Modartt user forum - Modartt strategy

Re: Modartt strategy

2018-06-21T12:51:33Z

bm wrote:

During your short test with Steingraeber, did you use this model directly with its code length of 2.72m compared to the actual Bösendorfer 200 recordings?
...
Regards,
Bruno

Hello Bruno. My approach is a bit simpler than yours...I merely offer the pianoteq model a different starting point to do its real time modeling. Starting from the closest instrument in sound and size gives better results since so much of the real-time sound generated is not due to the starting spectrum itself and out of reach, as is the attack sound itself. In fact, what I did is a little more complex Steinway B fxp than can be done by hand. Reducing the string length works of course, but it's better if you can avoid it.

Since we seem to have similar background and interests, I suggest you email me (in french) through the user list pianoteq emailer so we don't bother other people less interested in these matters. I'll be happy to go into more details if you want.

Regards,

Gilles

Re: Modartt strategy

2018-06-21T06:38:02Z

In addition, here is the position of RODE M5 mics that was used for the capture of Bösendorfer 200 used in my .fxp (the real piano samples have already been put on the site, I only used the first moments, for each note, 'as short as the note was acute)
this is a very approximate XY because I do not have a bar long enough (I have no professional experience of sound recording some piano),
regards,

Bruno

BOSEN200_MIC-POSIT_VIEW-1_20180309.jpg
BOSEN200_MIC-POSIT_VIEW-2_20180309.jpg
BOSEN200_MIC-POSIT_VIEW-3_20180309.jpg

Re: Modartt strategy

2018-06-21T05:52:49Z

During your short test with Steingraeber, did you use this model directly with its code length of 2.72m compared to the actual Bösendorfer 200 recordings?
(I say that these are raw recordings, but I only use the beginning of each note about 0.7 s for the first octave, dividing by half this duration with each new octave, and using a parallel window 2 times shorter with each new octave (65k then 32k up to 0.512k for the highest notes)
For my part, with the model Bluethner I first reduced the size of the strings to 2m (more exactly 2.1m) before making the comparison, so that the high partial frequencies can remain comparable between the real piano and Pianoteq.
I also avoided using for the partials simple multiples of the fundamental, by aligning me on the frequency of the partial as indicated by pianoteq for each partial in the spectrum profile window. To integrate a few more partials than what Pianoteq presents today in its spectrum profile window, I extrapolated the frequencies of the top partials not presented in the pianoteq window with a polynomial function (trend) by relying on the partial frequencies described by pianoteq. (Pianoteq seems apriori capable of exploiting partials well beyond the limit of the partial No indicated in the spectrum profile window, but on the other hand does not know (today) not manage them with the function of rescaling, which requires to reconstruct either this rescaling function itself) I set myself today a limit of 250 partial and a maximum frequency of 45Khz (that neither my ears nor my scan card from micro can not handle ...). A difficulty is to consider the string stiffness that seems different from one piano model to another, and I guess may be different with Bösendorfer. To adjust the length of strings to use with Pianoteq, I check if the frequencies of partials a bit high (around 8khz) correspond between pianoteq and the real piano for level peaks, and for the rest trust my ears. (Hence my approximation of 2.1m for piano pianoteq compared to the real piano of 2m). This is totally empirical and maybe a wrong approach, I do not really know.
Regards,

Bruno

Re: Modartt strategy

2018-06-20T14:12:58Z

bm wrote:

(If Gilles had the opportunity to test the Bosendorfer 200 on Steingraeber with his software, that would be great)

I did consider it but the distance is too great with the Bösendorfer 200, a smaller piano with a much different bass tone, one that is much more like the Steinways.

Re: Modartt strategy

2018-06-20T13:43:54Z

bm wrote:

This attempt to create a .fxp from a Steinway B model is very interesting. I tried at the beginning without success to model some notes of this Bosendorfer from the model Steinway D (too different).
Gilles, at what stage of the envelope ADSR and on samples of what length do you capture the notes on the real piano and on Pianoteq (before comparison of the spectra). If this sample length varies according to the pitch of the note, do you give yourself a particular rule for choosing the start and duration of sample capture? What is the spectral analysis function do you use (rectangular)? The choice of this spectral function (rectangular, Hanning, Blackman, ..) is it a feedback of experience after comparison with other methods? Which method do you use to initially set the recordings of the real piano and Pianoteq at exactly the same volume level before comparing the spectra for a given note?
...
Regards,
Bruno

As I write in the PtqSpecProf pdf doc, I tried many lengths but short ones (around 1.5 seconds) give good results as well as being practical to shorten the total work. Your samples are longer but I still compared with shorter ones from pianoteq, what is captured really is an average spectrum at the start of the note, the decay is to be handled by pianoteq itself. I can use all the common windows, but for a short sample rectangular works fine. My default number of samples for the FFT is 16384, but I can go higher. I found the relative volume has no real importance (apart from the spectrum content itself) since the spectrum is more or less auto-scaling from 0 dB down. I also center the resulting changes around the average value.

This is all amateur work, I am not an expert in signal processing, but conversations with Philippe at the time revealed no bad mistakes on my part if I remember well.

One very useful tool in the Spectrum Profile tool is the rescale function. I use it a lot when the given result sounds bad. It keeps the change in color found but reduces it if it is not natural. The tricky octave is the middle one, where the balance is difficult to get. In the lower registers there is a lot of harmonics so the result usually is ok as is. In the higher registers, differences in spectrum are less important than the attack itself.

Re: Modartt strategy

2018-06-20T13:24:45Z

dklein wrote:

Very nice, Gilles, and so much 'cleaner' sounding than the preset based on the Bluethner for the Bosendorfer 200 - though the tone is different, the Bluethner version sounds as if I am listening through a sheet of drywall/gypsum board. It's so hard to get these things "right"! I have frequently lost patience when I try to tweak things.
I love the Steinway B, ever since version 6.1, which really breathed new life into it. Closest thing to my Mom's 1957 "M".

You are right that the Bluthner version is much less precise, but remember that it was done by comparing with pianoteq version 4. I did not redo the analysis I just rendered the same fxp with version 6.2.0.

The "big and round" sound obtained with Bluthner is still interesting for some repertoire, but of course the Steingraeber has attained such a degree of clarity that everything pales when compared to it, even the more recent Steinway B...

EDIT: I think I read you wrong and you were talking about Bruno's fxp, not mine for the 290...sorry. I guess the difference is due to the distance between the Büthner, a large piano with the much much smaller 200...

Re: Modartt strategy

2018-06-20T12:19:09Z

In fact, any good transient designer will do... NI has one that is pretty great as well.

Re: Modartt strategy

2018-06-20T10:38:09Z

Philippe Guillaume wrote:

dklein wrote:
OK, then let me re-phrase the question - What controls other than hammer noise can 'sharpen' the attack of each note? A sharp attack appears to be part of the brilliance that is magical when standing or sitting very close to a real piano.
Maybe effects (EQ, compressor)? Post-processing? The "distance" between a raw recording and the final sound, whether it be music, samples library..., is simply amazing.

Hi Philippe
Hi dkein

I think Flux Bitter Sweet transient designer will be your best friend to control notes attack in Pianoteq without compressing the sound nor modifying the hardness of the hammers.

Cheers

Mickael

Re: Modartt strategy

2018-06-20T08:01:04Z

dklein wrote:

OK, then let me re-phrase the question - What controls other than hammer noise can 'sharpen' the attack of each note? A sharp attack appears to be part of the brilliance that is magical when standing or sitting very close to a real piano.

Maybe effects (EQ, compressor)? Post-processing? The "distance" between a raw recording and the final sound, whether it be music, samples library..., is simply amazing.

Re: Modartt strategy

2018-06-20T07:24:14Z

OK, then let me re-phrase the question - What controls other than hammer noise can 'sharpen' the attack of each note? A sharp attack appears to be part of the brilliance that is magical when standing or sitting very close to a real piano.

Re: Modartt strategy

2018-06-20T06:31:29Z

dklein wrote:

By the way, Julien and Philippe, Gilles brings up a good point - can you grant us a control for note attack? I had found that to be an issue when I was trying to emulate my own piano - the 'liveness' of the real thing is often conveyed by note attack and then decay. In Pianoteq, the Hammer hardness is one way to influence attack, but then it doesn't really allow for both a hard attack and then a soft decay, which is key for an 'in-close' live piano in front of you.

I would say the decay does not really concern the hammer noise, but essentially the string partials, which can be controlled by several parameters including unison, direct sound duration and soundboard impedance.

Re: Modartt strategy

2018-06-20T06:09:03Z

This attempt to create a .fxp from a Steinway B model is very interesting. I tried at the beginning without success to model some notes of this Bosendorfer from the model Steinway D (too different).
Gilles, at what stage of the envelope ADSR and on samples of what length do you capture the notes on the real piano and on Pianoteq (before comparison of the spectra). If this sample length varies according to the pitch of the note, do you give yourself a particular rule for choosing the start and duration of sample capture? What is the spectral analysis function do you use (rectangular)? The choice of this spectral function (rectangular, Hanning, Blackman, ..) is it a feedback of experience after comparison with other methods? Which method do you use to initially set the recordings of the real piano and Pianoteq at exactly the same volume level before comparing the spectra for a given note?
It will undoubtedly be necessary in the future to make a dedicated topic for these types of more technical questions that move away from the queries of users of this forum on Moddart's strategy.
For my part, my goal is not to make "alternative" piano models to those of Moddart necessarily much better equipped (especially with the mathematical modeling of the interactions between components of an instrument), but to learn how to help tools of Pianoteq pro to better discover how it is possible to create new sounds, starting (1st step for me) to get as close as possible to real instruments, which is possible only with access the instruments in question because it must be able to play or possibly recordings of real instruments of good quality but it makes the work more difficult without the feedback of the actual keyboard and the ability to control the sound recording.
I also wish, for sentimental reasons to try to "freeze" in fxp the timbre of the instruments that have rocked my childhood, before these instruments age too much, may be a day less well maintained while I'm not sure in the long run to recover them all.

I would like to try the Steingraeber model but I will need more time because my current treatments are only partially automatic. (The blackmann-harris spectrum analysis I'm using comes from Audacity) which forces me to export / import .csv for each note.
(If Gilles had the opportunity to test the Bosendorfer 200 on Steingraeber with his software, that would be great)
My professional activity leaves me only "windows" of an hour and a half very early in the morning and I still retain some of this time to learn to play better or compose (with Pianoteq of course). Near each lunch, I play again on the real Bosendorfer 200 family to keep in touch with the real sound that should guide me to improve my .fxp. To deepen my learning of the timbre of an instrument on a given note, I found it interesting to put on Audacity the free plug-in TRT Nova, which allows to listen in a comparative way in real time a recording by analyzing a thin slice of frequencies (that can be varied in real time) with a slope of 72 db / octave outside the frequency band analyzed. It helps me a lot to understand (unfortunately for me not related to mathematical modeling) how each string can "sing", with still very different modulations between Pianoteq and a real piano. The study of the ADSR envelope is also interesting, the structure of the attack phase remains very different for the low and mid-low notes, the decay phase is generally much shorter with Pianoteq than with a Bosendorfer 200 and the Pianoteq pro adjustment of the impedance & Q factor does not seem to allow to adjust it, acting mainly on the sustain phase (if there is a setting of the decay duration I am a taker). Reminder: I am only an amateur curious to better understand how is formed the sound of a piano, which is also a help to play discovering new sounds that can be reproduced on the keyboard in a composition or interpretation .

Regards,

Bruno

Re: Modartt strategy

2018-06-20T03:35:51Z

I see the birth of some sort of "Bosendorfer ? ...me too" movement in this forum.

But I bet the imperial Bosendorfer add on probably will be created more later, when Modart technology will be even closer to please the few ultra hard critic guys who love samplers. And a Bosendorfer it's large, have extra strings, and I bet it will take some few move computer power.

Anyway it's interesting to see you guys trying to emulate a imperial Bosendorfer using the already exiting models.

Re: Modartt strategy

2018-06-20T02:22:34Z

By the way, Julien and Philippe, Gilles brings up a good point - can you grant us a control for note attack? I had found that to be an issue when I was trying to emulate my own piano - the 'liveness' of the real thing is often conveyed by note attack and then decay. In Pianoteq, the Hammer hardness is one way to influence attack, but then it doesn't really allow for both a hard attack and then a soft decay, which is key for an 'in-close' live piano in front of you.

Re: Modartt strategy

2018-06-20T01:52:35Z

Very nice, Gilles, and so much 'cleaner' sounding than the preset based on the Bluethner for the Bosendorfer 200 - though the tone is different, the Bluethner version sounds as if I am listening through a sheet of drywall/gypsum board. It's so hard to get these things "right"! I have frequently lost patience when I try to tweak things.

I love the Steinway B, ever since version 6.1, which really breathed new life into it. Closest thing to my Mom's 1957 "M".