encore

Some months ago I asked Eugene about the slow attack and release problem with kX way of handling Soundfont. He investigated in it and as I recall, another way of calculating the attack/release was used. But yesterday I tried to use some soundfonts with kX (version 3434) and the attacks were still not the same as the APS (and probably the Creative) drivers. The attack of the instruments still wasn't quick enough. As you can imagine, this affects f.ex. drum-soundfonts a lot.

So, basically, I'm just wondering what's going on, on this subject? Needless to say, Eugene and the rest of the kX-crew have done a lot of fantastic work with these drivers, but there are always details.

// encore

Eugene Gavrilov

I've tried to follow the CL/E-mu parameters as close as possible
however, it is not very easy to get the formulas during the experiments...

what's the present state of the problem? could you pls give a more detailed explanation? (not just 'incorrect' or 'slow')
that is, 'how much slower?' or 'how much faster?'...
can you re-produce the bug by using Vienna?
does it happen on the preset layer? global zone? instrument layer?...

/E

encore

Okay, Eugene. I'll take a closer look on the subject when I've got the time. Might take a few days but I'll try to investigate a little bit. Unfortunately (or fortunately? ) I had so much trouble installing both the latest kX drivers and the APS-Live! drivers on the same system that I gave up and decided to run the kX-drivers on both cards - so I can't easily compare to the other drivers. However, I know what it SHOULD sound like (that's why I asked in the first place) so I'll probably find the envelope-misbehavour anyway.

Like Arnold said; I'll be back!

encore

Okay, Eugene. I've been testing some values in Vienna now and also compiled an mp3 and a soundfont for you to try out. Maybe someone with more Live-cards (and different sets of drivers) could try this soundfont out as well and see if there's any difference between how the soundfont-engine handles 'em.

I made a soundfont with 2 different samples and 4 instruments of them each, named like this :

DX7 Bass 0.001
DX7 Bass 0.01
DX7 Bass 0.1
DX7 Bass 1
Square 0.001
Square 0.01
Square 0.1
Square 1

The value after the name is the value I've used on the Attack and Release. Ex: "DX7 Bass 0.001" has the attack and release set to 0.001 s (which is the smallest possible value) The following mp3 is the result of me playing these "DX7 Bass"-instruments with the kX-drivers. On the Aps and Creative drivers, the "0.001"-value would sound more or less like the original sample in the attack. On the kX it sounds a lot more filtered. And as I measured on "Square 0.001" the attack was more something like 0.006.

As I've tried Vienna out, it's impossible to get a quicker attack than this, I also tried to edit the multiplier value of the attack (on an Melodic Preset in the Melodic Pool), but that didn't increase the attack either.

If there's more that I can do to help Eugene, please let me know. Unfortunately I cannot install the old APS-drivers or Creative drivers again, but maybe someone else on the forum can help us out to test that.

Here are the soundfont and the mp3 demonstration-file.

http://home.tiscali.se/encore/kx_envelope_test.sf2 - (33 kb)
http://home.tiscali.se/encore/kx_soundfont_testing.mp3 - (721 kb)

// encore

PS. DX7 is the name of a well-known Yamaha synthesizer from the 80's - it has nothing to do with DirectX 7 or similar. DS.

Eugene Gavrilov

thanks for the files -- I'll investigate the problem

/E

Eugene Gavrilov

so, in order to fix this issue I need some feedback from the users

[note: certain experience is required]

please download the following files:
http://kx.maincore.nl/spy_att.exe
http://kx.maincore.nl/giveio.rar (install the giveio driver. if you are experiencing problems, search this forum for details. search the net for a better distribution)

=====

test sequence:
1. install creative drivers
2. install giveio
3. run vienna and load the soundfont found in the previous posting
4. run 'spy_att.exe'

it should display the current voice state as well as the hardware register responsible for attack time

(for instance, for 0.001 ms it usually says "(xx) 0000ff7f Y", where 'xx' is voice number and 'Y' is located near the active voice
now start changing the attack time (in vienna)
write down the corresponding hardware register value

so, we will get a translation table [usually 127 elements long]

unfortunately, I cannot currently perform this test myself due to certain configuration issues

=====
the test should be performed for 10k1 and 10k2 cards separately (because the formulas might be different)

you can re-install kX drivers and repeat the test (to find any mismatches)

it might be a good idea to check =all= the soundfont parameters this way -- we are seeking for volunteers

/Eugene

p.s.
spy_att.exe window size should be at least 100x30

p.p.s.
it is recommended to use the same note and use your mouse for triggering it, not your midi keyboard

encore

Great Eugene! I hope some of the Creative-driver users are willing to help us.

Btw, I think I've might asked you this question before but I can' remember if I got it answered. Anyway, the question : How come it takes longer time to load soundfonts (the system even seem to lock up for a little while) with the kX-drivers vs the apsLive/creative drivers? Have you thought about it and considered making kX load soundfonts quicker?

Sorry again for my half-bad english.

// encore

Eugene Gavrilov

[short answer]:

we use a different method (compared to CL drivers) for memory management

[technical details]:

Microsoft OSes have an internal limitation of the amount of memory available for kernel-level drivers
there are two main 'pools': paged & non-paged
we use non-paged pool to avoid any hard-disk swapping during MIDI playback
CL drivers seem to be using paged pool instead

the M$ limitation exists both for paged & non-paged pool
when our driver asks Windows Kernel to allocated memory, there might be some
lock-up - like situations, since it takes time for the kernel to allocate the memory and
re-map pages...
so, it is not the kX driver that causes the limitations or lock-ups, but m$ OS kernel

it seems that CL drivers use a different method for allocating memory...
I currently don't have any ideas about it -- if someone is more aware about it, let me know

[very technical answer]:

you can browse msdn.microsoft.com for details
-- search for the ExAllocatePool, NonPagedPoolSize, NonPagedPoolQuota articles for details

kX Driver simply calls 'ExAllocatePool' with NonPagedPool flag...

/Eugene

encore

Eugene : Ah, thanks for the answer. Even though I don't know much about the technique behind Windows, that helped me to understand it a bit better. I was always under the impression that the "lagging" was when kX were stuffing megs of soundfonts into the memory. So the lockup is actually when allocating and remapping pages?

Again, I'm pleading to anyone in the kX-forum partially using the Creative drivers and who are experienced . Please do the test that Eugene stated above. Personally, I can't understand how people (using the kX-drivers for music purposes) have been able to continue making music without noticing the lack of attack on their soundfonts. Maybe that's because it's so obivous to me, I don't know.. Anyway, if nothing happens the next couple of weeks, I'll personally go through the hassle, uninstall the kX-drivers, uninstall one of the cards (okay, maybe that isn't necessary) and reinstall the Creative-drivers again. But I'm hoping someone else who's having a perfect Creative/kX setup would do the kX-community a big favour and do the test. Well, you can always hope.

Come to think about it, my brother uses a SB Live! with the Creative drivers.. maybe... hmmm..

HarmLammers

I would like to do the tests, but I can't install giveio. I've copied giveio.sys to c:\windows\system32\drivers, but when I try to run giveio.reg I get the message "Cannot import giveio.reg: Error accessing the registry.". I am using win XP pro and do have administrator rights. What could be the problem?

Eugene Gavrilov

>> giveio

under certain circumstances, giveio doesn't install correctly
download 'loaddrv.exe' instead (http://kx.maincore.nl/beta/loaddrv.exe)

copy giveio.sys to c:\winnt\system32\drivers (of course, use the correct folder name corresponding to your configuration)
run loaddrv
enter 'full path of driver': c:\winnt\system32\drivers
click 'install'
click 'start'
(if it fails, reboot, re-run loaddrv and click 'start')

/E

dungle

I will be happy to do the tests after exams (mid november).
I have two Sblives one running kX(of course) and the other running CL drivers.
(this is for SF compatability [loading and editing with some apps] at the moment).
My system has various bootable partitions, the ones I use for music are
win2000 and XP.
I have used SF's many times simultaneously on both cards and have noticed only very small
differences in sound rendering. Some presets perform badly on kX (crackling loops) but nearly all sound perfectly.
I have to suggest the envelope problem is not universal and may only affect certain card/OS's maybe.
best regards

HarmLammers

I have giveio.sys running now and spy_att is running.When I load kx_envelope_test.sf2 in vienna it gives "(xx)0000ff7f" for the first four channels, but when I try to change the attack in vienna, nothing changes (I've both changed it in the instrument pool and in the preset section), so what do I have to see and what do I have to write down?

Eugene Gavrilov

>> in vienna it gives "(xx)0000ff7f"

does the sound itself change?
do you see all the 64 channels on the screen?

if the sound changes, you see the 'Y' letters, but the value is still '0xff7f', then CL is using a =DIFFERENT= register for attack time... that's strange

/E

HarmLammers

I've donne the tests on a Live (CT4620) and an Audigy (SB0090). The kX drivers seem to be quite compatible with the original CT drivers, but especially the CT Audigy driver has a much better resolution. So if my results are right there must be no difference in attack between soundfonts played using kX drivers and soundfonts played using CT drivers. Or there must be a problem with another parameter...

My results:
0000ff[XX] / CT Live / CT Audigy / kX Live/Audigy
7f / 0,01 / 0,01 / -
7e / / 5,30 / -
7d / 5,71 / 5,53 / -
7c / 5,98 / 5,78 / 6,01
7b / 6,44 / 6,03 / -
7a / 6,62 / 6,30 / -
79 / 6,90 / 6,58 / -
78 / 7,09 / 6,87 / 7,00
77 / 7,64 / 7,17 / -
76 / 7,82 / 7,49 / -
75 / 8,10 / 7,82 / 8,01
74 / 8,46 / 8,17 / -
73 / 8,83 / 8,53 / -
72 / - / 8,90 / 9,00
71 / 9,2 / 9,30 / -
70 / 10,12 / 9,71 / 10,01
6f / - / 10,14 / -
6e / - / 10,59 / 11,01
6d / 11,04 / 11,06 / -
6c / - / 11,55 / 12,01
6b / 11,95 / 12,06 / -
6a / - / 12,59 / 13,00
69 / 12,88 / 13,15 / -
68 / - / 13,73 / 14,00
67 / 13,79 / 14,34 / -
66 / 14,72 / 14,97 / 15,01
65 / 15,64 / 15,64 / 16,01
64 / 16,55 / 16,33 / 17,01
63 / 17,47 / 17,05 / -
62 / 18,39 / 17,81 / 18,00
61 / 19,31 / 18,59 / 19,00
60 / 20,23 / 19,42 / 20,01
5f / 21,15 / 20,27 / 21,01
5e / 22,07 / 21,17 / 22,01
5d / 22,98 / 22,11 / 23,01
5c / 23,90 / 23,09 / 24,01
5b / - / 24,11 / 25,00
5a / 24,83 / 25,18 / 26,00
59 / 26,66 / 26,29 / 27,00
58 / 27,58 / 27,46 / 28,00
57 / 28,51 / 28,67 / 29,00
56 / 29,43 / 29,94 / 30,01
55 / 31,27 / 31,27 / 32,01
54 / 32,18 / 32,65 / 33,01
53 / 34,02 / 34,09 / 35,01
52 / 35,85 / 35,60 / 36,00
51 / 37,70 / 37,18 / 38,00
50 / 40,45 / 38,83 / 39,01
4f / 42,29 / 40,54 / 41,02
4e / 44,14 / 42,34 / 43,00
4d / 45,06 / 44,21 / 45,01
4c / 46,90 / 46,17 / 47,00
4b / 48,72 / 48,21 / 49,00
4a / 50,55 / 50,35 / 50,99
49 / 52,40 / 52,58 / 53,00
48 / 54,24 / 54,91 / 55,00
47 / 57,01 / 57,34 / 58,00
46 / 59,77 / 59,87 / 60,01
45 / 61,59 / 62,53 / 63,00
44 / 65,29 / 65,29 / 66,01
43 / 68,03 / 68,18 / 69,02
42 / 71,70 / 71,20 / 71,99
41 / 75,39 / 74,35 / 75,00
40 / 79,97 / 77,65 / 78,01
3f / 85,51 / 81,08 / 83,03
3e / 88,27 / 84,67 / 85,02
3d / 91,01 / 88,42 / 88,99
3c / 93,79 / 92,34 / 93,03
3b / 97,43 / 96,42 / 96,98

I got this result by changing the value of the attack in the instruments pool section of vienna.

Eugene Gavrilov

could you please download the soundfont found in the beginning of the thread and check it?
(perhaps, 'encore' can provide us with a test MIDI file or give more instructions on reproducing the bug)

/E

encore

At the moment I'm just one big questionmark. Could this be a problem only on my machine? Very strange.

Okay, but I'm testing by doing this; start Vienna, open the kx_envelope_test.sf2 mentioned above, extract the "dx7_bass.wav"-file into a comfortable directory. Then play the "DX7 Bass 0.001" instrument from Vienna and compare that to f.ex. Winamp or SoundForge playing the original "dx7_bass.wav". Just by trusting my ears, I can hear that the attack on the soundfont isn't 0.001 at all, because on my system (check out the mp3, also above) the attack-rate in reality is a lot slower than that. When I measured the attack in Soundforge, this also confirms my suspicions.

Btw, I haven't used the original Creative-drivers in years, just wanted to make a note on that. It's mainly the APS-drivers I'm comparing to.

Eugene, if you come up with more good test-ideas, don't hesitate to tell me.

// encore

HarmLammers

This is strange: I've done a few more tests with both the CT and the kX driver, and found out that the problem is not in the register given by spy_att. But you are true, the CT driver gives a much shorter attack (audiable) than the kX driver. So the problem must be in another part of the SoundFont implementation. Eugene: do you have an idea what else it could be?

I have the same problem as Encore; I found it very hard to translate a song from a sample based music program to a midi song using the same samples as a soundfont. The problem was that with that conversion I lost al the power of the beat... So if it could be fixed that would be great and I am willing to do some more tests to obtain result.

Harm

encore

Harnlammers : Well it's nice to hear that I'm not the only one having/noticing this problem. I also want to reflect the possibility, if kX's attack and release are not equal to Creative's, maybe there are more Soundfont-parameters that are not equal either? Maybe we should check 'em all while we're at it? Everything from LFO-speed to filters... Just to make sure.

*dreaming of using the kX-drivers for perfect soundfonting, one day.. Yummy!*

Eugene Gavrilov

ok, guys

I'm glad you have found the idea of checking all the hardware parameters interesting
(since I got practically lost in them appr. 6-8 months ago and simply gave up)

I will prepare a small list of hardware & software parameters for us to check them

stay tuned

(and, please (!) remind me to do that if you don't get the list within 1-2 weeks)

/E

HarmLammers

Eugene, if you write another testing program, could you make it possible to select the soundcard it is using? I have four soundblasters in my computer and I had to remove all but one to use spy_att. That would make testing a lot easier to do... I have another soundblaster so in that case I could install a fift one with CT drivers just for testing purposes.

Eugene Gavrilov

>> you write another testing program, could you make it possible to select the soundcard it is using?

you won't believe, but the spy_att.exe already supports multiple cards
moreover, it =does= support any register numbers (I've forgot that I had already included this code long time ago)

so, command line options:

spy_att.exe $<card_number> 0x<register_number>

that is, to test card 0 and register 14:
spy_att 0x14

to test card 2 and register 2a: [hex numbers]
spy_att $2 0x2a

cards are numbered from 0

============

moreover, I've finished a [draft] version of the SoundFont guide -- found here:
http://kx.maincore.nl/sf.rar
[Microsoft Word documented packed with WinRar]

============

everyone willing to help us to improve SoundFont compatibility now has all the necessary information to do that
1. download spy_att.exe
2. download giveio driver & loaddrv.exe
3. download sf.rar document

good luck!

/Eugene

HarmLammers

Hi Eugene,

It would be hell of a job to investigate all these parameters this way. It must be possible to do it in a much easier way. Perhaps a little program could be written that reads all these parameters and writes them to a log file everytime there's changing one. In that way one could make changes in Vienna everytime he presses a key after changing something there would be a new row in the log log file containing the changing registers. This way it takes much less time and it would also be possible to see if changing a parameter in Vienna changes more than one register at a time. I think writing such a small tool would not take too much time and makes investigating the differences in soundfont handling between kX and CT much easier and less time consuming.

Perhaps I should learn myself C++ so I could write these simple tools myself. I used to program a lot in Turbo Pascal, but I stopped programming when windows 95 came... I don't know if I could find the time ones to learn how to program in a windows environment...

Harm

JoshuaChang

waiting for a better synth sound

__________________

encore

HarmLammers : That sounds like a good idea, to automate the checking. I know the problem, I've just manually transfered a lot of sounds from one synthesizer to it's cartridge. Took ages, and wasn't fun either. I mean, computers are BEST at automatical operations, lets use it. The little time extra it takes to write a program is quickly saved anyway. And yea, making a keypress (or similar) trigger a change would be nice.

HarmLammers

Eugene: could you please let us know if you are willing to write such a tool for investigating the behaviour of the CT soundfont system versus the kX soundfont system?

encore

*waiting for Eugene's reply as well*

Eugene Gavrilov

>> waiting for Eugene's reply as well*

oops

I thought I had already posted a reply.. probably, I forgot

sorry, I cannot promise to write such a tool very soon
there's a third-party tool available -- I will try to get it (with the sources) -- cannot promise either

/E

Lex Nahumury

Just posted here to get threads notification.

/LeMury

Eugene Gavrilov

so, any volunteers to update the tool? people with VS6.0 are highly welcome!

/E