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

This apparent velocity discontinuity may not be real - the difference in timbre between successive notes may be due to the modelling of the hammer hitting strings that are already vibrating. It would be interesting to repeat the test with a longer gap between notes, to ensure that the notes are COMPLETELY deadened. If the release duration (i.e the rate at which the dampers damp the strings) can be adjusted, it might be prudent to set it as fast as possible for this particular test.

Greg.

Nothing in terms of spectral views - the notes each look fairly random phase-wise. But the obvious velocity discontinuities (I would call this "layer switching" if it were a sampler) are rather strange for what is considered & sold as a 100% modeler. You can easily hear them as timbre variations, and easily see them in the waveform (amplitude / time - oscilloscope) view in Audition.

Ooooooops

1000 apologies.

Here is a stereo file:

http://www.box.net/shared/1r7xiqd8s3

AKG K271 Studio.

Julias O. Smith III discusses what you are talking about as a realistic effect that should perhaps be eliminated in modeled pianos. Start the string over rather than add/subtract energy and create unexpected results.

I'm certainly not trying to harp on the V-Piano, just trying to figure out what is going on in there. Admittedly, the DPBSD is more geared to revealing the limitations of samplers...

Here's what I'm seeing in Audition:

You can see, particularly in the lower (corresponding to the stereo right channel) waveform (envelope actually, this zoomed out) the strange velocity notes that look kind of squashed, while the others look pretty much like they are uniformly increasing. I'm think I hear timbre group changes there too, but maybe it's the gin talking.


So slow the MIDI tempo down and send me a sample already. ;-)

Ahh, much better! I'll review it tomorrow, thanks!

What I'd like to see, ideally, is for the tempo to stay the same, but a distinct gap between each note. Also, I personally can't send you a V-Piano recording, at least not until you send me a V-Piano.

Greg.

Subjectively, I don't find anything in the performance of the V-Piano that suggests velocity switching although I don't hear velocity switching on my Yamaha (which has the same main voice as the P155 I believe). It may be due to the low level of the recording I made but also I was surprised to see that the dynamic range was not greater than some other pianos. Subjectively, the difference between the quietest key strike and loudest is immense.

Lawrence, I used V1 Impactance I think. I turned all effects/reverb etc off but did not even look at what the other stuff such as string resonance, damper resonance etc was doing.

I'll do another recording at a higher level using Vintage 1. Yesterday I did a factory reset so I won't touch any other setting other than to turn reverb off.

I think Greg's point is interesting. If you rock a glass of water and set it washing from side to side you can almost stop it if you introduce another movement out of sync with the washing of the water at the right time. Maybe striking the strings whilst they are vibrating could induce a similar phenomenon, creating a similar graphic effect to velocity switching. Can you tell I'm not a physicist?!!

I'll also reduce the tempo to 60bpm and upload that too to see if there is still evidence of what might be velocity switching.

Unless I can increase the digital output I'll have to record from analogue outs.

Cheers All,

Steve

Steve,
If you play repeated notes on a real piano, with the sustain pedal down, you can hear the effect very easily - each note sounds a bit different. Yes, I think your glass of water analogy is right on. It is my understanding that digital pianos and synths commonly emulate this (very well, IMHO) by simply overlapping "voices", such that as you play each note, the sound of the previous note is allowed to continue to sound. The summation of the voices creates a subtle "phasing" effect which changes every time you play a note. It is plainly obvious to me that this is occuring in two software samplers that I have - I can watch the voice count keep increasing as I play the SAME note repeatedly with the sustain pedal. (even Pianoteq, which is modelled, increases it's voice count a bit, so it may even be doing it the same way) Without the sustain pedal, you would have to play repeated notes very "legato" in order for this to happen, of course.

I am by no means certain that this effect is occurring in your V-Piano recording - it's just a possibility I am raising.

Greg.

Check the Utility / Output Gain setting. Adjusting that may increase the digital output as well.

I hate looping, but this was pretty well done, better than the P-155 IMO (but any looping not done down near the noise floor is a deal-killer for me). Nice blending of velocity layers, except at the high end.

Techniques and implementation almost identical to what Yamaha does in their DPs, which is very interesting.

I wonder why more DPs don't at least try to simulate sympathetic resonance? Too subtle and they figure few people will miss it? Perhaps, as it is often implemented a delay effect, it interacts poorly with the (generally rather poorly done) reverb?

-------------
- Kawai MP5 -
-------------
FILE & SETUP:
- dp_bsd_v1.3_kawai_mp5.mp3
PROS:
- Nice long note decay.
- Looping isn't too obnoxious.
- Large dynamic range (~42dB, vel=1:127).
- This is a very smoothly blended multi-velocity layer sample set.
- Responds to partial pedaling.
CONS:
- Looped, but pretty well done.
- Samples are 3 to 1 seconds (lo to hi) with 1 to 0.5 seconds of crossfade.
- Obviously stretched, group transistions farily audible.
- Stretch distances: 5,3(x8),2,4,3(x17),2 = 29 groups.
- Velocity layer switch fairly audible (timbre change) @ vel=104.
- No key-up or pedal samples.
- No obvious sympathetic resonance.
OTHER:
- Volume in MP3 file very good, though there are HF (PC soundcard?) tones near the noise floor.
- Date reviewed: 2010-02-09

I'm not sure whether Audition provides a way to script this, but I thought it might be useful for the layering test, to normalize the volume of each note. Sort of like "manning the volume knob", but done automatically. That way, there's only one variable (timbre) to consider, rather than two (timbre and volume.)

Here's an example of what I mean:

http://voltara.org/pw/yamaha_ydp223_gp1_velocity_normalized.mp3

I manually trimmed out the layering test in Audacity, exported to WAV, then ran a shell script (Linux user here) that used the "sox" utility to: split the WAV file into 60 pieces, one for each note (tempo 120 = 0.5 seconds each); adjust the volume of each WAV file to 90% peak; then paste all the pieces back together.

Listening to the result, it sounds to me like either a 2-layer sample with a broad crossfade in the middle, or maybe a filtered 1-layer sample.

I would be curious to know whether the spectral phase graph of the post-processed file reveals any more detail than the original.

Andy

I figured it out - I had a small amount of feedback through the digital I/O in the Echo MIA mixer. Cans sound fine now.

I do it by applying one of several custom envelopes I've developed in Audition to the signal. It amounts to pretty much the same thing, but is actually a bit better for diagnostic purposes because it is a smooth envelope which can reveal small amplitude changes between layers. And you're right, it makes it easier to listen for timbre changes if the volume isn't constantly changing. It also makes the intensity of the spectral views more constant.


Yes, that is what it sounds like to me too. I guess I'm taking Yamaha at their word with the 3-layer thing, as the clues as to where the layers blend are too vague to see or hear. I don't see reference to the layers in the manual, but Sweetwater states "3-layered piano samples with DSS (Dynamic Stereo Sampling)".

Most DPs that have obvious layer switching tend to devote more of the layers to the upper end of the velocity, where the strings and soundboard display more non-linearity I suppose and get much brighter. Thus my comment No real variation in timbre at higher velocities - what you and I are saying boils down to the same thing. I am saying it in the context of other DPs, but but you are describing the effect more directly.

I looked at and listened to your file, but it didn't reveal anything I haven't already seen and heard in Audition. I do appreciate you going to all that effort though. And I'm very glad to see others doing some analysis on these files, and am hoping for more debate and perspectives on what goes on in DPs by studying the sound samples.

http://www.mediafire.com/?sharekey=23fab7ed7f98c959d9d5c56d04dfa8b09482e6c03f68ebb40ac99885da44e881

Dewster,

I think the above should be a link to my shared MediaFire account.

The folder now contains three files...I deleted the original V-Piano file. I increased the output gain as per Lawrence's instructions to +12db. It is not as loud as I would've liked but I think it is significantly better.

All files recorded direct to CD via coaxial digital out from the V-Piano using Vintage 1, the default grand piano. Reverb (or Ambience as Roland terms it) is off.

1. V-Piano BS Detector at 120 bpm
2. V-Piano BS Detector at 60 bpm
3. Just the velocity section of your midi file played at 20 bpm

Cheers and thanks again,

Steve

Yes, the level is much better, thanks so much for going the extra mile!


OK, the velocity variations for the repeated notes in the layer detection test are different in all of the separate files. Which is really interesting.

I think Roland is (pick one):
a) not aware of it
b) trying to model reality
c) ignoring or content with an emergent behavior from their model

I think a combination of b & c is the most likely answer.

Too bad many hate the midrange, the V-Piano is head and shoulders above any DP sampler I've seen/heard. 3 second (max!) samples + loops suck.


------------------
- Roland V-Piano -
------------------
FILE & SETUP:
- dp_bsd_v1.3_roland_v-piano_vintage1.mp3
- S/PDIF to CD recorder.
- Factory settings except gain +12dB & reverb off.
PROS:
- Beautiful long note decay.
- No looping (modeled).
- No stretching (modeled).
- Sympathetic resonance, though very subtle.
- Responds to partial pedaling.
- Good dynamic range (~36dB, vel=1:127).
- Nice "loom of strings" pedal down sample, pedal up sample more of a knock.
CONS:
- Key-up sample not audible.
OTHER:
- Quickly repeated notes have somewhat random velocities (modeled reality?).
- Date reviewed: 2010-02-09

I've only listened to the 20bpm velocity reocrding, and I cannot hear anything resembling a velocity "layer" switch - it sounds extremely smooth. I listened both to the original recording as-is, as well as a shortened version where I stripped out a lot of the sustain from each note, to make it easier to detect changes in timbre by simply listening to the recording without performing any analysis.

Greg.

Hi Dewster,

Thanks for taking the time to analyse the Kawai MP-5

Just for your info, I set Damper Resonance, String Resonance and Key Off Effect, all to the midway setting: 5/10. Not sure if I should have set them all on max?

The HF noise you heard/saw, seems to be a result of using USB out, rather than the standard DIN MIDI connectors.

Usually I use MIDI cables into an Edirol UR-80 audio/midi interface, but in this instance, I used the usb out from the DP, straight into a new, fairly well specced PC.

I noticed the noise, too, but when I un-plugged the usb connection, and used the din connectors, the noise disappeared.

Normally I guess we are all more interested in the default, just to know what it will sound like "out of the box" as we assume the factory set it that way in the first place for a reason. But I'm very interested in whether these effects exist and how they sound, so if you feel like doing it again with them turned up, I'll definitely listen to it and archive the second sample with that noted.

Funny, our P-120 has a ton of digital bleed into the audio when we use the Yamaha "to-host" connection (glorified RS-232 with proprietary connector) - so much that we can't use the resulting audio in a solo recording. The MIDI connection is much quieter though.

I had an Edirol UA-25, kind of wish I hadn't sold it.

Hi Dewster,

Here is the link to the updated Kawai MP-5 file (take 2)

This time recorded with standard MIDI cables.

Damper Resonance, String Resonance, and Key Off Effect, all set to the max.

http://www.box.net/shared/29quyz7h2h