Just wondering what other good methods you guys have to removing DC offset on the master?

I've tried using various filters, but I'm looking to remove it without the use of a filter, one plugin i found uses 1 block of the asio block size to determind the average offset then nulls it in the next block but it doesn't work great for non-fixed DC

Voxengo 's Elephant is cool because of the choice of frequency down to 1hz and various slopes but there is no way to use it for DC removal only, the limiter is always on :(

Best tool for removing DC bias?

removing DC Offset post

Beware these offline "DC-offset detectors". Most of the time, they yield misleading results. In fact, it's more of a coincidence to ever see them show a zero.

Problem is that they all work by simply taking the geometric average of the AC signal (i.e. averaging the waveform itself). They assume that a DC free signal yields exactly 0, and the average tells how strong the DC content is. But given a finite signal, i.e. a signal that has been cut in a way or another, this average doesn't directly represent "DC that should be corrected".

Imagine a sine wave signal that has exactly the duration of one cycle (or any multiple of it). When measured, the average sums to exactly zero. Fine.

Now take this same signal, but cut it deliberately, and measure it: It now has DC. Depending on the duration of your signal, and the way you cut the start and end, it can even show a substantial amount of DC. But the sine wave hasn't been shifted vertical, we just cut the start or end of some of its cycle(s).

Technically, this really is nothing one should shift get crazy about*.
A fool proof way to detect DC is to low-pass (set to its lowest value. Optimally 0Hz but that's not possible. 1-10 Hz should do it), and measure its RMS.


*A really bad idea would be to inversely offset the signal by this aforementioned detected "error", in an attempt to "fix" it. This operation will, in the majority of the cases, add DC on its own. If you truly want to eliminate DC, hi-pass with a first order filter. DC is a 0Hz "tone", not a vertical shift.

Oh man, I can only tell you if you have DC Offset on your master then you're too late. Go back to mix stage and find your track, find source (wave, sound) which has DC Offset. You correct that on that one (or more) wave.

I primarily work in Ableton Live and I can tell you how I deal with that there. Find source, call Utility tool, check DC offset (as for remove!) and render it (record it to another new track) or as Live call it "freeze" and flatten. Simply record to a new track using Utility with DC Offset removal checked. That newly recorded waveform is now OK. Now all you have to do is replace old one with that new one. And check everything in that track or bus (group) if you had that one with a few others (for instance) playing as a drum bus or bass. Or whatever.

I am sure any DAW is equipped with same utility tool as Live is!

Yeah i actually have only started paying it more attention now, in the past I never worried as much about it unless there was a click/pop from silence.

In iZotope RX 7 the wave stats reflect DC offset in percentage, alot of files i tested would on average show like +0.125%. So it got me testing some DC removal methods to see if I could achieve 0% and if that would affect the signal in doing so.

What LP filter do you suggest to test RMS levels for that 0hz DC? Most filters only go down to 20hz and what RMS level would indicate DC?

Honestly issues with DC level, which could be problematic in practice, are rather rare. If that happens (typically result of some previous effects with asymmetric saturation, which doesn't have its own DC output filter), I would use IIR filter for the removal.
That is safest way, exactly because if the offset is changing, static compensation offset for whole clip will cause more problems than it could solve.
As always, use your ears, try to compare before and after. It's pointless to fight with issues are so low in level, it doesn't cause any practical problems.

Speaking of Elephant, there is no bypass feature for limiter section, but if you set it to clip mode with no oversampling, no gain change and ceiling at zero dBFS, the section will pass incoming audio completely unaltered (of course just when it doesn't peak over digital zero). So that way you can use Elephant just for DC filtering or dithering, if you want.

Or as someone mentioned in similar previous thread, free DMG TrackControl has also HPF, which can be set as low as 1 Hz with adjustable order.

Michal

I did try Elephant in Clip mode because i thought the same, I believe the clipper threshold is actually -0.1 even when the knob is at 0. A simple null test using some normalized sines confirmed that.

TrackControl is great!

Again this was more of an experiment to testing out the methods of DC removal, I'd rather not stick a HPF on the master but if a situation arose where I had no choice but to, its good to know the best options available!

I think, you did something wrong there, I also verified that by null test some time ago. It's bit transparent, unless you hit the threshold.
Anyway, I've maybe didn't tested it with signal so close to digital zero, my signal was something like -0.2 dBFS. I'd need to retest with level closer to 0 dBFS.

Michal

Yeah i’m pretty sure my test was correct, just a pure sine wave that was normalized to digital 0.
My meters read -0.1 after running through elephant in clip mode, and thats with the input & ceil both set at 0.00

A high-pass filter on both channels will remove voltages at 0 Hz. Picking the lowest corner-frequency available usually sounds least bad. The Mastering Massivo's 12-Hz filter is pretty good.

Also, you know, music is a-symmetrical AC about the zero-crossing, causing an acoustic polarity that's audible when inverted, so there should be an AC offset, and one therefore doesn't usually want to 'fix' an offset measured above 0 Hz. Hopefully, auto-detection is only looking at the no-Hertz potential.

Using a coupling cap on left and right analog signals out of the Ampex ATR-102's single-ended (and better-sounding) outputs eliminates their DC offsets, which are otherwise always there and are normally filtered by, both, coupling caps and the transformers, up in its penthouse. 'DC' said the penthouse cavity is good for storing a box of Q-tips - but he also said to use cc's, down below...

I've just tested it again. You're true, clip mode kicks somewhere around - 0.1 dBFS. As I've mentioned before, I tried it with -0.2 dBFS signal, which was bellow that and didn't show any change (eg. there was null up to lower bound of FFT meter).

Anyway if you have signal, which peaks higher than that, you can simply set higher threshold via input and output gain there. So for example, if you set input gain to 1 dB and conversely output gain to -1 dB, you effectively move threshold one dB higher. So even signals peaking to full digital zero won't be affected.

Michal

Airwindows DCVoltage, a free plugin.
It adds/increases/decreases/removes DC offset

IIR based high-pass filters are more effective, less killing CPU, easier, doesn't require high orders, 1st order or 2nd order (phase shift increases!) at 1-5Hz or so. Use this!
FIR based filters require toooo high IR length, less accurate but do job too, but more CPU killers, latency etc. Simpler too adjust phase modes. Experiment with this one! Or use IIR.

I think you've got it round the wrong way, i think it should be

Input gain -1db
Output gain +1db

I tested that and just for mathematics / ocd sake it doesn't completely null to -inf its a very very very low inaudible -217db haha but there is still something going on, thats with just the limiter on and nothing else, when i normalize that null test the resulting signal just sounds random single sample distortion

I had a look at this, i think its only useful for nulling out fixed DC scenarios

Yes, sorry, it should be reversed, as you've pointed out.

Yes, that's pretty much null in any practical sense.. residuum you've see, is likely due to gain manipulation back and forth with associated conversion from dBs to multiplier with some rounding. Absolutely inconsequential.
As my analyzer operates in 32 bit float (its floor is about -180 dB), it was outside of its scope.

Michal