Modeling a sampled piano

This is a huge, deep, and fascinating topic, that I don't believe I could begin to cover in a single post.  I will give a short summary of how I create presets, and I hope that will give you some ideas.

I start by finding source material (which you've already done), and this can be a terribly difficult step for obscure brands or one-of-a-kind instruments.  In your situation, I would make several recordings on the CA97 both of the sound and the MIDI data.  This will give you something that you exactly replicate in PTQ.  When I'm using recordings, youtubes, or playing an instrument in person that I want to recreate, it's impossible to create a "perfect" replication of the source, but since you're starting with digital playback data in the CA97 engine that you can save as MIDI data to the computer, you can very easily overdub the recorded sound from the CA97 and the real-time sound of PTQ playing the MIDI data from the CA97.  You can loop something like that in DAW for as long as you need while testing PTQ settings and really, really get a precise result.

After getting all of my sound reference material where I can work with it, I gather any technical information that I can.  For example, the SK-EX is 276cm in length, and has agraffes to note 57 and a front and back duplex. https://kawaius.com/product/sk-ex/  Photos like this one: https://i.pinimg.com/originals/cc/00/c7...05cf22.jpg show you where the dampers stop (PTQ uses the MIDI key number not the keyboard key number), and the felt shows you how many strings and which strings have additional harmonics that are voiced or felted.  Look at the pictures of the piano as much as possible to deduce particular details about the design: where do copper-wound strings start and end (changes volume and character), where does the bass bridge break from the treble, does it have a capo d'astro bar or agraffes (which control how it's tuned and some aspects of the sound), etc., etc., etc.  The pictures will tell you tons of information that no spec-sheet will offer.  Any data that you extrapolate from the instrument design (like length, last damper, etc.) can be entered directly into PTQ settings as a starting point--usually under "Design" in the UI.

I find it nearly impossible to do any of this with PTQ Stage.  Standard gives you moderate control and good results, and Pro gives near total control and excellent results.  I highly recommend Pro for this kind of project, but Standard can suffice.  In the past, I've had the best results copying the Shigeru Kawai SK-EX with the K2 library in PTQ.  However, I did that two or three years ago, and I haven't made it back to reworking those particular PTQ 5 profiles for PTQ 6.  The Steingreaber, Bechstein, and Petrof have all been released since I made my presets, so it's possible that they will work well--they're definitely worth giving a try but the K2 would still be my first choice for this.  The Steinway D isn't necessarily a bad place to start either.  I'd avoid the Bluenther for this, because the SK-EX has a duplex and no aliquots, and the default spectrum profile curve (only available for editing in Pro) for the Bluethner is rather aggressive and gives a very particular sound, while the K2 has a very flat default curve which will be easier to work with.  I suggest starting with loading a "Player Perspective" profile or something similar since your CA97 will likely default to that type of mic position.

Once you've selected with PTQ piano you'll use as a starting point, you compare the sound you're getting from PTQ and the sound from the CA97 until you find settings that match.  Enter any details from the technical references (like length, etc.), and then work with the other settings.  Design is a good place to start, and then you refine the voicing, spectrum profile, and resonance EQs.

I will work on a preset until I can't really tell the difference between the original recording and PTQ.  I also ask members of my family to listen, since new opinions can be very helpful.  As a final thought, books like Samuel Wolfenden's Treatise on Pianoforte Construction or Del Fandrich's posts on the PianoWorld forum are goldmines for how to design pianos, which really helps to refine one's use of the settings in PTQ.  This lecture series is also very good too: https://www.speech.kth.se/music/5_lectu...tents.html

Good luck, and keep us posted on how this project goes!

You're welcome, and also, you could post what the settings within the Virtual Technician in the CA97 you've used, and the community may be able to give you estimated equivalent settings.  While the names inside the software likely will be different, there are usually fairly easy ways of translating one digital system nomenclature into PTQ-friendly terms.

I can also dig around for my Shigeru Kawai SK-EX K2 preset, in case that speeds up the process for you.  It's old, but it should be a decent start.

Yes my software could be a start and I'm ready to help if needed.

If davidizquierdo82 can use an external program to run the MIDI file documented on the site to be played by his keyboard and record the output as a wav file, he could do the analysis and easily produce a first draft fxp starting from a suitable Pianoteq instrument that he owns.

I'm willing to do the analysis of the wav file if Java is not available to him or if he finds the process too complicated.

i apologize if this is a silly question, but is the source code available, e.g. on github or something similar?  i would  really like to look at what's going on under the hood.  if it were a public project maybe people could contribute to it.

Thanks so much for your appreciated help.

I've been working a bit on the steps mentioned above and created a public folder (microsoft drive) where I'm sharing any progress I'll be making:

https://1drv.ms/u/s!ArFgjlBHEgvrgQ9Su0W...N?e=tY0jsa

For now, I've managed to create a complete sample of the 88 notes, 60 seconds each

In the folder, you'll see the MIDI file that I created for that purpose, as well as the WAV file that resulted from rendering it.

As my piano is set default to 432Hz (intentionally), that's what I used for the rendering (I don't know how much important that is to create a model) but I could get it to 440Hz if needed.

Also my Virtual Technician settings:

01. Touch Curve: Normal
02. Voicing:Normal
03. Damper Resonance: 8
04. Damper Noise: 2
05. String Resonance: 7
06. Undamped String Resonance: 9
07. Cabinet Resonance: 5
08. Key-off Effect: 9
09. Fall-back Noise: 6
10. Hammer Delay: 4
11. Topboard Simulation: Open3 (max) 
12. Decay Time: 8
13. Minimum Touch: 1
14. Stretch Tuning: Normal
15. Temperament: Equal
16. Temperament Key: C
17. Key Volume: High Damping
18. Half-Pedal Adjust: 5
19. Soft Pedal Depth: 3

Reverb and Effects are OFF in the samples (although I have them ON when I'm playing), settings:

Type: Lounge
Depth: 5
Time: 6

All rendered notes in this folder are at a velocity of 100

My next steps are rendering them at 25, 50 and 125 as further references.

Also, I'm going to add two or three complete songs rendered (MIDI and WAV) although it's gonna be a bit tricky since I'll be rendering a MIDI created on other piano at different velocities than it would have been played on mine, thus resulting in a sorta awkard dynamics, but it's good enough though as a reference of the piano sound I want to achive as a whole.

I don't know whether I'm missing anything, I'll keep reading any suggestions or comments you may want to make.

PS. In case MODARTT reads this, I'd be pleased to provide as many samples as required if it helps creating an actual instrument you can then sell, call it SKG..

Regards,
David

Hi Gilles,

Thank you sir, I've checked the website and used the MIDI file there to render it as suggested, I've uploaded both MIDI and WAV files (440Hz) to a folder called Gilles_Collab:

https://1drv.ms/u/s!ArFgjlBHEgvrgQ9Su0W...N?e=tY0jsa

The MIDI file is basically the same I designed for this purpose (if you check the other folder), only I've made the notes last longer (1 minute) so that I could capture as much information as possible. In the resulting separate WAV files you can see how detailed (and yet not enough in the bass notes) the samples are.

Splitting the entire audio file into separate notes is not a big deal since I used an automated process, so just let me know if you need the WAV in your folder to be cut into small chunks.

IMHO the rendered notes when using your MIDI file, are too short, they are not showing the decay signal properly, which I assume is not critical for a first draft of the instrument.. but maybe crucial for the final result when designing the resonnances, not sure though.

JAVA program that you included in the website seems very cool stuff but I was not able to open it (Windows 10), not sure whether because of my system or I'm just too clumsy for that but I'd definetely love to learn about it and how it actually works. Also, as it's a spectrum analyzer, whether wouldn't there be any other options inside Ableton (my DAW) or any other to do the same job?

Another question rises in my head, wouldn't we need samples of other velocities too as signal behaviour may change according to the different velocity layers? Just a thought.

Regards,
David

Thanks for the suggestion dude, but I don't find the compatibility mode unlike other programs where I know how to set it, but here's no options..

I don't even believe pppp is possible, it's just a mith to make us mad frustrated trying all time.. xD

Anyway, thanks for the explanation, I guessed so far but I'm still thinking the signal behaviour might differ at lower velocities.. just for the sake of getting as much an accurate model as possible (if feasable).

Regards,
David

Hi again David.

I found the Steingraeber a really very good starting point for your samples and I didn't even have to do any correction, the fxp is straight out of PtqSpecProf. As I mentioned in the documentation, I prefer the stereo output to minimize phase cancellations that can happen when using mics. If you want to use the Sound Recording option, you might have to move mics around a bit.

Since I saw on your YouTube channel a rendition of a Chopin Waltz, I tried to find it in the Minnesota collection but to no avail, so I used a Mazurka that is similar in atmosphere in my opinion.

If you compare side by side the Steingraeber E-272 preset I started from with the resulting one, you will hear the slightly rounder tone that your samples have compared to the Steingraber. I hope you like the result and can use it as a basis, if not I can try another Pianoteq instrument.

Regards,
Gilles

https://forum.modartt.com/file/3ysgn70v
https://forum.modartt.com/download.php?id=3808

Hi again Gilles,

I managed to run your JAVA application, and I just can say wow it's really smart logic behind it.

I'm missing though the backward button (previous note), or better yet, the capability of chosing the note you need/want.

I rendered a Steingraber scale (your MIDI) and used it as source waveform, and loaded up my rendered scale CA97 as target. What I understood if I'm right, the
analysis starts from note 24 (C0) and goes all the way up until 107 (B6) as per MIDI notes in your file

My doubt, I'm copying the resulting spectrum (the corrections) and I'm pasting them on each note individually on PTQ (spectrum profile tab), and it's giving a modification of frequencies not only in the affected note, but along the rest of the notes too.. is it correct what I'm doing?

Now, what if I used the single note files (the trimmed big ones) instead of the scale? I'm pretty confident I'd be getting different results.. the question is, would it be more accurate or doesn't make sense the extra effort an time that it takes the whole process?

PS. I've been listening again both rendered WAV (CA97 vs. FXP) and they really really sound very close, specially when not much notes played. I've observed though that the FXP sound is more punchy in the middle notes, a little bit more aggresive. Also, the bass is brighter in the FXP I guess, I assume that can be corrected..

This is the FXP I came out with, no reverb no effects, used a "Steingraber Player" patch, and also this is a MIDI file rendered with it (Chopin Ballade 4):

https://1drv.ms/u/s!ArFgjlBHEgvrkG4IThf...4?e=dbd0Xu

https://1drv.ms/u/s!ArFgjlBHEgvrkG8Bpi25wsuVsVNZ

Regards,
David

Good that you can make it work. Yes there is no previous note button the process goes relentlessly forward for optimization reasons due to the older slower computer so you have to be careful in the process.

You can use what I called the File function to analyze separate notes of any size in any order as long as you take care to specify the correct MIDI number. Feel free to try that with your separate notes but I doubt the result will be much better, slightly different maybe.

By the way, at the time it was only possible to paste one note at a time in the Spectrum Profile so that directed the design. Later the possibility was added to paste many notes at a time so I could rewrite the process to be just an app without GUI that generates the whole 7 octaves at once and do one paste, but the slower note by note method allows to validate visually that the spectrum is indeed correct.

If you see that a paste seems to affect following notes it's because you forgot to first use the Edit All Notes option available in a menu when you click on the bottom rail. It preselects all notes so each change affects only the current note.

I have to go for now, I'll listen to your generated fxp later and come back to you.

Yes there is a lot of phasing artifact in what you produced, the only reason is you probably started the analysis with a player preset with Sound Recording on (mics). Summing this to mono inside PtqSpecProf generates a lot of cancellations so you absolutely have to start with a stereo or better a mono preset. This is what I did.

I guess my offer was the best possible then, so yes you better keep playing the samples, make great music and wait for Modartt to maybe offer a Shigeru Kawai in the future. I think there are all kinds of copyright problems though that may prevent that from happening. Not all companies are as accommodating as the ones that have already approved the modelling from Modartt.

As for your questions on the Edit All Notes option, see page 66 in the manual where the interpolation is mentioned.

Regards,
Gilles

Hi Gilles,
I saw this post and downloaded the PtqSpecProf.jar, terrific software, it ran on w10 (fully updated) with JDK8, not bad for a decade old jar! The only glitch was when using (24-107) .wav files 16 bits 44.1KHz, “NextNote” at 106 resulted in a stall (not a crash) due an “index out of bounds” java exception. No biggie, I did single notes for 21-23, 108 and the missing 107, just fine! This was using the SevenOctaves100.mid @ 40bpm, for a full keyboard. The results were extremely encouraging, in trying to model akai’s “splendid grand” samples to a Grand Steinway B preset.

So, thank you so much, this made things so much easier...

88 copy-pastes is a little daunting, but here’s a tip: if each paste is on a single line in a .txt file, this can be easily changed via search/replace to create a full keyboard spectrum file for easy loading in a preset.

Regards,
Tez

Yeah, I noticed that, when running java and not javaw... But this dummy didn't think to do a "Console" window copy! Very nice tip, that speeds things up a whole bunch, now I can try multiple presets quickly. My Akai samples I think are Steinway B, but there's little history on their exact origins, so now I've got more flexibility. I'm using stereo .wav files, so I assume as you said previously, pteq stereo is the best, but volume is a bit tough as it seems pteq has slight random fluctuations on a per note played basis, any recommendations that could help?

Cheers,
Tez

Oooops, I had not tested that before, and the syntax for a group paste is slightly different than a single one, so one has to do some editing before pasting the whole group of notes, so not really so easy. Sorry.