|
Post by psionic11 on Jan 8, 2010 1:21:54 GMT
This must be some kind of promotion, because on the regular Sonart site, this same sampleset is listed at regular price. sonart.cc/shop/product_info.php?products_id=210It's a big download. Hopefully after conversion there will be at least one patch that will fit on the Fusion. EDIT --- ah, forget about it. The samples folder alone is 1.67GB, and the individual note sizes range from the smallest being ~400kb while the largest ones are at 9,400 kb. This is nice raw material, but to squeeze it down to less than 64MB for the Fusion is a tall order...
|
|
|
Post by psionic11 on Jan 8, 2010 3:11:13 GMT
Steve, or any other Fusion multisamplist out there, I have some questions regarding assembling a multisample.
Say you have multisamples spread out every major 3rd, in 5 different velocities for each note.
I'm guessing standard practice would be to make the sampled note itself the root note for the zone, then also use it for the next 3 notes up. After that, the next note would be the next sampled note, which in turn would populate its little zone until you hit the next note, correct?
For example, let's say you have sampled notes of A1, C#1, F1, A2, C#2, F2, A3, and so on. Would standard practice follow this pattern then:
zone1, based off of A1 ===> A1, A#1, B1, C1 zone2, based off of C#1 ==> C#1, D1, D#1, E1 zone3, based off of F1 ===> F1, F#1, G1, G#1
And so on.
Given that a sampled note can be stretched lower and sound decent better than it can be stretched higher and sound decent, would it be advisable perhaps to instead arrange the zones this way instead:
zone1, based off of A1 ===> G0, G#0, A1, A#1 zone2, based off of C#1==> B1, C1, C#1, D1 zone3, based off of F1 ===> D#1, E1, F1, F#1
If so, how would you implement this in the Fusion's sampler? Does it have something to do with "Split Point Shift"? I have a sneaking feeling that you use the split point to tell the Fusion where to start using the root note, and only on one "side" of the note. Meaning, a split point of -2 keys means that the A1 would render a zone filled with G1, G#1, and A1.
Am I anywhere near the ballpark?
|
|
|
Post by Shreddie on Jan 8, 2010 4:39:05 GMT
Given that a sampled note can be stretched lower and sound decent better than it can be stretched higher and sound decent, would it be advisable perhaps to instead arrange the zones this way instead: zone1, based off of A1 ===> G0, G#0, A1, A#1 zone2, based off of C#1==> B1, C1, C#1, D1 zone3, based off of F1 ===> D#1, E1, F1, F#1 That's basically what I do, I go both sides of the 'root'. Though I normally go ... -1 semitone, root, +1 semitone, ... I tend to keep it even both sides most of the time. As for the number of semitones either way, it depends on the samples and/or how big/small/detailed I want the multisample to be. Some samples don't like being transposed to far while others can go a fairly long way. I'm not too familiar with 'split point shift', isn't that for transposing all sample zones or something like that? I'm not sure I've used it... At least not for some time. How I would (and do) do it is by setting the root note of each sample then setting high and low ranges (note and velocity if there are velocity layers) for each sample... IIRC it's under the 'zone' tab.
|
|
|
Post by Shreddie on Jan 8, 2010 4:50:37 GMT
Oh... And... This is nice raw material, but to squeeze it down to less than 64MB for the Fusion is a tall order... Not really... Just dispense with some of the samples and select appropriate loop points, with additional loop levelling if needs be. Patience in doing it is all you really need. Once upon a time a whole piano program could be squeezed into less than 4mb!
|
|
|
Post by psionic11 on Jan 8, 2010 5:12:49 GMT
Aye, I took a close look at the sample organization and found that it can quickly be pared down to a manageable size. For one thing, it has 20 sampled velocities per note!! There is a 5 velocity chunk which has 170MB, so even further tightening down should indeed be possible.
|
|
|
Post by kpr on Jan 8, 2010 14:19:18 GMT
Nice venture, psionic11, and a challenge too :-) Your plan for the concept sounds not bad, but I would recommend following:
- Use a sampleplayer like Kontakt as sandbox for all your test versions. You will need test versions.
- View the entire sample content closely to get an impression of what you have.
- Consider a making of at least three versions, one of them will make it into the Fusion. Perhaps two.
- Avoid standard mapping like C1, D#1, F#1, etc. in this case. The goal is overall keyboard consistence and sometimes a C1, E1, G1, etc. sounds "better" than anything else. Also bass range allows less samples compared to the middle range. Do all this with all dynamic versions.
- Use a calculator to keep the overview about the MB weight of the final results. A little extra space is always useful, just in case if you need to insert a few samples here and there. Think small, MB wise.
- Awave and other software allows dynamics already in the editor before converting the data to Fusion format. This helps to test the versions with the sampleplayer (Kontakt, etc.) instead of moving the test data to the Fusion.
- A MB size somewhere in the middle is not the best idea, but it is a idea though. Better is one that is small as possible as well as one that is close to the Fusion's maximum RAM size. But test it with the sampleplayer to find out if the bigger MB size is worth the result.
- The Fusion's strength is the synth parameters. Use them to "design" the piano. It saves MBs too. Especially the envelopes and key tracking parameters as well as the velocity functions will make it. Invest time into this part of the venture.
Good luck :-)
Klaus
|
|
|
Post by Hollow Sun on Jan 8, 2010 22:07:52 GMT
Though I normally go ... -1 semitone, root, +1 semitone Likewise here. That way, the samples are never transposed more than +/-1 semitone. I find that C, D#, F#, A across the octaves is the ideal (reverse engineer some of my stuff to see) - that way, nothing is ever transposed more than +/-1 semi and most sounds (including vocal and other tricky ones) can handle that without noticeable artefacts. As Klaus says, you don't need quite so much detail/many samples in the bottom end. Also, experiment with sample rates - you might be surprised how little most sounds are affected at 32kHz and it can save LOTS of memory. When you've looped, be sure to get rid of everything after the loop end - it will never be heard and is wasting memory. Cheers, Steve
|
|
|
Post by psionic11 on Jan 9, 2010 2:59:20 GMT
Thanks for the suggestions. The project is going to involve pre-recorded, pre-looped samples of the Yamaha C7. There are 20 velocities and a bunch of release and other noises (which I'm going to ignore).
I'll be dealing with A, C#, and F across the octaves. I'll experiment with down-sampling to 32kHz, and will also test several versions externally before I import into the Fusion.
Since there are only 3 samples per octave, not ideal for a piano, I'm likely going to go with +1 semitone, -2 semitones per sample. To minimize this, I'm thinking of possible staggering this out with another spread for each velocity layers, probably going with 5 velocities. By stagger, I mean, trying +2 semi, -1 semi every other velocity layer. Of course, trial and error and ear are going to be the final judge.
I've only ever read about multisampling. The only imports I have now into my Fusions are soundfont conversions via Translator Free, so I've not done the real work of assigning zones and such. Nevermind actual, painstaking sampling of an instrument across octaves and tweaking loop points. Baby steps.
|
|
|
Post by Shreddie on Jan 9, 2010 8:32:55 GMT
Since there are only 3 samples per octave, not ideal for a piano, I'm likely going to go with +1 semitone, -2 semitones per sample. To minimize this, I'm thinking of possible staggering this out with another spread for each velocity layers, probably going with 5 velocities. By stagger, I mean, trying +2 semi, -1 semi every other velocity layer. Be careful staggering things like that. On some notes where the velocity layers are from different notes/samples, you may find that there is a noticable change in timbre from one velocity layer to the next... Try it by all means but personally I'd keep the same spread for the same root note for all velocity layers. Ooooh! I'll give you an insight to how I feel about it... Tedious, fun, mindnumbing, very rewarding. I do quite enjoy it sometimes, especially when the sample editing is out of the way... And I love it when I get to the actual patch creation but be warned, it'll test your patience at times!
|
|
|
Post by kpr on Jan 9, 2010 12:24:32 GMT
I've only ever read about multisampling. Does this mean the C7 venture is the first time ever that you handle seriously with huge multisamples? If so it could help to try something easier as a start, perhaps a Minimoog sound based on 12 single samples or so. Good luck anyway :-)
|
|
|
Post by markone on Jan 11, 2010 10:23:23 GMT
Good find!
I'm downloading it now, though, I'll probably end up just loading it into ESX24, and using mainstage
|
|
|
Post by psionic11 on Jan 15, 2010 21:16:30 GMT
I stand corrected -- the samples are not pre-looped. They're one-shot, some lasting as long as 27 seconds (bass notes).
So far I've auditioned all the samples by lining them up together by note, but starting with the softest velocity and ending with the lowest. This lets me see which velocity samples I can do without. I think I can get a decent piano multisample with only using 6 of the 20 velocities given.
I'm also finding that going mono and 32khz is good enough.
I'm trying out Awave for looping. Ack, yet another tool to learn. It never stops, the learning, does it?
|
|
|
Post by Shreddie on Jan 16, 2010 18:06:12 GMT
It never stops, the learning, does it? No.
|
|
|
Post by kpr on Jan 16, 2010 19:41:05 GMT
I stand corrected -- the samples are not pre-looped. They're one-shot, some lasting as long as 27 seconds (bass notes). So far I've auditioned all the samples by lining them up together by note, but starting with the softest velocity and ending with the lowest. This lets me see which velocity samples I can do without. I think I can get a decent piano multisample with only using 6 of the 20 velocities given. I'm also finding that going mono and 32khz is good enough. I'm trying out Awave for looping. Ack, yet another tool to learn. It never stops, the learning, does it? Do not use Awave for looping, it's auto-loop function needs way too long to get a result. And mostly the first result is lame. Other Awave functions are superb though and I use it a lot. You could grab the freeware EndlessWAV for great automatic loops. Doing appropriate sustain-loops for Grand Piano samples is a challenge. Depends on the sample length you can grant for your venture. If you only have to loop the portion after let's say 5 seconds, then it is quite easy. But if you need to save memory you must consider a sustain-loop at about 2-3 seconds for the mid range and about 5-6 in the bass. The problem has a name: Decay. There is no chance to get a decent sustain-loop because the fall of the volume amplitude, so you will get hard jumps in the first place. You can try to equal that with a compressor, but this is the hard way. You can try to use loop crossfade, that is the soft way. This can be used to create a kind of moving of the harmonics as exactly this happens during the original piano tone decay. But it needs a bit experience, so be patient with yourself. 6 velocity steps is still a lot considering the Fusion RAM (ok, depends on the size), and already for the sample organisation job. If you want to transfer the stuff to the Fusion, then 3-4 are ok. But they must be well choosen, we all know this when looking back to the factory Holy Grail, where the mf is a bit to far away from the f sample. But working with the filter in cooperation with the envelope you can "fake" it quite well. Indeed mono can work, although a Grand Piano is stereo in the real world. 32 kHz? Mmmm, not my favorite sample rate when it comes to Grand Piano. There are a lot of harmonics especially in the attack portion, in the harmonic structure of the lower range as well as in the most upper tones. You can always cut down the sample rate, but I would really recommend to start with 44.1 kHz. Nice to see you're going on with this project! And you can get YOUR personal superpiano, after all.
|
|
|
Post by psionic11 on Jan 16, 2010 21:56:53 GMT
I've even considered NOT looping, just implementing a fade out. Then perhaps layering the result with another Fusion waveform, to let that handle the task of the final looping duties.
As for mono/32kHz, it still sounds very decent doing an A/B comparison... there is enough harmonic activity that it still sounds pleasant and authentic. Re-creating stereo afterwards in the Fusion is the easy part.
I'll try EndlessWAV. I'd like something simple and direct, to see the waves, and its properties, and do what I need to by hand. If I do need to loop, perhaps the "bidirectional" or reverse option will work to mask the amplitude jumps...
As a last resort, I could just do much fewer velocity samples, and do different versions, to get them into the Fusion separately and later re-assemble them into custom pianos as needed. Like, hard velocity for rock piano, soft velocity for simple classical.
I will definitely keep in mind using filter tracking as an option, but that can only do so much. I already have several piano soundfonts from my project last year, and they work well for limited application, but they lack the dynamic expression of a real piano, despite achieving the necessary velocity responses in both amplitude and filter brightness...
|
|