DC Offset Removal - To do, or not to do ...

Special Interest Groups Audio Processing
1

Two of my soundcards (the onboard SoundMax in my desktop and my external Edirol UA-1EX) exhibit very small amounts of of DC Offset on the signal - so small as to only be really visible under ridiculously high amplification of a quiescent signal. Amazingly the Realtek in my laptop appears to be clean in this respect.

The question then: is it worth removing this very small DC offset immediately after capture - or am I better off leaving the signal un-modified in this respect?

WC

2

It depends on what else you are doing.

If you are only doing cut/paste/delete/trim type edits to a single audio track recording, then I would say it’s not worth doing, assuming the DC offset is very small.
If in this case the final destination format is 16 bit, then the best results would be to set high quality conversion dither to “none” so as to avoid unnecessary dither noise.

In other cases it is less clear cut. A very small DC offset is unlikely to have any noticeable detrimental effect on the sound quality, though I don’t see that removing the offset will do any harm unless you are working with 16-bit tracks (any type of processing on a 16-bit track will add dither noise if dither is enabled but will ameliorate quantize errors).

If the audio track is normalized to 0 dB it is possible that any amount of DC offset could cause clipping but for a small DC offset the amount of clipping will also be extremely small and is unlikely to be audible.

Removing DC offset before Noise Removal can be beneficial, though for very small amounts of DC offset the benefit seems negligible.

I’m trying to think of a case where there are significant (audible) benefits to correcting very small amounts of DC offset.
I think a reasonable test for whether or not there will be a benefit would be as follows:

  1. Find a section of silence (or near silence) in the recording.
  2. Select part of that silence (not all).
  3. Apply DC offset correction to the selection.
  4. Play all of the silence.
    If there are noticeable clicks at the transition between the offset corrected and not corrected portions then it is worth correcting (Undo and reapply to the entire track).
    If there are no noticeable clicks then it’s not worth bothering with.
3

For my LP transcriptions my workflow is

  1. Capture with Audacity set at 44.1kHz 32-bit float
  2. Export a 32-bit WAV
  3. Process the 32-bit WAV though Brian Davies’ ClickRepar (which woks at 320bit - actuall 24-bit plus 8-bit mantissa according to Brian’s manual)
  4. Import the CRed file back into Audacity and delete the original capture track
  5. Deal with the intertrack gaps
    5.1) replace the inter-track “silence” with genuine silence
    5.2) optional short fade-in to clean the track start if required
    5.3) Fade-out the track end
  6. Amplify to -3.0 dB
  7. Export multiple WAVs at 44.1 kHz 16-bit PCM stereo using triangular dithering.

Basically I try to do as little processing as possible so as to retain maximum quality information from the original LP.

Similar process for FM capture except the ClickRepair steps 2,3 and 4 are not needed unless the BBC plays an old scratchy record (which they still do from time to time) ! But sometimes i do try to remove FM hiss if I can get a good noise sample from the capture.

Thanks for your thoughts on this Steve - and thanks for the test methodology.

WC

4

“Signed 32 bit” or “32 bit float”?

I presume that you mean “which works at 32 bit” ?
24-bit plus 8-bit mantissa is “32 bit float”.


I doubt that a very small amount of DC offset will make any noticeable difference, (but it depends on what you mean by “small”).
It may be worth testing the Click Repair step to see if there is any noticeable difference if you remove DC offset before Click Repair.

5

I have the “32 bit float” encoding and “WAV (Microsoft)” Header selected in the “Specify Uncompressed Options” dialog box - is that wrong?

Since ClicKRepair is working at 24 bit float + 8 bit mantissa I’m guessing that I’ve been making the right export.

WC

6

Steve,

I tried your test on both my soundcards.

  1. With the Edirol no audible clicks even at high amplification (50 db)

  2. With the onboard SoundMax no audible clicks at normal sound levels - I need to amplify by 25 dB to hear a barely audible click at the transitions - and this is way above any amplification I use in practice.

WC

7

Yes that’s correct. As there is also a “Signed 32 bit PCM” option I thought I’d just check.

So there’s not likely to be any significant benefit in using DC offset correction (unless you’re using the on-board sound an applying very high levels of amplification).