Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards The DPBSD Project!

Yeah I like the .mp3 files too, to do a subjective compare.

I also like the pictures, but don't really know what they mean, so some of your recent markups of the pictures has been interesting.

I've downloaded a number of items from mediafire, and have especially listened to the samples from different DPs now that I have a decent pair of headphones.

I've found the "readme" file, and the pictures posted here (with all the "squiggly lines" + comments + cursor locations), to be most instructive.

I hope that this will positively impact future DP R&D so that dewster, one day in the year 2020 (if not 2019), will finally upgrade to a new DP. But with coming inflation, and the collapse of the US economy, things look bleak.

I love them! I don't need the pictures, but for a simple reason: I also use audacity to check for several issues on the file, so I see the same pictures "live" on my computer within audacity! So I can better zoom into some of the examples and see what exactly is going on there, especially in the damper resonance part. I now understand exactly how you do analyse those files, and can easily reproduce it or even analyse new files by myself. It's really not rocket science if you know a little bit about theory of sound.

I would suggest to even more expand the midi file to test for even more real acoustic behavior and if the DPs reproduce them correctly. I have some additional tests in mind that I also perform on my own DP and I'm therefore very curious if other pianos do perform as well as CA63 in these tests or maybe even better.

--> String resonance: Press a key silently and see what happens with string resonance, what keys are producing ringing sounds on the silent undampered key etc., so pressing the appropriate keys that should produce harmonics.

--> High notes and string resonance: Check the very high notes which are not dampered at all if they also produce string resonance when depressed. This should not be the case.

--> String resonance and damper resonance: If you press a key "silently" with damper down and then press other keys that should resonate with this silent key and then set damper up, check if the key does still resonate.

It would also be very nice if the test would not only cover those technical things, there is no way I can judge on the sound quality that way. At the beginning of the test there should be a short real playing sample (~1min) which covers a subset of real live situations: Playing pp to ff, with damper up and damper down, some chords, some arpeggios from low to high, some staccato, some legato, that's it pretty much.

I agree, it's basically pattern recognition, a bit of sound theory, and some knowledge of how DPs and real pianos work. And it's really not that complicated.

I'm glad someone else is doing the analysis too, and with a different software tool. Keep me honest kawaian!


Those seems like they might be so subtle it would be hard to hear them. I do appreciate the suggestion though.


Test #2 in DPBSD v1.5 is something like that:
2. C2 & C3 & C4 & C6 & C7 @ vel=1 (& G6 @ vel=15); C5 @ vel=127; keys up (& G6 @ vel=15)

Another test might be to find out where the transition between the dampered and undampered notes is.


Right, I think everyone (myself included) would really like to hear what it sounds like doing something more musical. I've looked around for something like that but haven't found anything yet. Do you have any suggestions?

Did I say audacity? I mean Audition, sorry, audacity does not have those analysis capabilities. So we actually use the same software for analysis, but still it's quite obvious and reproducable how this works.

Oh noes, we're both using the same "unproven" software!!

I find the analysis pictures helpful, and the audio file is only helpful to me if it's available in the context of a few test songs, and not just walking the notes up and down the range.

Yeah, I know, Adobe is like a little rat-shop...

A rag-tag bunch of pikers probably running things out of some garage somewhere. Their 2009 fiscal revenue was only $2.946 billion USD - no one should take them or their software seriously. It's not like anyone has even heard of them, much less use their products.

I would really love to have DPBSD MP3s of:

- Korg SP-250
- Yamaha P-85

Could anyone perhaps furnish them?

Yamaha CLP-990 Review

Nan kindly provided us with a DPBSD MIDI file of the Yamaha CLP-990. This is a 2002 vintage DP with a five layer velocity sample and no stretching, which makes it a rather advanced and interesting specimen for its time.

In their literature Yamaha even says how much memory they allocate to the piano voice, 80.4MB, something they tend shy away from these days. Doing the math, this works out to about 1 second of attack + loop sample time on average for each note layer. The average attack + loop times I measured are on the order of 1.7 seconds, so perhaps they aren't using 16 bit samples, or are employing additional compression such as delta encoding or similar.

As you might expect, the samples and loops are too short not to hear obvious artifacting, though the decay is fairly long. The velocity layers aren't blended, but the steps aren't too objectionable. It responds to partial pedalling, but the effect is fairly subtle. There are no extra sounds like key up or pedal up/down, though it passes both sympathetic resonance tests, as well as the pedal down silent replay test.


Spectral phase view of note C3, cursor at end of attack sample | beginning of loop sample.


Spectral frequency view of velocity layer test, all four layer switches are visible here.


Spectral phase view of stretch test, all notes appear and sound separately sampled. Mid keys shown here, lows and highs are similar.

More analysis pictures, the MP3 file, and the text review file are available at the share point.

------------------
- Yamaha CLP-990 -
------------------
FILE & SETUP:
- dp_bsd_v1.5_yamaha_clp-990.mp3
- Floppy drive (MIDI file), headphone output.
- Recorded by "nan".
PROS:
- Passes the pedal down sympathetic resonance test.
- Passes the key down sympathetic resonance test.
- Passes the pedal down silent replay test.
- Responds to partial pedaling.
- Fairly long decays on the order of 3/4 Pianoteq or longer.
- This is an unblended 5 velocity layer sample set.
- No visible or audible stretching, all notes sampled.
- Large dynamic range (~58dB, vel=1:127).
CONS:
- No obvious pedal up/down samples.
- No obvious key up samples.
- Obviously looped, both visually and audibly.
- Sample & loop lengths are rather short.
- Sample lengths are (C2:C9): 2.2,1.9,1.6,1.6,1.5,0.9,0.8,? seconds.
- Loop lengths are (C2:C9): 0.83,0.6,0.56,0.5,0.49,0.4,0.4,? seconds.
- Visible & audible layer switch @ vel=24,36,44,52
OTHER:
- Something like a small amount of reverb in the test file.
- Notes played @ vel=1 produce no sound.
- MP3 levels: peak @ -12dB, noise floor @ -76dB.
- Date reviewed: 2010-03-14

Korg SV1 Triple Voice Review

jve has kindly supplied us with (count 'em) three DPBSD MP3s of the new Korg SV1 - thanks jve!

The voices are:
"Grand Piano 1" - a Japanese grand
"Grand Piano 2" - a German grand
"Electric Grand" - a CP80

I fully reviewed all three, and I made a bunch of pix of the Grand Piano 1, but only a few of the other two. Very long attack and loop samples for a DP that sound pretty good - it's really too bad about the excessive stretching and some abrupt and very audible velocity switches.

I've never played a CP80, but jve owns a CP70 and says the SV1 sounds as though it goes into the final decay stage too early, resulting in a too percussive sound. I thought the initial decay of the grand piano patches sounded a bit steep too, with a swelling rebound later. From the spectral views the CP80 voice could likely be a mono sample enhanced with a stereo simulator.

The key and pedal noises are presented here at the default level and sound to me a bit too loud, though the levels of these sounds are adjustable.

As usual, more analysis pix, the MP3s, and all reviews can be found at the share point.

Grand Piano 1

Spectral pan view of the note C4. Cursor at the sample attack | loop boundary. Sample lengths are 3.8 | 3.0 for attack | loop which is better than usual for a DP.


Spectral phase view of the entire looping test compressed 20:1. Attack and loop samples are clearly visible here for the lower notes (redish sections are the noise floor).


Spectral phase view of the stretch test mid notes. Except for the very highest notes the lower and higher sections are similar. Stretching is audible in the lows and mids.


Spectral frequency view of the entire layer test. The audible and abrupt velocity switch at vel=88 is very clear here.

--------------------------
- Korg SV1 Grand Piano 1 -
--------------------------
FILE & SETUP:
- dp_bsd_v1.5_korg_sv1_grand_piano_1.mp3
- Yamaha N12 digital mixer, Adobe Audition.
- Recorded by "jve".
PROS:
- Pedal down sound is a nicely done "loom of strings".
- Pedal up sound is "stringy clunk".
- Key up "clunk" sound.
- Passes the pedal down sympathetic resonance test.
- Responds to partial pedaling.
- No strange pedal sounds during partial pedaling.
- Nice long decays on the order of Pianoteq.
- Decent dynamic range (~34dB, vel=1:127).
- This is an unblended 4 velocity layer sample set.
CONS:
- Fails the key down sympathetic resonance test.
- Fails the pedal down silent replay test @ pedal up.
- Looped, though very well done with long attack and loop samples.
- Attack sample lengths are (C2:C9): 5.0,3.8,3.8,3.4,3.0,2.5,?,? seconds.
- Loop sample lengths are (C2:C9): 3.8,3.5,3.0,3.0,2.4,1.9,?,? seconds.
- Obviously stretched, low and mid group transistions farily audible.
- Stretch distances: 3,4,4,3,3,2,3(x16),2,1(x19) = 42 groups.
- Visible and audible velocity layer switch @ vel= 44,88,116
- Layer switch @ vel=88 is fairly abrupt.
OTHER:
- Notes played @ vel=1 produce no sound.
- MP3 levels: peak @ -1dB, noise floor @ -75dB.
- From the SV1 manual: "This is a very accurate sampling of a Japanese grand piano, suitable for pop and jazz."
- Date reviewed: 2010-03-15


Grand Piano 2

Spectral frequency view of the entire layer test. All three velocity switches are visible here, including the fairly abrupt sounding switch at vel=116.

--------------------------
- Korg SV1 Grand Piano 2 -
--------------------------
FILE & SETUP:
- dp_bsd_v1.5_korg_sv1_grand_piano_2.mp3
- Yamaha N12 digital mixer, Adobe Audition.
- Recorded by "jve".
PROS:
- Pedal down sound is a nicely done "loom of strings".
- String damp sound is realistically "buzzy".
- Key up "clunk" sound.
- Passes the pedal down sympathetic resonance test.
- Responds to partial pedaling.
- No strange pedal sounds during partial pedaling.
- Nice long decays on the order of Pianoteq.
- Decent dynamic range (~34dB, vel=1:127).
- This is an unblended 4 velocity layer sample set.
CONS:
- Fails the key down sympathetic resonance test.
- Fails the pedal down silent replay test @ pedal up.
- Looped, though very well done with long attack and loop samples.
- Attack sample lengths are (C2:C9): 6.9(?),3.8,4.3,?,?,?,?,? seconds.
- Loop sample lengths are (C2:C9): 3.6,3.1,2.6,?,?,?,?,? seconds.
- Obviously stretched, low and mid group transistions farily audible.
- Stretch distances: 5,4,3,3,3,3,2,2,3(x13),1(x24) = 45 groups.
- Visible and audible velocity layer switch @ vel=44,88,116
- Layer switch @ vel=116 is fairly abrupt.
OTHER:
- Notes played @ vel=1 produce no sound.
- MP3 levels: peak @ -1dB, noise floor @ -80dB.
- From the SV1 manual: "This is a very accurate sampling of a famous German grand piano, perfect for jazz and classical music.
- Date reviewed: 2010-03-15

Electric Piano

Spectral pan view of the entire layer test. The single velocity switch is obvious here.


Spectral frequency view of the entire layer test. The two layers are nicely blended, and the switch isn't visible here.

---------------------------
- Korg SV1 Electric Grand -
---------------------------
FILE & SETUP:
- dp_bsd_v1.5_korg_sv1_electric_grand.mp3
- Yamaha N12 digital mixer, Adobe Audition.
- Recorded by "jve".
PROS:
- Pedal down sound is a nicely done "loom of strings".
- Pedal up sound is a "thump".
- Key up "clunk" sound.
- Passes the pedal down sympathetic resonance test.
- Responds to partial pedaling.
- No strange pedal sounds during partial pedaling.
- Decent dynamic range (~28dB, vel=1:127).
- I believe this is a blended 2 velocity layer sample set.
CONS:
- Fails the key down sympathetic resonance test.
- Fails the pedal down silent replay test @ pedal up.
- Rather short decays on the order of ~1/2 Pianoteq.
- Appears to be a mono sample enhanced with stereo simulator?
- Most likely looped, though very well done with long attack and loop samples.
- Obviously stretched, most group transistions farily audible, even highest ones.
- Stretch distances: 8,8,3,3,3,4,4,3,4(x6),2,6,4,5,11 = 19 groups (waveform display).
- Visible and audible velocity layer switch @ vel=54
OTHER:
- Notes played @ vel=1 produce no sound.
- MP3 levels: peak @ -2dB, noise floor @ -76dB.
- From the SV1 manual: "Released in the mid-’70s, this was the very first portable piano really close to the real thing."
- Date reviewed: 2010-03-15

I beg to differ:
These should *always* resonate, wether the key or pedal is depressed or not.
And the resonance sound should continue when all keys and the pedal are released.

I believe this is a typical component of real piano sound, it adds a lot of realism to the sound, because there wont be total silence, when the last key is released and it would sound more lively.

Peter



Peter

DPBSD MIDI File v1.6 Released

I just updated the DPBSD MIDI file to version 1.6. An updated readme file reflects the changes as well.

This adds some finesse to the key down sympathetic resonance test by raising the sympathetic keys one by one.

It also adds a brief pedal partial damping test, which our P-120 passes.

====================
= Revision History =
====================
v1.6 - 2010-03-16:
- Added brief pedal partial damping test.
- Added progressive C7, C6, C4, C3, C2 key lift to key down sympathetic resonance test.
- Reversed the order of the pedal down sympathetic resonance test stages (to put loudest first).
- Added a short G6 to signal the end of each test.
- Added C2 @ vel=100 to peak recording level (test 0).

dewster, thank you for posting those interesting findings from the SV1 test.

I wonder why the acoustic piano sounds are not blended, while the electric piano apparently is.

Also, would you like to hazard a guess as to why vel=1 produces no sound?

Cheers,
James
x

My theory is a CP80 is a much simpler instrument, sound-wise, than a grand piano (guitar-like electric pickup) so blending two velocity layers with a filter is a lot easier. I believe the sample is mono, which makes the blending even easier. It must require some kind of spectral morphing when one blends the stereo layers of a real piano sample.


Seeing as how note-on vel=0 has been historically co-opted for note-off, if I were building a DP, I'd probably design the keyboard controller to generate note-on vel=1 for playing notes silently. That would differentiate the two scenarios, and allow the sound generator to behave more realistically like an acoustic piano given various scenarios.

It's really too bad MIDI note-off isn't used because the "velocity" could be used as an indication of damping timing / strength, further adding to realism.

dewster, thank you for the explanation.

So to clarify, is it particularly unusual for the SV-1 to not produce any sound at v=1, or is this standard for most/all DPs?

Cheers,
James
x

I can only detect this in DPBSD v1.5 and higher (when the key down sympathetic resonance test was implemented) and since its introduction I've reviewed 5 DPs. Kind of hard to get a good statistical feel for how usual / unusual this is with that small number, but here are the lists so far:

Silent MIDI note-on @ vel=1:
- Korg SV1 (Grand Piano 1 & 2, Electric Grand)
- Yamaha S90 XS Natural Grand S6
- Yamaha CLP-990

Play faint note with MIDI note-on @ vel=1:
- Galaxy Vintage D
- Yamaha AvantGrand N3

How about pianoteq?

James
x

The v3.5.3 demo plays faint notes with vel=1.