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

I'm just not seeing in their products what they bring to the table that is exceptional or different.

If they could come up with a very compact efficient solution, with no looping and with all of the other standard features fully functional - kind of a SuperNATURAL piano on the PC - then it could have some real value. As it is, it's a very mediocre offering compared to the other PC sample/player sets.

Are you sure that there are "velocity layers" in truepianos?
To me this looks like only one single sample with some kind of filter.

I recall downloading the TruePianos trial for Mac OS X a few months ago - the installer was relatively small (around 50-60mb or so), giving me the impression that the package did not use/contain samples.

Cheers,
James
x

Quote from manual:

http://www.truepianos.com/download/manual.pdf

That's because of lots of stretching and only one velocity layer.

Where there is stretching and looping, there is by very definition sampling going on. TruePianos says they use sampling in their FAQ. They also suggest that they use other techniques, whatever they are. I think they are like Yamaha, and are abusing the word "modeling" to make their products seem more exotic than they actually are.

I can only speculate, but a multi-velocity sample of a real piano is almost certainly the input to their process. After some analysis and processing it may just boil down to a filter - who knows? To me it would be more impressive if there was no stretching, and if all the other features (e.g. pedal down sympathetic resonance) worked as they should.

Exactly. DPs do this too to reduce the sample memory size. But why sample memory remains so sacred is a mystery for the ages.

If you were to play an acoustic piano, I bet you'd not see any "layers". I think what they've done is interpolation. Call it "layer morphing".


About Kawai Jame's comment that the installer was so small it may not be sample based. I'd add "maybe not time domain samples" What if they use frequency domain samples at a low 10 per second rate and from that they could generate the high rate stereo sound. In short what they store is exactly the kind of data Dewster is plotting. It is very compact and captures the tone well.

Almost certainly? Why?
And what do you mean "only speculate"? Can't you see that there are no differences between samples on all velocities in Truepianos?



What are you talking about? Of course I can see (and hear too) the differences between layers - if they are. Look again at CP1 layers:


You see? This is "interpolation".

Now compare with Truepianos:



Interpolation, morphing? Between what? They all are the same!

upd
And, by the way, here is how it should look on "acoustic piano" - VSL Vienna Imperial with ~100 layers.

As ChrisA suggests, it could also be morphing between two samples, one high velocity and the other low velocity. If it is just a filter, then some analysis most likely went into designing that filter. But either of these processes would most likely start with a multi-velocity sample set at it's origin.


There is a timber difference with increasing velocity. Here is a highly compressed sample of the TruePianos velocity layer test, where the volume increase between the notes has been removed:

http://www.mediafire.com/?hje4gfn3zzt

And here is a spectral frequency view of that MP3 file:


Clearly there is a timber difference. They obviously don't use a separate sample for each velocity as the VSL Vienna Imperial does. And they are doing it more smoothly than the CP1 does. Beyond that we can only guess at what the actual process is.

dewster

The only reason why we hear and see timbral differences is filter (EQ), I already said that!

I can't understand why are you ignoring the fact that all the samples are the same, except filtering.

You think all the piano samplers are using multi-sampling? No, Yamaha CLP230 uses only one layer, and P85 too, as far as I know.



But that's what I'm talking about - there is no even two different samples.

EQ is a very good guess, but it is still a guess.


The lowest velocity and the highest velocity could be two different phase aligned samples that they crossfade between. That method would have two different samples.

Wouldn't that kind of 'harmonic alignment' lead to a direct patent infringement? (except when they have a license of course)
http://www.samplemodeling.com/en/index.php

Adding all the attack and loop lengths together and dividing by 8 gives an average total sample time of:

9.8 + 6.5 + 4.8 + 4.3 + 4.0 + 1.8 + (1.8) + (1.8) = 34.8
2.4 + 0.9 + 1.1 + 1.2 + 0.8 + 0.3 + (0.3) + (0.3) = 7.3
(34.8 + 7.3) / 8 = 5.26 seconds

5.26 sec * 24 stretch groups * 2 ch (stereo) * 2 bytes/sample * 44100 samples/sec = 22 MB

I just downloaded the demo and the Diamond.data file is 70 MB. Plenty of room in there for even three layers of samples.

From the link:

1) "Time alignment of the phase of a set of musical sounds to be used with samplers" . Patent-pending. Filed by Giorgio Tommasini, as of September 23th 2004.

2) "Determination of modal resonances and body impulse response of a musical instrument by analysis of sounds performed with pitch changes. Application to the synthesis of vibrato & portamento with samplers". Patent-pending. Filed by Giorgio Tommasini and Stefano Lucato, as of December 20th 2004.

Attempts to patent the most obvious and straightforward approach to solving a problem are such a crock.

Both pending since 2004 - I wonder what stage of the patent process would they be at now?

No, it's not a guess - it's a logical conclusion from the obvious fact of absence of differences between samples. The only difference is in frequency view - so its just a frequency filter there.



The lowest and the highest velocity samples must have different spectral pictures.

I own Truepianos and have used it for more than 2 years for joyful play. I really love the sound signature and the expressivenss of the sound. And no, it's not just 50-60MB, that's only the application itself. After that you have to download the individual pianos, each of them is about 200-300MB. I know, it's still not near any big multi velocity software piano. I don't know how they do it, and I don't like the fact that they treat this as a big secret.

Overall I was really pleased by the sound for more than two years. The only big drawback was that pedal resonance or any other resonance is not detectable at all, even if you enable it within the software - I have the feeling as if they just claimed to have this implemented whereas there is nothing implemented in this respect.

I'm pretty sure pesk has the full version. Here are the sizes of the *.gs2 files he reports:

Amber (117 MB)
Diamond (70 MB)
Emerald (84 MB)
Sapphire (107 MB)
SapphirePed (107 MB)


I can sort of barely see some difference between the pedal up and pedal down spectral views, but I sure can't hear any difference. It's like the option checkbox isn't connected to anything.

Sorry, my mistake, you're right.