[feature request] 64bit floating audio engine

[shadiradio]

You can't even make cuts (with no fades) and export a sample accurate result:
http://www.kvraudio.com/forum/viewtopic ... 9&t=482943

do you imply that an objective test is not possible because of that?

Oh not at all - I think this is quite specific to making cuts and probably isn't related to summing. I was just venting a little frustration about the attention to accurate audio in general (and in this case, audible).

Although, in relation to the transient issues above, I have had Bitwig render clips differently every time I bounced in the past. I noticed it when sometimes during playback some transients would sound weak, and sometimes strong (different every single time while on loop), and I was shocked to realize that it was different with each bounce as well, as I assumed this was only a playback issue. That one freaked me out, and the sample errors at cut locations (a regression from v1) freaked me out more.

[Cyoon]

ok, so you mean Bitwig doesn't export consistent results, thus it will always differ to other tools (daws) which does?

sorry, just trying to understand..is that still the case with the exports?

[shadiradio]

ok, so you mean Bitwig doesn't export consistent results, thus it will always differ to other tools (daws) which does?

sorry, just trying to understand..is that still the case with the exports?

In the case of sample cuts/splits, it will definitely differ in exports from other DAWs (I keep testing it against Reaper). I just re-did that test with 2.3.1.

I have general trust issues with Bitwig's accuracy since I've followed it and used it from day one, but they have fixed many things from those days (and also regressed in some areas like the cuts). This doesn't stop me from using it because IMO the workflow for music composition trumps the sample accuracy and trust issues I have, but I definitely don't use it for sound design / dialog / linear editing / etc.

I can't say whether or not it affects the test in this thread, but I definitely wouldn't rule it out. If it does affect this test, that means it's not a 32-bit vs 64-bit thing, but some other bug or artifacts caused by the way Bitwig renders things.

[antic604]

As you can see there is a difference when transients are hiting.
There's just 3 simple tracks with one plugin, imagine what would it be when you mix a 50 track project with 100-200 plugins inside.

By default Bitwig applies fades at the beginning and end of the clip, I've no clue if Reaper does the same. If so, are they of the same length?

[meta-redundant]

Sorry folks - and I really don't want to beat a dead horse - but in the interest of clarity and objectivity, let's briefly outline:

1) OP talks about 64 bit summing.
2) Many say there's nothing to it.
3) add'l party does a test involving 10 different clips spread across the timeline in 2 DAWs.
4) Conclusion is made that that LF content below -100dBFS in a null test is proof of something.
5) This "something" is then pegged on to Bitwig.
6) Further conclusion is made that the LF below -100dBFS is due to loss of transients (explain that one...)
6) OP concludes that this "something" is also proof that 64 bit summing it totally a real thing.

C'mon folks, isolate variables...lots of disparate ideas going on here.

Some food for thought:

--Personally, I would not have used a time-variable plugin as the test case, especially an analog (chaotic) model with possible non-linearities...something like a digital linear phase EQ is a better choice...unless you could get the analog model to print the same every time within the same DAW. But, moot point I guess.
--Pan laws: are they the same?
--Is the spectral difference an average over time, or instantaneous?
--Fun fact: value of -100 is below the audible floor, and in fact many null tests consider values of -100 or less to be "equal." Some DAWs (IIRC Sonar is one) even treat everything below -100 as infinite zero.
--We're all friends here.

[majkel]

sensej81 made a test with 'analog' off plugin, so a plugin schould work in a 'transparent' mode (only compression algorithm, but both have same settings)

[meta-redundant]

sensej81 made a test with 'analog' off plugin, so a plugin schould work in a 'transparent' mode (only compression algorithm, but both have same settings)

At the risk of being pedantic, the "analog" button you're talking about on this plugin controls which AC power supply hum you'd like (50 or 60Hz), it has nothing to do with the "analog" nature of the plugin's processing.

Again, likely a moot point but with tests of this nature it's important to be accurate.

[Enrize]

People, you're so funny. While others making hit records on their iPhones you're discussing how 32-bit engine doesn't make your sound crisp enough! Please continue

[meta-redundant]

People, you're so funny. While others making hit records on their iPhones you're discussing how 32-bit engine doesn't make your sound crisp enough! Please continue

Thanks for this

Yeah, back to music making...

[majkel]

quick test of build in samplers:
pink noise sample inserted into a build in samplers> making a comparison by voxengo span. (i forgot to change volume on reaper sampler so voxengo metering showing a differences in a gain level but thats not a game changer in a test)

bitwig pink noise sampler.jpg

reaper pink noise sampler.jpg

where a high freq on a BW sampler ??

btw. im a pretty sure thats a 'fanboys' doesnt need a high freq but a music producers and studio engineers 'sometimes' would like to work with whole audio spectrum..

[majkel]

At the risk of being pedantic, the "analog" button you're talking about on this plugin controls which AC power supply hum you'd like (50 or 60Hz), it has nothing to do with the "analog" nature of the plugin's processing.

you right, but im pretty sure those supply hum are not constant only a floating point algorithm so those could make a difference in a null test so a 'analog' was a switched off to phase cancelation test.

[JHernandez]

Tried it myself, sampler in Bitwig does do that but the tracks don't. Not that I can hear it so far.


Not that it changes anything on my end, it doesn't concerning this test, but what does this mean?

Pic related.

Screen Shot 2018-03-11 at 6.53.33 PM.png

It came up while testing so I figure, why start a new thread? Apologies.

[Hez]

I think they updated the filters in the built in devices recently, my guess is to reduce resonance range, as I found the filters broke up quite easily in v1.

Regarding all the non 64 bit summing issues brought up in this thread, these are really important to investigate! Inconsistent transients and bounces of the same clip etc. are problems that should be reported and fixed.

Guessing that the sampler high end thing is just part of the way they've voiced the sampler though, I imagine you'd see a similar thing with various hardware samplers.

[trancema]

quick test of build in samplers:
pink noise sample inserted into a build in samplers> making a comparison by voxengo span. (i forgot to change volume on reaper sampler so voxengo metering showing a differences in a gain level but thats not a game changer in a test)

bitwig pink noise sampler.jpg

reaper pink noise sampler.jpg

where a high freq on a BW sampler ??

btw. im a pretty sure thats a 'fanboys' doesnt need a high freq but a music producers and studio engineers 'sometimes' would like to work with whole audio spectrum..

Hmm...That is really interesting finding. I've checked it too on v2.3.1 and can confirm.
I also did same tests with few other samplers and DAWs, I apologize for off-topic, here are (non-scientific) results:

- Ableton's Simpler -> good, same like original
- Studio One - Sample one -> bad, similar like Bitwig
- Cubase 9.5.2 sampler -> good, same like original (sounds "cleanest" and closest to original to my ears)
- Logic EXS24 -> good, same like original
- TAL Sampler -> sample hold (default) somewhat similar to original, but not the same, other variations has different character, but that is the point of this sampler, not accuracy
- Nuance -> good, same like original

[Tj Shredder]

quick test of build in samplers:
pink noise sample inserted into a build in samplers> making a comparison by voxengo span. (i forgot to change volume on reaper sampler so voxengo metering showing a differences in a gain level but thats not a game changer in a test)

bitwig pink noise sampler.jpg

reaper pink noise sampler.jpg

where a high freq on a BW sampler ??

btw. im a pretty sure thats a 'fanboys' doesnt need a high freq but a music producers and studio engineers 'sometimes' would like to work with whole audio spectrum..

Its the sampler not the audio engine...
It shows white noise with a tad more highs btw...
At which sample rate? The Bitwig sampler is not their high light anyway, maybe a humble attempt to avoid aliasing...

The discussion is a bit funny. I know two technical reasons for going 64-bit. None has to do with sound.
1. Is sample accurate navigation for files longer than about 6 minutes.
2. Is about avoiding denormals, but I think it doesn't apply in the audio engine, more in plug-ins, and there are other ways to avoid them easily.
The first argument of majkel was a good one to go from 16-bit fix to 32-bit float. You would need a billion times a billion tracks to get an audible difference just by suming errors between 32-bit and 64-bit float...; - )
You might not like the answer, but proffessionals are caught by the same psychological ticks as all the other sheep herds. Though they could resist monster cables, I guess they are just too expensive...
I still prefer 1-blade razors over even 2-blade, not to talk about 7-blade, or do we have even more now?