On Windows 7, Audacity v2.3.2:
I frequently encounter audio files with some DC offset.
I want to remove it.
All the posted advice I’ve seen says use the Remove DC Offset feature of Normalize.
But when I do that, there is no change in visible DC offset.
Same when I undo it.
I know, you’ll say upgrade to 2.4.2, and of course I’ll do that eventually.
However, all the advice claiming it works was posted long before 2.4.2, also before Win 8 and Win 10.
How are you measuring that?
Where does the audio come from?
Note that a sound may be asymmetric without any DC offset. There’s no reason why positive going peaks and negative going peaks have to be the same size.
It’s not unusual for home microphones to have very low pitch sound errors and noise which can look like DC offset depending on where in the show you measure. Those errors can throw off effects, filters and corrections which is why getting rid of very low pitch trash is the first step in Audiobook processing.
Koz
filters and corrections which is why getting rid of very low pitch trash is the first step in Audiobook processing.
And, since DC is “zero Hz” a high-pass filter will remove any true DC offset. If you have deep bass that you want to retain use a 20Hz high-pass.
Note that you can’t hear zero-Hz but you might hear a “click” at the beginning & end when it kicks-in and kicks-out. Aa high-pass filter may not remove the click at the beginning so you might have to mute the glitch after filtering.
A 20Hz high-pass filter may “help” with an asymmetrical too but it won’t change the sound. An asymmetrical waveform can be normal and it can sound normal or it could be an indication of distortion.
You won’t often see asymmetry in commercial music releases because they are highly processed, plus they are a mix of sounds and most asymmetry would be buried in the mix.
Some “cheap” soundcards may [u]clip[/u] asymmetrically when over-driven.
Thank you all for your quick and caring responses.
Unfortunately, your suggestions are not working for me.
I will answer better tomorrow. Please forgive my slowness!
Of all the pieces I aired this morning in Truth and Justice Radio, this one appears to have DC offset:
Like several others I’ve encountered, trying Normalize, with Remove DC Offset checked, has no effect. For this one, I went ahead and tried High Pass Filter for 0, 20, and 40Hz, 40db attenuation, each with no effect. I understand why home mikes, or passive recorder mikes, can do this while in use. But this was NOT a home mike situation.
In past broadcast pieces that I didn’t set aside for reexamination, I’ve encountered at least one with very clean, heavily compressed audio, in which the entire left side is centered but there’s a point in the middle after which the entire right side is cleanly elevated. Trying Normalize, as above, had no effect. Because the excursions are separated, I believe that failure can’t be attributed to low-frequency noise.
I would like to see some audio with DC Offset that is successfully removed by any of the above procedures. I have yet to find any.
(If you want to know more about Truth and Justice Radio, go to truthandjusticeradio.org.)
Why do you say that? There is no DC Offset visible in that screenshot.
I see a non-symmetrical voice. Many announcers have that. I used to be able to recognize which of two favorite announcers was speaking because one had pronounced biased blue waves and the other didn’t. Neither had DC offset.
This is what happens when you edit a DC Offset presentation to a clean one.
You can’t fix that later. You have to “cure” the offset in broken files before you edit with them.
Koz
THANK YOU Steve and Koz for those answers. But …
(1) The file behind my pic is of two DIFFERENT speakers, each speaking 1 to several minutes. Could both of their voices be equally nonsymmetrical, both hairy at the bottom?
(2) Are you sure Remove DC Offset works ONLY before any editing? It doesn’t make much sense to me. But if so, that would probably explain why I’m getting no effect. However, (a) where is that in the FAQ or manual? and (b) why is it buried in the Normalize operation?
I will try to fish up the original just so you can see why I ran Compress twice (after which I noticed the hairy bottom).
Are you sure Remove DC Offset works ONLY before any editing?
It’s only effective before editing. Remove DC doesn’t go down the show syllable by syllable adjusting the waves. It takes the whole show’s waves gently up or down until that little space on the left between sound silence and “0” timeline goes away.
This, badly illustrated, is the offset. That downward positioning of the waves. Silence is supposed to settle at 0.00 on the timeline. The stuff on the left doesn’t.
The sound on the right never had any space there, so it’s normal.
None of this has anything to do with the tips of the waves. The quality of the voice or instrument determines that, and no, they don’t have to match at all. If you start worrying about the fuzzy quality of the blue waveforms, it’s going to be a very long day.
This is where we find out if you exported a perfect quality, WAV (Microsoft) 16-bit copy of all your work before you started to mess with it. Once you close Audacity, you lose UNDO and we can’t take most effects, filters, and corrections out. If you really damage the sound with no backups, you may be recording it all again.
Koz
The flat top to that waveform suggests that the recording has been pretty heavily processed, making it hard to guess why it looks like it does.
DC Offset is when the average of all samples is greater or less than zero (silence). With a “normal” waveform, the sum of all samples in the positive side of the centre line will balance the sum of all samples on the negative side.
Here’s an example of an asymmetric waveform that has zero DC offset:
The effect will “work”, but once you have started editing and processing, the DC offset may well have created numerous clicks which are then difficult to get rid of. DC offset, if present, should be dealt with as the first thing before any other editing or processing. Best of all is to not have DC offset in the first place, but we don’t always have that option 
THANK YOU Steve for your continued followup on this.
OK, I get what you’re saying. Recorded audio doesn’t necessarily center, or appear to center, at 0. So the extra hair at the bottom should not be viewed as abnormal, nor a sign of DC offset. And finally, so my original complaint is not valid.
I do remain curious why Remove DC Offset is offered as an option for Normalize.
Just to be complete, I found the original, yet-to-be-processed-by-me mp3, and here it is, not zoomed in:
I’m sure you can see why I used Compress. The guest’s voice was much much much louder than the host’s.
(He was using riverside.fm for the first time.)
Now, I can’t help but notice that the positive peaks are bigger than the negative peaks.
Is it possible that voice recordings typically have that phenomenon?
Could it be that peaks in voice recording are that way because that’s the way a human larynx works?
(I have to confess that I wasn’t paying sufficient attention during my first 16 years on air!)
Stan
Haha - that has been fiercely disputed for at least a decade 
Personally I think that DC Offset should be a separate effect in its own right.
Natural sound sources are rarely “exactly” symmetrical, though most usually appear fairly symmetric. Brass instruments are often noticeably asymmetric.
Some microphones and some sound cards tend to produce asymmetric waveforms.
Just as a topic of interest but not much practical value:
If a sound is asymmetric but has no DC offset, it is due to the phase relationship of the frequencies that make up the sound. Applying an “all-pass filter” may “correct” the symmetry (or make it worse). An “all-pass filter” allows all frequencies to pass without boosting or cutting, but varies the phase according to frequency.
If you would like to try playing with an all-pass filter, try running this code in the Nyquist Prompt effect (see: Nyquist Prompt - Audacity Manual)
;version 4
(allpass2 *track* 100 2)
The numbers “100” and “2” are arbitrary. Different numbers will shift the phase differently.
Before you ask - It’s probably not possible to predict what numbers are required to make a waveform symmetrical.
I do remain curious why Remove DC Offset is offered as an option for Normalize.
I assumed it was because you could get more gain with equal positive & negative peaks. Of course that assumes the waveform is otherwise has symmetrical peaks.
I really appreciate you guys’ caring help.
Today, finally, without meaning to, I actually RECORDED a short piece of hum, which had DC offset. I employed Remove DC Offset and IT WORKED!