions previously tried using optical sensors in an action mockup.  It had a very temperamental & unstable alignment.

The typical mechanism now has 3 carbon dots hitting a PC board in different positions of the key travel.  They detect acceleration, velocity, damper position.  The Yamaha CP series probably has the best synthetic actions.  Helas MIDI doesn't encode the acceleration.  That would probably take some hacking to expose.

The sound actually was affected by how quickly you released the key &  where in the keypress you accelerated.  Whether the goal of the action should be to eliminate such effects or enhance them has been debated for all keyboard instruments.  With a binary only synthesizer, the effects would require capturing all the possible outputs from the synthesizer with some kind of tone changes & making a sampler that blended the tone changes.

 It would not be possible to create an acceleration based tone changer with samples & lions have no practical example of tone quality changing with acceleration.

-----------------------------------------------------------------------------------------------------------------------------------------

The 3rd problem is the controllability.  Lions have never been able to control loudness on electric actions as well as a Steinway, all the way from the K3 to the CP series.  The CP's have an exponential growth in loudness with velocity, exploding suddenly at around 3/4 of maximum velocity.  The 80's synthesizers had very little response to velocity in any range.  It's believed some kind of MIDI translation table might solve this problem.  It might also be caused by a change in samples at certain velocities.  That would be solved by a software synthesizer.

-----------------------------------------------------------------------------------------------

The initial workflow would involve capturing all the MIDI codes in parallel with the stock PCM audio.  Then the MIDI codes would be converted to a PCM audio file offline.  Timestamps would be adjusted to match the PCM audio.  This wouldn't require any hacking.  It could use the stock binary.  Just getting una corda & sympathetic resonanace would be an incremental improvement.

Realtime synthesis would be a later goal & that would require hacking to run without a GUI.