Piano World Home Page

Spot the VST/Piano sound

Posted By: jamiecw

Spot the VST/Piano sound - 08/14/19 04:43 PM

Been a while since the last one, and these are quite fun:

https://www.dropbox.com/s/xk2n3bm95lzm6fg/Prelude%20-%20take%201.wav?dl=0

https://www.dropbox.com/s/oo810m2bea28ceh/Prelude%20-%20take%202.wav?dl=0

https://www.dropbox.com/s/3ey6s0hfql9tg15/Prelude%20-%20take%203.wav?dl=0

Let's see who can tell what these are (looking at you CyberGene smile )...all three files were played with their respective VST/Piano etc. and no changes were made (bar from adding some built-in reverb).

I'll give it a day or two and then will publish answers..

Cheers,

J
Posted By: Harpuia

Re: Spot the VST/Piano sound - 08/14/19 09:41 PM

Number 3 is pianoteq. Can’t tell the first two smile
Posted By: magicpiano

Re: Spot the VST/Piano sound - 08/14/19 09:49 PM

To me all 3 are different presets of Pianoteq sounds... laugh
Posted By: CyberGene

Re: Spot the VST/Piano sound - 08/14/19 10:15 PM

Almost missed this thread. Not sure it’s the right piece for demos since there are only short notes while longer sustain would have been better as well as varying texture rather than almost monophonic arpeggios. And we need more octaves... smile

Anyway, to me 1 sounds best, 2 is so-so, 3 is ugly. I like how 1 sounds balanced with no exaggerated dynamics. 2 and 3 suffer from some harshness where louder notes stick out like they are artificially amplified selectively. And the general timbre of 3 is kind of boxy. Hard to say what is what from this type of demo but if I have to guess, 3 is Pianoteq.
Posted By: MacMacMac

Re: Spot the VST/Piano sound - 08/14/19 11:49 PM

The three are distinctly different ... but CG is right: the tune being played doesn't reveal much. (Was that even real music?)
Posted By: navindra

Re: Spot the VST/Piano sound - 08/15/19 03:31 AM

Resonance sounds great on #2 to me. #1 feels like a tone generator and #3 is terrible, whatever it is. I do hope #3 is PTQ, though. I wouldn't be surprised if all 3 were PTQ.
Posted By: Maconi

Re: Spot the VST/Piano sound - 08/15/19 03:47 AM

1 and 3 sound the same to me. Both sound pretty flat. 3 is just louder and might have some sustain/reverb?

2 has a lot more dynamics. Still not the best piano sound I've ever heard but better than 1 and 3 to me.
Posted By: clothearednincompo

Re: Spot the VST/Piano sound - 08/15/19 08:44 AM

#1 Sounds fine, not Casio or sample based Roland
#2 Help! My head is stuck inside a piano! Is this is a "loudness war" version?
#3 Same as #2

They all have a very similar timbre.
Posted By: QuasiUnaFantasia

Re: Spot the VST/Piano sound - 08/15/19 08:58 AM

#1: Brittle sound, and very artificial
#2: Much more realistic, but a rather hard sound
#3: Intermediate between the other two in every respect.

Personally, I would prefer #2. It would drive me to insanity, if I were forced to use #1 for longer periods of time.
Posted By: CyberGene

Re: Spot the VST/Piano sound - 08/15/19 09:38 AM

Originally Posted by CyberGene
Almost missed this thread. Not sure it’s the right piece for demos since there are only short notes while longer sustain would have been better as well as varying texture rather than almost monophonic arpeggios. And we need more octaves... smile

Anyway, to me 1 sounds best, 2 is so-so, 3 is ugly. I like how 1 sounds balanced with no exaggerated dynamics. 2 and 3 suffer from some harshness where louder notes stick out like they are artificially amplified selectively. And the general timbre of 3 is kind of boxy. Hard to say what is what from this type of demo but if I have to guess, 3 is Pianoteq.

I wrote this last night, listening to the files through my AirPods. I just re-listened to them through my HD-650 and while my comment still holds true, there's this noticeable difference in volume levels. You should have normalized all files. 2 and 3 are too loud compared to 1.

I think the 3 is even nastier through proper headphones. 1 is a slight bit too clean and dry but it seems to respond best to dynamics and the soft timbre is most to my taste. 2 is not bad in terms of timbre but there's something wrong with its dynamic response, it's too sensitive and jumps out unnecessarily loudly and that ruins it. Also I think the reverb in 2 is not very well done. Judging from the proper dynamic response in 1 and very unnatural dynamic response in 2 and 3 I would assume 1 is your digital piano's internal voice, whereas 2 and 3 are VST-s and you need to rethink your velocity touch settings.
Posted By: jamiecw

Re: Spot the VST/Piano sound - 08/15/19 10:29 AM

Originally Posted by CyberGene

I wrote this last night, listening to the files through my AirPods. I just re-listened to them through my HD-650 and while my comment still holds true, there's this noticeable difference in volume levels. You should have normalized all files. 2 and 3 are too loud compared to 1.

I think the 3 is even nastier through proper headphones. 1 is a slight bit too clean and dry but it seems to respond best to dynamics and the soft timbre is most to my taste. 2 is not bad in terms of timbre but there's something wrong with its dynamic response, it's too sensitive and jumps out unnecessarily loudly and that ruins it. Also I think the reverb in 2 is not very well done. Judging from the proper dynamic response in 1 and very unnatural dynamic response in 2 and 3 I would assume 1 is your digital piano's internal voice, whereas 2 and 3 are VST-s and you need to rethink your velocity touch settings.


Kudos CG - I don't know how you do it but you do have a knack picking these out...

1) was built-in CFX sound of P515 (played on P515) - with edited paramaters: Concert Hall/Reverb 15 - recorded to USB (and transferred to dropbox)

2) was VI Labs Ravenscroft 275 (played on SL88 Grand) - with TW Hall Warm/Reverb 20 - recorded using GarageBand DAW (no other added effects bar a little compression)

3) was Steingraeber E-272 using the Venue Reverb from the Ant. Petrof Venue preset - recorded directly to PTQ (end exported as WAV)

I have learned a little something from this, the CFX with the P515 is probably the best combo I have, in terns of sound/action/eveness ratio to the SL88 with the VSTs but I still don't understand why the P515s USB recording does not produce a decent output volume, when I play it over speakers vs hearing it back is lacklustre in comparison and just doesn't do the playing any justice whatsoever.

Quick question, is the answer to normalise the recordings and if so does that not affect the dynamics of the performance or should the gain be used instead? Other recordings of the P515 especially on YouTube are so much louder with clarity, I know there is something missing...
Posted By: QuasiUnaFantasia

Re: Spot the VST/Piano sound - 08/15/19 10:43 AM

Originally Posted by jamiecw
Quick question, is the answer to normalise the recordings and if so does that not affect the dynamics of the performance or should the gain be used instead? Other recordings of the P515 especially on YouTube are so much louder with clarity, I know there is something missing...


I don't know what specifically "gain" means here, but in normalization the dynamics will generally be increased. Imagine the sound level originally only reaches half of what is possible. Normalization will then double the sound values, so the difference between no sound at all, and maximum sound level, will double. This increase in dynamics implies a loss of sound quality, because digital sound moves in steps, and in this present scenario the steps will become twice as large. Whether this loss of quality is truly audible is far from certain.
Posted By: CyberGene

Re: Spot the VST/Piano sound - 08/15/19 10:55 AM

You should definitely use normalizing, it's a non-destructive process that doesn't change anything (well, there is rounding, but anyway) besides detecting the peak value and then amplifying the entire file so that the peak value hits a target, e.g. the max value without clipping which in the digital domain is 0dB. It doesn't change dynamics because the entire file gets equal digital gain. BTW, when exporting from GarageBand, the files are by default normalized to 0db (or maybe -1dB, that doesn't matter) which is why 2 and 3 are already normalized. Normalization is a two-step process: first you go through the entire file to detect peak velocity and thus calculate the gain and then apply the gain. Which is why digital pianos can't do that because they stay on the safe side and record relatively quiet on the USB file and don't know how loud you would play. After you stop playing, all is finished and I guess it would be a bit complicated for the piano to re-read the file from the USB, detect peak and normalize. On Kawai pianos you can apply a fixed gain to counteract the usually quiet USB volume but that's not wise because it would lead to clipping and that's already a non-recoverable distortion.

Long story short: after you record to USB, just open the file in Audacity and apply a normalization to -1dB which is a pretty standard procedure. (Don't mistake it for dynamic compression which applies different gain to different parts of the file to make it sound always loud.)
Posted By: CyberGene

Re: Spot the VST/Piano sound - 08/15/19 10:59 AM

Originally Posted by QuasiUnaFantasia
I don't know what specifically "gain" means here, but in normalization the dynamics will generally be increased. I

No, that's not true. Dynamic range is defined as the difference between the loudest and quietest. Imagine that in a file the loudest part is -30dB and the quietest is -40dB. The dynamic range is 10dB. What normalization will do is detect the peak value of -30dB, and if the target value is -1dB will apply 29dB gain to the entire file, so the new peak value will be -1dB and the new quietest value will be -11dB, so the dynamic range is again 10dB, it's not changed.
Posted By: jamiecw

Re: Spot the VST/Piano sound - 08/15/19 11:40 AM

Originally Posted by CyberGene

Long story short: after you record to USB, just open the file in Audacity and apply a normalization to -1dB which is a pretty standard procedure. (Don't mistake it for dynamic compression which applies different gain to different parts of the file to make it sound always loud.)


Thank you CG - this does make a huge difference! Just normalised a file in Audacity and this brings the performance that much closer to how I hear it when I play - examples of pre and post:

Before normalisation:
https://www.dropbox.com/s/mfxci28dpdc5nmp/BLUESINTHEATTIC-P515old.WAV?dl=0

After normalisation:
https://www.dropbox.com/s/9pl15ceqzfhq4la/BLUESINTHEATTIC-P515.wav?dl=0

I am liking the CFX even more now! :-)
Posted By: QuasiUnaFantasia

Re: Spot the VST/Piano sound - 08/15/19 11:44 AM

Originally Posted by CyberGene
No, that's not true. Dynamic range is defined as the difference between the loudest and quietest. Imagine that in a file the loudest part is -30dB and the quietest is -40dB. The dynamic range is 10dB. What normalization will do is detect the peak value of -30dB, and if the target value is -1dB will apply 29dB gain to the entire file, so the new peak value will be -1dB and the new quietest value will be -11dB, so the dynamic range is again 10dB, it's not changed.


I get your point entirely, but you're forgetting that the quietest level is zero, and when zero is multiplied by anything, it still is zero. Therefore any change in the maximum sound pressure level must necessarily alter the dynamics.
Posted By: QuasiUnaFantasia

Re: Spot the VST/Piano sound - 08/15/19 12:10 PM

Originally Posted by QuasiUnaFantasia
Originally Posted by CyberGene
No, that's not true. Dynamic range is defined as the difference between the loudest and quietest. Imagine that in a file the loudest part is -30dB and the quietest is -40dB. The dynamic range is 10dB. What normalization will do is detect the peak value of -30dB, and if the target value is -1dB will apply 29dB gain to the entire file, so the new peak value will be -1dB and the new quietest value will be -11dB, so the dynamic range is again 10dB, it's not changed.


I get your point entirely, but you're forgetting that the quietest level is zero, and when zero is multiplied by anything, it still is zero. Therefore any change in the maximum sound pressure level must necessarily alter the dynamics.


I just noticed that we're not talking about exactly the same thing. You made specific reference to the concept "dynamic range", whereas I am talking about dynamics quite generally. That may be the cause of the difference in our definitions.
Posted By: CyberGene

Re: Spot the VST/Piano sound - 08/15/19 12:44 PM

Yeah, there's some misunderstanding here. There's always a noise floor. Which is why dynamic range is defined as the difference between the loudest value and the quietest value (i.e. the noise floor) and since you also apply gain to the noise floor, the dynamic range is the same. As to dynamics... well, depends on how you define dynamics. But when you normalize you also increase the noise floor, so SNR stays the same. Remember there's no such thing as "zero". There's a level below which you can't discern between signal and noise. Even digital files have noise floor determined by bit-depth, 6dB per bit, so -96dB noise floor for 16-bit audio.
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 09:21 AM

Originally Posted by CyberGene
You should definitely use normalizing, it's a non-destructive process that doesn't change anything (well, there is rounding, but anyway) besides detecting the peak value and then amplifying the entire file so that the peak value hits a target, e.g. the max value without clipping which in the digital domain is 0dB.


This is definitely a good idea in general. In this case, where comparison of recordings is the goal, it would even be better to normalize based on a statistic of the average value in the recording, e.g. rms amplitude, instead of the most extreme value. This is more complicated because you need the determine the rms amplitude of each recording, and find a set of scaling scaling factors that equalizes rms across recordings but that at the same time do not lead to samples being larger than 0dB in any recording.
Posted By: Tyrone Slothrop

Re: Spot the VST/Piano sound - 08/16/19 09:26 AM

Originally Posted by pianogabe
Originally Posted by CyberGene
You should definitely use normalizing, it's a non-destructive process that doesn't change anything (well, there is rounding, but anyway) besides detecting the peak value and then amplifying the entire file so that the peak value hits a target, e.g. the max value without clipping which in the digital domain is 0dB.


This is definitely a good idea in general. In this case, where comparison of recordings is the goal, it would even be better to normalize based on a statistic of the average value in the recording, e.g. rms amplitude, instead of the most extreme value. This is more complicated because you need the determine the rms amplitude of each recording, and find a set of scaling scaling factors that equalizes rms across recordings but that at the same time do not lead to samples being larger than 0dB in any recording.

Are there any tools/apps that do this automatically across a given set of recordings? If not, then there might be an opportunity for someone to make an app. A slight enhancement would be to use the rms amplitude of each recording to to the same level, but where as a group, they are as high as they can go with none of the recording in the given set of recordings, clipping.
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 09:42 AM

I am not that familiar with commonly used GUI audio software. I usually work with Python, in which this is trivially easy to do. But how to make this accessible for the average user? In principle it should be possible because a simple GUI library is included in Python, and I believe you can make an installer that includes python itself (although I haven't tried this myself). But would this be the easiest route?
Posted By: Tyrone Slothrop

Re: Spot the VST/Piano sound - 08/16/19 09:48 AM

Originally Posted by pianogabe
I am not that familiar with commonly used GUI audio software. I usually work with Python, in which this is trivially easy to do. But how to make this accessible for the average user? In principle it should be possible because a simple GUI library is included in Python, and I believe you can make an installer that includes python itself (although I haven't tried this myself). But would this be the easiest route?

The easiest route would be to give it a set of files: a.mp3, b.mp3, c.mp3, ..., z.mp3, and have it create (via something like "rmsnormalize a.mp3 b.mp3 c.mp3 ... z.mp3") a new set of files, a-norm.mp3, b-norm.mp3, c-norm.mp3, ..., z-norm.mp3, which are all appropriately adjusted. No other parameters are necessary if the rms amplitude for each file is set to the same maximum value for which none of the files clip. There is only one such value so the Python program can determine that and create all the files. Why don't you make that script and we can try it out on jamiecw's original 3 files and see if it does the thing we expect?
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 12:18 PM

I am confident that I can write that in a small python script to show the principle. But there are a few caveats: 1) users would then need to have Python installed already, and ideally also the numerical computing library Numpy, 2) mp3 is a non-free format (at least it was, not sure about the status now), and its support in open source software such as Python is not ideal (would require installing additional packages). The latter can be solved by just requiring that people use PCM (WAV) format. But all in all it would require people to install a number of packages and work with a command line.

I had another idea. For my work I use Praat a lot ( http://www.praat.org ), which is scientific speech analysis/synthesis software. It is scriptable, and I already have an old script lying around that does rms equalization (including scaling to 0dB). I can have a look at that again, and modernize it. The advantage of Praat is that it is multi-platform, free, and has a GUI. For music audio users the interface and all the options may be a bit overwhelming, but for this purpose you just have to open a bunch of files and run a script.
Posted By: Tyrone Slothrop

Re: Spot the VST/Piano sound - 08/16/19 12:53 PM

Originally Posted by pianogabe
I am confident that I can write that in a small python script to show the principle. But there are a few caveats: 1) users would then need to have Python installed already, and ideally also the numerical computing library Numpy,

You should be able to do both using pyinstaller.

Originally Posted by pianogabe
2) mp3 is a non-free format (at least it was, not sure about the status now),

"MP3 technology became patent-free in the United States on 16 April 2017 when U.S. Patent 6,009,399, held by and administered by Technicolor, expired."

Quote
and its support in open source software such as Python is not ideal (would require installing additional packages). The latter can be solved by just requiring that people use PCM (WAV) format. But all in all it would require people to install a number of packages and work with a command line.

There is this thing and this also.

Quote
I had another idea. For my work I use Praat a lot ( http://www.praat.org ), which is scientific speech analysis/synthesis software. It is scriptable, and I already have an old script lying around that does rms equalization (including scaling to 0dB). I can have a look at that again, and modernize it. The advantage of Praat is that it is multi-platform, free, and has a GUI. For music audio users the interface and all the options may be a bit overwhelming, but for this purpose you just have to open a bunch of files and run a script.

Can it be packaged into a single executable like pyinstaller would do with python scripts? Because if it can, then it really doesn't matter the technology behind the executable.
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 01:12 PM

Originally Posted by Tyrone Slothrop
Can it be packaged into a single executable like pyinstaller would do with python scripts? Because if it can, then it really doesn't matter the technology behind the executable.


Not that I know of. Praat is already one executable. But there would be a separate script (text) file. So you would have to open, Praat, then the script file, the the audiofiles you want converted and then click "run".

I have no experience with pyinstaller, but see that e.g. librosa and pysoundfile are not on the compatibility list (https://github.com/pyinstaller/pyinstaller/wiki/Supported-Packages), and these are libraries that do require non-python files. But it can't hurt to try. Perhaps it works.

I will be on holidays for the next two weeks, but see if I can try pyinstaller today.
Posted By: Tyrone Slothrop

Re: Spot the VST/Piano sound - 08/16/19 01:17 PM

Originally Posted by pianogabe
Originally Posted by Tyrone Slothrop
Can it be packaged into a single executable like pyinstaller would do with python scripts? Because if it can, then it really doesn't matter the technology behind the executable.


Not that I know of. Praat is already one executable. But there would be a separate script (text) file. So you would have to open, Praat, then the script file, the the audiofiles you want converted and then click "run".

I have no experience with pyinstaller, but see that e.g. librosa and pysoundfile are not on the compatibility list (https://github.com/pyinstaller/pyinstaller/wiki/Supported-Packages), and these are libraries that do require non-python files. But it can't hurt to try. Perhaps it works.

I will be on holidays for the next two weeks, but see if I can try pyinstaller today.

OK. If you run out of time, let me know because then I'll take a crack. Could use some little project to improve my Python skills.
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 01:29 PM

pyinstaller does not work out of the box for me, if I import librosa or pysoundfile, or even just numpy. I am sure the latter should be fixable because it is supported, but alas, on Ubuntu, it does not work as is; it exits with errors. It does work if I just import the wave library (which is part of the standard library) so wav files could be read and written. However without numpy things will be cumbersome.

I will see if I get it to work in Praat. You could have a look if on your machine if you can get an executable from python using pyinstaller. Just a file with two lines for testing:

import numpy
print(numpy.__version__)

This would be enough to see if the pyinstaller can work. I am sure the actual Python code to do the equalization is easy.
Posted By: bSharp(C)yclist

Re: Spot the VST/Piano sound - 08/16/19 01:46 PM

You could also use ffmpeg I think, and perhaps a windows batch file.

https://trac.ffmpeg.org/wiki/AudioVolume

There is also ffmpeg-normalize for Python.

https://github.com/slhck/ffmpeg-normalize
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 02:26 PM

Originally Posted by bSharp(C)yclist

There is also ffmpeg-normalize for Python.
https://github.com/slhck/ffmpeg-normalize


Very interesting! This is still command line based but it seems to install fine if you follow the instructions, and is easy to use, once installed.
Posted By: Tyrone Slothrop

Re: Spot the VST/Piano sound - 08/16/19 03:11 PM

Originally Posted by pianogabe
Originally Posted by bSharp(C)yclist

There is also ffmpeg-normalize for Python.
https://github.com/slhck/ffmpeg-normalize


Very interesting! This is still command line based but it seems to install fine if you follow the instructions, and is easy to use, once installed.

So, does this mean we are about to have some friendly competition?

I'd open an Eclipse instance except that those who program Python in Eclipse have "forgotten the face of their father."
Posted By: Abdol

Re: Spot the VST/Piano sound - 08/16/19 03:23 PM

You didn't specify if you have played them individually or used the same midi file for all of them.

#3 is the best in my opinion. It's either played better or the VST has better sensitivity range. There is more expression in it and the altitudes in the dropbox depict this as well.
Posted By: pianogabe

Re: Spot the VST/Piano sound - 08/16/19 06:53 PM

Originally Posted by Tyrone Slothrop

So, does this mean we are about to have some friendly competition?

I'd open an Eclipse instance except that those who program Python in Eclipse have "forgotten the face of their father."


laugh

I will leave tomorrow morning early, so won't have time to produce something useful in the next two weeks. I haven't tried it but the ffmpeg-normalize package looks very good (and has even more advanced options than rms normalize). I guess an improvement would be to implement something that has a graphical user interface. Praat, or Audacity would be nice. I do not know if the latter has sufficiently functional scripting options.
Posted By: Maconi

Re: Spot the VST/Piano sound - 08/17/19 08:36 PM

I've always wondered why someone hasn't made a piano MIDI "Pangram" that covers just about every note played using every technique possible. Then you could just feed that MIDI into any VST of your choice and see how the VST recreates it (to see which VSTs fail at recreating various techniques and what not).
Posted By: Frédéric L

Re: Spot the VST/Piano sound - 08/17/19 09:22 PM

What do you mean by “every technique” possible ? Some VST have already a staccato vs. non staccato articulation. If you add the velocity axis with a phrasing axis (staccato, normal, legato, tenuto...) 1/ you will have more samples to record and store, and 2/ no way to communicate efficiently the second axis value from the controller to the VST.
Posted By: Maconi

Re: Spot the VST/Piano sound - 08/18/19 12:11 AM

Originally Posted by Frédéric L
What do you mean by “every technique” possible ? Some VST have already a staccato vs. non staccato articulation. If you add the velocity axis with a phrasing axis (staccato, normal, legato, tenuto...) 1/ you will have more samples to record and store, and 2/ no way to communicate efficiently the second axis value from the controller to the VST.


Yeah I was just referring to how some demos I've seen people post are almost all staccato or what not. A good "template" would have examples of all forms of articulation just to show how the VST handles short or sustained notes and various velocities.

(linking for general posterity, obviously you know what these are)

https://en.wikipedia.org/wiki/Articulation_(music)
© 2019 Piano World Piano & Digital Piano Forums