An Idea for Programing Samples in Giga Sampler

[Kenn159]

Have you ever noticed when playing, say giga piano, that when you are playing a song and the song plays the same notes repeatedly in sequence ,that you can hear the same exact note sample over and over with out variation . The only way that note[lets say c-4] will not sound like the same sample everytime is if you play that note with mutiple velocities.
So if your song reqires that note to be played with the same stick force [velocity] then your ear really perceives that this is the same sample playing over and over again. The Idea of a sampler with multivelocity capability in therory would make ,say, lets use a snare drum, sound more real ,asuming every time you hit the key or electronic drum pad, that you would hit it with a different enough velocity to trigger different samples . But once again if your hitting it with the same amount of stick force your hearing that same static sample over and over again , and it defenatly sounds like a sample not a real snare.
So my Idea is ,a instrument sample would sound more reilistic if , along with having multivelocity samples , to have multiple samples for each velocity zone maybe 3 or so .And a way that giga sampler could randomly go between those 3 samples in that same velocity zone. So if a piano had 6 velocity zones , then that would be 18 samples per note , 3 notes per velocity zone . [I wish I had a dollar for everytime Ive said velocity here] not to mention extra notes for damping and resonance etc.

[Kenn159]

Im not sure how to edit my ad , so im listing underneath it that I meant strike force not stick force

[Bob Campbell]

that, Kenn is a CLASS A suggestion.

very, very, good idea. you should email that to Nemesys, or even better, patent the idea. I thought the same thing about 10 years ago using an Akai, but never considered there would be a system that could actually use it one day.

Bob

[Georg]

Very good Idea!!

Georg

[Joris Vincken]

If Nemesys hasn\'t already read this, please email it to them. This is one of those ID\'s that would make GS a lot more musical that it already is.

[SCARBEE]

This summer 2000 I (SCARBEE) will release a Bass-CD-ROM for Giga-sampler using a very interresting programming teqnique, which will solve this problem and others too. However, I cannot tell about it yet, as you can not patent these ideas - I guess. But expect a demo withing a month on my homepage. The homepage is not up yet, as I have to finish the CD first.

Thomas

[Jan]

Hi Kenn159,

I totally agree with you when you say that with sampling we only get to hear a limited set of variations of an instrument over and over again, and that the ear registrates this. For instance, I cannot stand to play the piano I have in my workstation for a long time due to ‘ear fatigue’, something that does not happen when I play an acoustic piano. That is why Gigasampler is such a relief: I am much less troubled by this ‘ear fatigue’, although it still is not a replacement for the real thing. I think that your idea of horizontal sample alignment is a very nice idea, but not very practical. For instance, suppose that you would build a piano according to this principle of horizontal sample alignment, it would probably occupy a space of 30+ Gbytes, for only one instrument. With the current technology it is not possible to use instruments this large. When it comes to snare drums, another option is available: simply map the alternative set of velocity switched samples of the same snare drum to another key. This way you can have one set of samples under, another set of variations under G and so forth. Of course, this is not true of sampling for instance a flute. But again this has the drawback of creating huge files for just one instrument. With Gigasampler and your theory we are approaching the limits of what one can achieve with sampling technology. It is simply impossible to record all the variations of an instrument in one file, because the number of variations are infinite. What is needed here is a model that describes accurately how an instrument will behave, with the samples as foundation material. One can use the samples as basic material and allow a behavioural DSP-algorithm to modify the sound. Drawback of this method is that –from what I have heard- it is a herculean task to construct an algorithm that accurately reproduces the desired sound, but this mariage of sampling and acoustic modeling is the way to go for better results in the future. Still, I like your idea and I would definitely post it to Nemesys or put it in the wish list.

Jan

[Kenn159]

Thanks for all the input on this subject. In responding to Jan , I dont think that we need a sample for every possible variation in attack and nuance that a instrument can produce , that would be mathematically impossible . But theres a big differance between every possible variation and hereing the same as in one sample over evertime you hit that note with the same amount of velocity. For instance if your playing a sampled instrument and that instrument has 3 velocity zones per note .On note c-4 you have one sample to be triggered from 0-40 and another from 41-80 and another one from 81-127 velocity, all on the same c-4 note.And lets say your playing your average song that 90 percent[in some cases 100 percent] of the velocity used to play the notes trigger samples in the 41-80 zone area .Then the end result to the ear will be that you are playing a sampled electronic instrument. because every note on every key will trigger the exact same sample every time you play it with in that velocity. So a few horizonal samples for every velocity range with a way to randomly switch between them would prevent hearing the same repetitious sample over and over again within a given velocity. As far as memory goes I think giga stdio will have the amount of memory to achieve this , the specs on giga studio are up to 4.3or something gigs per sample . If you took giga piano at 1.2 gigs , if it had this capabilty it would at most be 3 times the size , thats 3.6 gigs .And you can easily put that on a DVD disk with a 5.1 gig capability or have it on multiple cd\'s .

[elle]

Would anybody care to test the idea on Michiel Post\'s 3 gig Pristine piano?

[tonylombardi]

I think that what we want more than anything else here is more velocity splits.
Ken\'s argument is based on the fact that, since currently most instruments only have a few velocity switches, we are triggering many of the same samples in a row when we strike a key multiple times.
However, this can be solved by simply creating more velocity splits. After all, if we\'re going to put more samples into RAM, we might as well assign them a velocity range.
For example, if we had 20 different velocity samples, I think it would be foolish to split these into 5 velocity ranges, and have each range randomly trigger 4 samples. If they\'re taking up RAM, and have different sound levels, why not just create a 20-level velocity range? You\'re much less likely to trigger the same sample twice in a row with so many velocity levels.
When sequencing, since it would be very cumbersome to make sure every note has a slightly different velocity, one would probably need to use some sort of \"humanizing\" program that would go through the song and randomly adjust the velocities by a certain amount to avoid triggering the same samples twice in a row. I know for Cakewalk there are CAL programs that humanize rhythm in a similar way.
Ken\'s idea should make a program more musical if there aren\'t many velocity splits. But if there are many, I don\'t think randomizing samples is necessary.

Anthony

[elle]

This sounds a bit like a power approach. It\'s time - as Joris mentioned - to get more intelligent algorythms into the game. What about \"Dynamic Sample Allocation\", where samples, normaly spread over the 0-128 velocity range, are autonmaticaly allocated to the - predicted - velocity range/zone, all with respect towards their output volume.
I think these algorythms should be linked (sold) as an \"playing style\" or velocity parameter element of the sample library.

[mitchb2]

This has been done on a small scale.
For instance, in the AO upgrade there are some instruments that have 2 different takes for each 16th note string.

By alternating the notes (or randomizing with a simple Cakewalk CAL), you can get less of that \"machine gun\" feel.

However, that\'s just a tiny step in the direction. We absolutely need the ability to embed different takes of the same note and same velocity to get the realism.

[Gulliver]

I think most of what has been discussed here could also apply toward sequencing rather than strictly sampling. It\'s not that hard or complicated to record a large amount of variations on indivisual notes but it is a challenge to apply all of this toward a more realistic performance. Sequencers, like samplers, have come a long way but mostly in terms of additional number of tracks, efx plugins, timing and syncing, etc.. Little has been improved in the performance front. One of the major bottlenecks is the ancient and antique MIDI standard (but that\'s another story) One can apply all kinds of different tricks with MIDI to mimick certain playing methods (read: Style Enhancer plugin for Cakewalk)
but that\'s all they are, tricks... I like the suggestion \'Jan\' made regarding smart \'DSP\' techniques to subtly change samples and considering the rate at which dsp research is going, I can imagine something like this coming into play within the next 3 or 4 years, easy... The question is going to be what will happen on the sequencing front?
If sampling is going to get more sophisticated there will have to be a similar improvement in sequencers to take advantage.

Regards,
Gulliver.

[Dave Of NemeSys]

We are aware of this idea of sequential triggering of multiple variations of samples in the same velocity zones. It won\'t hurt to email wishlist@nemesysmusic.com to show your interest in this idea.
Take care
Dave

[elle]

If adding \'intelligence\' is the next step, then the man-machine interface is certainly an area. Each library could have its own playing style algorythms that adds intelligence on when and how samples are triggered and \'massaged\'.

[SCARBEE]

[We absolutely need the ability to embed different takes of the same note and same velocity to get the realism.[/B][/QUOTE]

You can do that: ex: sample 1, velocity 100-127 - attentuation 4, sample 2, velocity 89-99 - attentuation 0.

This way sample 2 will play as loud as sample one at it\'s highest velocity (99

You just have to play around with the attentuation to get the right feel.

Thomas