Mixing a 16 bit file to 32 bits gives wrong result (FIXED)

[Art Evans]

As discussed on the KvR forum, if I put a 16 bit file on track one and the same file on track two, and render to 32 bit float, I get a result which is demonstrably not the doubling of the sample values - it's very slighly off. Doing the same test in Audition gives a demonstrably correct result. Why should that be?


FIXED (2.0)

[LOSER]

Originally Posted by Art Evans

As discussed on the KvR forum, if I put a 16 bit file on track one and the same file on track two, and render to 32 bit float, I get a result which is demonstrably not the doubling of the sample values - it's very slighly off. Doing the same test in Audition gives a demonstrably correct result. Why should that be?

How slightly is it off?
Also is Auditions rendering engine 32bit float or 64bit float (like REAPERS)?

[schwa]

art can you post the project with a little audio snippet?

[Dstruct]

Originally Posted by LOSER

How slightly is it off?
Also is Auditions rendering engine 32bit float or 64bit float (like REAPERS)?

audition = 32 bit

so yeah - it sounds like simple rounding errors/differences of both engines ...


nothing to worry about i think!?

[Art Evans]

Well, maybe not worth actually worrying about, but in my simple terms, if I'm adding two equal samples (say 5000 and 5000 sample values) I should get 10,000 and rounding error does not come into it. I'm raising it more as a matter of curiosity but sometime querying something like this reveals something else of more importance. Or not!

[Dstruct]

what do you mean with "it's very slighly off"?

[Art Evans]

Original sample value (24 bit range) = 2793984
2793984 x 2 = 5587968
2793984 added to itself in Reaper = 5588138.5
2793984 added to itself in Audition = 5587968

Conditions - 16 bit original file being mixed with itself and saved as 32 bit float.

[LOSER]

Originally Posted by Art Evans

Original sample value (24 bit range) = 2793984
2793984 x 2 = 5587968
2793984 added to itself in Reaper = 5588138.5
2793984 added to itself in Audition = 5587968

Conditions - 16 bit original file being mixed with itself and saved as 32 bit float.

Hmm... that seems worrysome. But have you "Reduce denormalization from plug-ins" enabled? (Prefs->Plug-ins, not that you didn't know that ). Since this adds a little DC-offset/Nyquist noise IIRC, it could be the source of that difference.

[Art Evans]

Aha. I do have that enabled. Would that account also for the fact that the problem I describe does not happen if the original file is 32 bit?

Gotta go to bed right now, will retest tomorrow.

[LOSER]

Originally Posted by Art Evans

Aha. I do have that enabled. Would that account also for the fact that the problem I describe does not happen if the original file is 32 bit?

Gotta go to bed right now, will retest tomorrow.

Yes, no matter what source format, the noise will always be mixed to the signal (just a low -300db(?) noise IIRC, to prevent the said denormals).

[Dstruct]

maybe justin should make this a bit more intelligent (so that this noise just gets added on channels where plugins are actually running/enabled)!?

on the other side i don't care. always had this setting disabled here

[Billoon]

Originally Posted by LOSER

Yes, no matter what source format, the noise will always be mixed to the signal (just a low -300db(?) noise IIRC, to prevent the said denormals).

That would probably account for it...

But its still occuring with "reduce denormal" unticked....so does that mean that "reduce denormal" is not switching off or its something else?

My suspicion is its not turning off...

With "reduce denormals" unticked...insert an empty item on a track and 2 js/vol plugs both set to +150db...hit play and there is (inaudible) signal present.

Ticking "reduce denormal" and trying the same test produces the exact same result.

Also curious is that the denormal reduction only seems to be affecting the positive side of the rendered wave, negative is spot on in my tests.

[Billoon]

Just tested with v1.24 (which has no denormal reduction and no signal detected at -300db) and its still happening....when summing integer files and rendering to float...positive values in float are out, negative-correct.

So it cant be the denormal reduction causing it.



\Still would be good if the off switch was fixed for the denorm. red. though.

[airon]

Maybe that's the reason I'm getting the wrong readings from files I KNOW don't cross a certain threshold, yet Reaper tells me they do.

In my case I was playing a 16-bit 48kHz WAV mix that's limited to -9.0dB. Measurements by Soundforge confirmed the peaks were at -9.0dB, yet I get -8.9 and -8.7 readings in Reaper on those mixes, and that's the peak meter.

Could this be related?

[Bulls Hit]

Originally Posted by airon

Maybe that's the reason I'm getting the wrong readings from files I KNOW don't cross a certain threshold, yet Reaper tells me they do.

In my case I was playing a 16-bit 48kHz WAV mix that's limited to -9.0dB. Measurements by Soundforge confirmed the peaks were at -9.0dB, yet I get -8.9 and -8.7 readings in Reaper on those mixes, and that's the peak meter.

Could this be related?

I don't think Reaper's meters are particularly accurate. I regularly see them hit the reds on 2 track mixes that have been rendered with a brickwall limiter on the master

[danbee]

Originally Posted by Bulls Hit

I don't think Reaper's meters are particularly accurate. I regularly see them hit the reds on 2 track mixes that have been rendered with a brickwall limiter on the master

Was the limiter set to 0db ceiling? If this is the case it's likely that inter-sample peaks caused the clipping, in which case Reapers meters are probably more accurate than most.

[Bulls Hit]

Originally Posted by danbee

Was the limiter set to 0db ceiling? If this is the case it's likely that inter-sample peaks caused the clipping, in which case Reapers meters are probably more accurate than most.

Nope, -0.1

[Art Evans]

As far as I'm concerned the original issue raised here is fixed in Reaper 2.0 beta 3 - many thanks.

[danbee]

Originally Posted by Bulls Hit

Nope, -0.1

You're probable still getting intersample peaks clipping. Try setting it down to -0.3db at least. If you're still clipping bring it down to -0.5db and so on until it doesn't clip.