There`s alotta talking shit about how FLS sux when dealing with audio. Just to prove it`s bollocks i made a test mixing down 3 stereofiles at in different hosts - samplitude(on friend`s PC), logic 5 and FLS Demo(consider buying it) to 16 bit mix(just to eliminate the bit-depth resolution since logic 5 can bounce max. to 24-bit). The difference between FL mix and Samplitude\Logic mixes(zero difference)was very small(about -90 dB at peaks) though existed. So i took an 32bit-fl.point wav.-file and inserted into Audiotrack in FL and rendered to a 32-fl.point file and again, the difference between the original and the Fl-rendered file existed and looked like peaks in the place of original file`s peaks. in samplitude opening a 32bit file and rendering it to a 32bit-f.p.-file and comparing 2 files showed no difference.
So guys - I mean developers of my best non-human-being-friend FLS - that`s not a deadly sin, but to me it`s important to know, what it`s caused by? of course, all of faders were set at 0, none of FX were used, i just tried to test summing algorithm.
One more question - what`s the resolution of pan knob and faders values in FLS - is it also in 32-f.p. or it`s always fixed - like 8 or 16 bit deep?
the last blabla - it would be nice to make max. MIDI resolution 960 PPQ.
FLS audio handling aint clear......
-
- KVRAF
- 6937 posts since 4 Jun, 2004 from Utrecht, Holland
A difference of -90dB in 16bits audio stream indicates a difference in the very least significant bit. Probably a rounding error that comes from internal calculation. 16bit streams are summed to an internal format (32-bits?) and then converted back to 16bits. There also may be some dithering applied in this process which could be the source of difference. Not to worry about it...
The difference you see in peaks probably shows what the various hosts do when the resulting mix clips, since you mix 3 tracks at a zero level. If the sum in float format goes above 1 it would be clipped when represented in 16bits. Various hosts could deal different with this, its about the design of the mixing algorithm. Hosts are allowed to handle it slightly different.
BTW: I'm not sure weather the tests you do actually address problems specific to mixing audio streams as opposed to mixing VSTi's. I'm not sure there really is an issue other than people cant find how to RECORD an audio track in FL.
The difference you see in peaks probably shows what the various hosts do when the resulting mix clips, since you mix 3 tracks at a zero level. If the sum in float format goes above 1 it would be clipped when represented in 16bits. Various hosts could deal different with this, its about the design of the mixing algorithm. Hosts are allowed to handle it slightly different.
BTW: I'm not sure weather the tests you do actually address problems specific to mixing audio streams as opposed to mixing VSTi's. I'm not sure there really is an issue other than people cant find how to RECORD an audio track in FL.
-
- KVRer
- Topic Starter
- 18 posts since 18 Oct, 2004
yes it`s clear but i didn`t MIX the file with another 1 in the second test, i just took a 32bit fp stereofile with no peaks exceeding 0 dB and rendered it to 32bit-f.p.-file. then subtraction - oops, there`s a difference WTF?! In samplitude rendered file the same test(substraction from the original) shows absolute 0 silence. audiotrack in FL must have not passed any FX and not dealt with velocity or filter like data in sampler.
So the difference must have occured in the panorama \ master bus fader calculation. I guess there`s no fft filter at any stage of such pathway in my test, like it`s in Reason where impulse signal shows in the rendered file a slight cut off starting from 18 kHz.
So the difference must have occured in the panorama \ master bus fader calculation. I guess there`s no fft filter at any stage of such pathway in my test, like it`s in Reason where impulse signal shows in the rendered file a slight cut off starting from 18 kHz.
-
- KVRAF
- 6937 posts since 4 Jun, 2004 from Utrecht, Holland
Ok, I didnt understand that correctly. Thought you were saying you mixed three files. Now I see you didnt mix them together but did the test with three files consecutive.
Probably FL does something to 32bits floating stuff, like oversample to 192kHz and resample back to the appropriate freq, or maybe even first convert it to 16bits
Yes, there should be no difference, but would you notice it in a fully completed mix?
I would conclude to only use 16bits audio files if you want FL not to mess too much with it
Probably FL does something to 32bits floating stuff, like oversample to 192kHz and resample back to the appropriate freq, or maybe even first convert it to 16bits
Yes, there should be no difference, but would you notice it in a fully completed mix? I would conclude to only use 16bits audio files if you want FL not to mess too much with it
-
- KVRist
- 41 posts since 19 Jun, 2003
Here is a question. If you rewire fruity into another host, say Samplitude, will it be rendering through Samps audio engine, and if so wouldn't that make it sound exactly the same as if it was rendering by itself in Samplitude, or am I wrong?
-
- KVRer
- Topic Starter
- 18 posts since 18 Oct, 2004
Shiiit. I Made 2 tests---
1. Mix 3 files to 1 16bit mix in 3 hosts.
2. Put 1 stereofile 32bit-f.p. as Audiotrack in FL, then render it back at 0 db faders to a 32bit-f.p. file.
1. Mix 3 files to 1 16bit mix in 3 hosts.
2. Put 1 stereofile 32bit-f.p. as Audiotrack in FL, then render it back at 0 db faders to a 32bit-f.p. file.
-
- KVRer
- Topic Starter
- 18 posts since 18 Oct, 2004
I guess it^s gotta be an issue for developers, huh?
I`d like to get answer from JMC, gol, Mirabebe or someone else from IL. I dunno if it`s a problem when calculating 1 file, but when mixing 20 tracks this artefacting might be accumulating? it depends on the stage which this change in signal takes place on. If signal quantized to fixed point format first with some calculation rounding \ error, then by summing 20 tracks it can get an serious issue?
I`d like to get answer from JMC, gol, Mirabebe or someone else from IL. I dunno if it`s a problem when calculating 1 file, but when mixing 20 tracks this artefacting might be accumulating? it depends on the stage which this change in signal takes place on. If signal quantized to fixed point format first with some calculation rounding \ error, then by summing 20 tracks it can get an serious issue?
-
tony tony chopper tony tony chopper https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=3103
- KVRAF
- 3561 posts since 20 Jun, 2002
-pick up a free compiler (I think borland gives older versions of delphi, there are loads of free C++, or I don't know give a try at visual basic)
-save a wave file as raw data so you can easily load it
-mix some of them by yourself in a 32bit floats buffer. It's a simple addition. You'll then understand that there's no magic, nothing done frequency-wise, nothing 'analog'.
There's no point in making tests, either you already think that 'audio summing' can be done differently in digital, or not, but your choice is probably already made. Nothing can convince you, except your own programming experience. Until then, keep thinking that sequencers have hidden compressors, different summing qualities, etc.
-save a wave file as raw data so you can easily load it
-mix some of them by yourself in a 32bit floats buffer. It's a simple addition. You'll then understand that there's no magic, nothing done frequency-wise, nothing 'analog'.
There's no point in making tests, either you already think that 'audio summing' can be done differently in digital, or not, but your choice is probably already made. Nothing can convince you, except your own programming experience. Until then, keep thinking that sequencers have hidden compressors, different summing qualities, etc.
-
- KVRer
- Topic Starter
- 18 posts since 18 Oct, 2004
gol thank you for da reply- but ima not diss FLS (shit - dats real bad mothafucka for beat making and my say virtual lover)))), and im not that clever at borlands an delphis to make some tests on code.
and im convinced - there^s no any compressors filters whatever in the signal path. who thinks there is,--- it^`s his own problem.
I just wanted to get an clear answer about the reason of non-0-mixing of the original 32-fp file and the rendered of 32-fp file.
yesterday i made 1 more test.
1/Take a stereo 16 bit file - limited to heavy -9,05 db RMS(but no peaks exceeding or reaching 0 dB) - in sound forge convert to 32 bit ieee fp file.
2/insert it as audiotrack in FLS without routing it to any fx - i.e. straight to the master bus.
3/render the audiotrack to 32(0.24)fp file with all faders set at 0db.
4/subtract the original file from the rendered file in Sound Forge(Wavelab).
5/check results - the rest is not silence - it looks like 1-bit deep noise(amplitude - 90,3 db peaks) of an 16-bit file but very similar with the test signal in spectrum(just like if the test signal was quantized to 1 bit file and then reduced in amplitude by 90 db).
As i know now, this is not accumulated when mixing many tracks - the rest of the substraction remains the same in amplitude and looks again like 1-bit version of the mix.
Though the audiotrack created from an 24-bit file (rendered to 0.24 32fp bit file and then truncated in Sound Forge to 24 bit) after substraction from the original file creates absolute silence. The same goes to the substraction of a samplitude mix from a FLS mixes created of 24-bit files - absolute 0, silence.
Hence I can make a hypothese, that it can be caused by following things:
1/ Result of the fader\panorama multiplication and the following rounding.
2/ As stated in the manual - FLS can read files of max. 24 bit resolution - so when i put a 32fp file into audiotrack or sampler - it`s converted\requantized somehow and than rendered back to 32fp file.
3/ bit distribution in the 0.24 32fp format and in the sound forge\samplitude (ieee format) is different(perhaps, ratio of mantissa \ fraction bit-depth in a bit-word is different).
to developers - 4 more close getting down to the nitty-gritty u might knock on my ICQ 254389470
PS - only 4 developers! unless u r a real sexy sista
and im convinced - there^s no any compressors filters whatever in the signal path. who thinks there is,--- it^`s his own problem.
I just wanted to get an clear answer about the reason of non-0-mixing of the original 32-fp file and the rendered of 32-fp file.
yesterday i made 1 more test.
1/Take a stereo 16 bit file - limited to heavy -9,05 db RMS(but no peaks exceeding or reaching 0 dB) - in sound forge convert to 32 bit ieee fp file.
2/insert it as audiotrack in FLS without routing it to any fx - i.e. straight to the master bus.
3/render the audiotrack to 32(0.24)fp file with all faders set at 0db.
4/subtract the original file from the rendered file in Sound Forge(Wavelab).
5/check results - the rest is not silence - it looks like 1-bit deep noise(amplitude - 90,3 db peaks) of an 16-bit file but very similar with the test signal in spectrum(just like if the test signal was quantized to 1 bit file and then reduced in amplitude by 90 db).
As i know now, this is not accumulated when mixing many tracks - the rest of the substraction remains the same in amplitude and looks again like 1-bit version of the mix.
Though the audiotrack created from an 24-bit file (rendered to 0.24 32fp bit file and then truncated in Sound Forge to 24 bit) after substraction from the original file creates absolute silence. The same goes to the substraction of a samplitude mix from a FLS mixes created of 24-bit files - absolute 0, silence.
Hence I can make a hypothese, that it can be caused by following things:
1/ Result of the fader\panorama multiplication and the following rounding.
2/ As stated in the manual - FLS can read files of max. 24 bit resolution - so when i put a 32fp file into audiotrack or sampler - it`s converted\requantized somehow and than rendered back to 32fp file.
3/ bit distribution in the 0.24 32fp format and in the sound forge\samplitude (ieee format) is different(perhaps, ratio of mantissa \ fraction bit-depth in a bit-word is different).
to developers - 4 more close getting down to the nitty-gritty u might knock on my ICQ 254389470
PS - only 4 developers! unless u r a real sexy sista