Very interesting stuff guys,
What I take from this thread is that its awesome that developers out there care so much about the quality of algorithms they implement so we can have better tools to make music with!
The aliasing thread
-
- KVRist
- 495 posts since 5 Sep, 2002 from Boston, Mass
If it sounds good it is good.
-
tony tony chopper tony tony chopper https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=3103
- KVRAF
- 3561 posts since 20 Jun, 2002
I own the cutest kittens ever (as far as cats can be owned):Hummm, where are my kittens pictures...
http://gol.gfxartist.com/
-
- KVRist
- 142 posts since 3 Jun, 2003 from Edmonton AB Canada
Urs, you need to investigate the difference between Spatial Aliasing and Temporal Aliasing. Period.
If something is good, I'll use it. I don't choose products solely on whether they alias or not. But sound quality always plays a factor in my decisions.
Saying that most people don't notice aliasing is not an excuse for complacency. "Good enough" is not a phrase I use often, especially when auditioning products.
If something is good, I'll use it. I don't choose products solely on whether they alias or not. But sound quality always plays a factor in my decisions.
Saying that most people don't notice aliasing is not an excuse for complacency. "Good enough" is not a phrase I use often, especially when auditioning products.
-
- KVRist
- 490 posts since 23 Sep, 2003 from Hungary
If someone likes the sound of non bandlimited samplers and wants to use the sound of them in a song/track as a solo instrument, thats no problem, it's a question of taste. But fitting an aliased sound in a mix lowers the dynamic range only. It strengths the amplitude of fundamentals you don't need at every part of the spectrum. So if the sound has to be the best in the mix, with the bigest dynamic range, then the best solution would be a sampler with perfect band limiting.
But i still think, that a bandlimited sampler would use too many RAM if used with big sample packages, maybe the best solution is to use seperated samples for every key.
But i still think, that a bandlimited sampler would use too many RAM if used with big sample packages, maybe the best solution is to use seperated samples for every key.
-
- KVRAF
- 3528 posts since 18 Apr, 2002 from British Columbia, Canada
indeed all this stuff is very interesting; you devs are all extrememly talented and intelligent people, and it is a gift to be able to share a forum with such minds.
something kept bothering me in this thread though, and that is a tendency to quote part of posts by William, and treat them like statements, when in context they were theoretical wonderings. I think that William had some very interesting and valid points to make, and was asking some questions that i too was very interested in hearing answered, but was replied to in such a dismissive tone that i found myself getting a bit pissed, and i am not even the target. But hey, you all (well, most.) seem to be getting along well again, so this is all water under the bridge hopefully.
And to be fair, I think that maybe wanting to see the pages taken down is a bit much, there, Will!
Certainly the research is quite valid, and the results are informative and useful in a technical sense; which i believe (
) is what it was meant for.
However it would be great to have maybe just one or two sentences to explain it just a bit to people who are a bit new to this stuff, or are not that technically savy. Perhaps a link to this thread?
And clearly the motive was not to mislead anyone (yeah, it's a big conspiracy to sell more XS-1's
), I think that much is clear to everyone now.
thanx for listening.
PS > William, looking forward to hearing the new mode on W-Station! You show yourself to be a wise person by your willingness to consider more points of view than you own; that is all too rare, these days. Good on you!
something kept bothering me in this thread though, and that is a tendency to quote part of posts by William, and treat them like statements, when in context they were theoretical wonderings. I think that William had some very interesting and valid points to make, and was asking some questions that i too was very interested in hearing answered, but was replied to in such a dismissive tone that i found myself getting a bit pissed, and i am not even the target. But hey, you all (well, most.) seem to be getting along well again, so this is all water under the bridge hopefully.
And to be fair, I think that maybe wanting to see the pages taken down is a bit much, there, Will!
Certainly the research is quite valid, and the results are informative and useful in a technical sense; which i believe (
) is what it was meant for. However it would be great to have maybe just one or two sentences to explain it just a bit to people who are a bit new to this stuff, or are not that technically savy. Perhaps a link to this thread?
And clearly the motive was not to mislead anyone (yeah, it's a big conspiracy to sell more XS-1's
), I think that much is clear to everyone now. thanx for listening.
PS > William, looking forward to hearing the new mode on W-Station! You show yourself to be a wise person by your willingness to consider more points of view than you own; that is all too rare, these days. Good on you!
-
- KVRAF
- 2988 posts since 11 Apr, 2001
The DRAFT mode in sfz uses the lousy linear interpolation (that's why it's called DRAFT). You can check test results of one-octave transposition for sfz in all modes here:Also, Rene, the Draft mode of SFZ uses which algorithm?
http://www.rgcaudio.com/sfztest.htm
CPU numbers are a bit outdated, they have over a year now.
-René
-
- KVRAF
- 2988 posts since 11 Apr, 2001
It seems to me that you're comparing a GENERATED waveform with a PLAYED waveform. That's very evident when you look at the spectrogram of the wave file you're so carefully prepared:Thank you. I don't want to start another battle. But how come playing SFZ Draft mode produces some very weird artifacts, that my own linear code won't?
Here is a picture. TOP is Wusikstation "lousy" linear-interpolation, BOTTOM is SFZ Draft.
http://www.rgcaudio.com/images/WusikstationVsSfz.gif
Please take a look at the right part of the graphic. In your 'superior' interpolation, new harmonics start to appear while we're going down in frequency. Those are of course not in the sample, right?
BTW, you're not using full samplerate bandwidth to generate those samples right? graphic looks like you're using half-samplerate.
I would please ask everyone to download the wav file WilliamK has posted. Then please listen to the three last notes of the ascending scale, first on channel left, then on channel right. That's the difference between lousy linear interpolation and badly-implemented-lousy-linear interpolation.
Can wusikstation load wave samples? I'll be pleased to accept a face-to-face challenge, with a third-party referee to check the validity of the test results if so.
-René
-
- KVRist
- 185 posts since 3 Nov, 2002
I like aliasing. Synths that don't alias, can't be trusted.
Now take it outside you two.
Now take it outside you two.
-
- KVRAF
- 11380 posts since 3 Feb, 2003 from Finland, Espoo
The software used is FL Studio. It's very quick and easy to do tests there. Try it yourself. Get demo, load any song with a lot of samples. Tune some just a few cents. Render two versions, one linear interpolation and one sinc 256.WilliamK wrote:That's actually a very good point, acurated. But may I ask you how is the diference you hear? What is the software you used for testing?bmanic wrote:Does this text belong in this thread? Well, in a way it does. I for one hear the sinc 256 vs linear interpolation in a FULL MIX if I pitch up a closed hihat or crash sample by as little as one semitone. Now add on top of that ALL the samples in a mix that I tweak a little to get the timing/length/timbre correct by tweaking a couple of cents. It does add up, quickly, and makes quite a difference. My 2 non scientific but ear VERIFIED (abx) tests.
bManic
Wk
the difference:
The end result at 256 sinc usually ends up being more smooth and tight. It's hard to describe but the mix just " geels " together much better and I hear more details also. The difference is actually quite similar to dithered vs non-dithered output (from 32 or 24bit to 16bit).
I wish I could send you my genelecs or a pair of dynaudio's to you. There you really hear the difference quite well. In a busy drum mix it's easiest to tell the difference (and the benefits) of 256 sinc vs linear. Hihats and cymbals end up being less harsh. Can't really put this into a sentence that makes any sense.
Cheers!
bManic