Noise Reduction/Removal

Hmmm. Yes… if I just split the track… :confused: How would I go about doing that?

BUT if you couldn’t hear the hum… and I also had someone else listen to one of the chapters where the hum is definitely there (I can see it on the sound meter and feel it vibrating my ear drum), and he didn’t notice anything wrong… would it be a fair assumption that your average audio book listener won’t be freaking out about it in the way that I am?

Maybe the undiscerning audio book listener won’t notice the hum?! …

They may not. But ACX will. Or the Robot will get there before human Quality Control.

We simulated what the robot is going to do with Will’s compilation testing tool called ACX Check (if I haven’t mentioned that already). It’s a plugin that appears under Analyze.

http://kozco.com/tech/audacity/Documents/acx-check.ny.zip

That may be the older one. The newer one is prettier, but says the same things. I gotta fix that one of these days.

ACX has a very simple weed-out process that catches ratty submissions. When you’re speaking, your peak sound values have to stay below -3dB, your volume or loudness has to be between -18dB and -23dB, and when you stop talking, what’s left (the background noise or Room Tone) has to settle below -60. If you do a very good quality job of announcing, you can crank out one of these compliant clips without a big deal, or your clip can be made complaint with a simple volume change. I, my very self, have done this, so I know it can be done. It’s not fantasy and it also happens to be broadcast standard.

Then it’s on to the human Quality Control. If you got past the Robot by applying legions of filters, effects, patches, modifications, and changes, you will most likely fail QC with “Excessive Processing” (he said, gazing up at the above thread’s list of notches and filters).

Without knowing any more about your setup than the amount of corrections it appears to require, I would think about changing it.


You can achieve the -3 thing with Normalize. I use -3.2 instead of -3 for safety. Conversion to MP3 later can mess with this setting.

– Select the whole clip or show by clicking just above MUTE.
– Effect > Normalize: [X]Remove DC, [X]Normalize to -3.2 > OK

Then Analyze > ACX Check and see what it says.

Koz

Thanks, Koz.

I want to make sure I’m understanding everything correctly. You ran the clip through ACX check and it failed based on the sound floor being above -60dB?

I’m actually pretty proud of my home studio setup, and consistently pass ACX’s checks with just a high- and low-pass filter :smiley: – the problem is that this project (~6 finished hours) has already been recorded in a different studio, over 100 miles from my house, and I’m just trying to salvage it. If I were to paste clean noise over all the dead space that currently has the hum, would that be a suitable fix?

If not, and I go the route Trebor recommended (splitting the track into two tracks - one above 200Hz and one below, etc.) - can one of you please break that process down into steps (as though you’re explaining it to a drunk kindergartener)? Y’all are my heroes. Seriously.

Also, Koz, in reference to your point about the -3 – you said you use Normalize to -3.2. I’ve been using the Hard Limiter at -3.2 and that seems to work. Would normalizing be better, and if so, why? Thanks! Also, happy new year!!

I’m just trying to salvage it.

I remember now. Sorry. This is me going in the corner, drinking tea and not “helping” any more.

Normalizing (and its cousin Amplify) just turn the volume up and down. They are not “processing” tools and after using them, you can come along later and put the volume right back where it was with no harm done. We can’t reverse processing tools like limiters, compressors and noise reduction (other than UNDO). This is where the sticklers will post that you can cause intentional damage with Amplify and you can remove battery damage with Normalize and that can’t be reversed. All true, but those are celebrity applications. In normal use for most people they just make the show louder and softer with slightly different philosophy between them.

Pasting The Blackness of Space Silence over all the hummy bits will trigger the “Overprocessing” rejection. You have to do an extraordinary amount of work without seeming as though you’re working hard. I’m a great fan of Noise Reduction over Noise Removal because in judicious, gentle amounts, Noise Reduction is effective and ACX can’t find it.

You are rescuing a show bordering on forensics. Not a job to be taken lightly [he said, backing out of the room].

Koz

I’ve been using the Hard Limiter at -3.2 and that seems to work.

Dueling posts.

I don’t know how Hard Limiter works. That’s one of the tools that got massaged over the last couple of Audacity versions.
The final arbiter is ACX Check, followed by listening to the work. That’s what’s going to happen after you submit.

Peaks may very well change when you create the MP3. That’s the reason for 3.2 instead of 3.0. If you shoot for -3.0 in the WAV master and the MP3 has any conversion error at all, you lose. -3.2 may not be enough, but it’s far better than shooting for the real thing.

One other note. ACX puts great stress on consistency throughout a reading and chapter to chapter. It should not be necessary to constantly adjust the radio in your car to hear the whole show. Back when I traveled more, I used to greatly enjoy the Sarah Vowel audiobook CDs.

Koz

Hmmmmmmm OK but it’s not the Black Silence of Space though; it’s room tone from that very studio (recorded at the very beginning of our session, without the hideous fan or me stepping on the headphone cords or the AC kicking on in the next room or whatever the hell that damned hum is). So that shouldn’t count as overprocessing, right??

And to your other point - when I try the noise reduction effect, it takes away the hum but totally distorts the voice… It’s so strange – there are entire chapters and minutes-long segments of chapters that have no awful hum at all, and then suddenly it’s back! And then just as quickly sometimes, it’s gone… Just trying to figure out the best thing to do to those hummy sections. You’re saying the room tone paste won’t cut it?

Room Tone does work, but it may sound funny when the hum comes back during words. That’s the kind of thing they listen for.

totally distorts the voice

It can do that when you get a bad sample or profile…or crank the first number, reduction up past about 15 or so. I used to recommend up to 18, but I don’t recommend that any more. It’s too likely to give you cellphone/wineglass voices if you go that high.

This may be old news:

Generic Noise Reduction of the Beast: 6, 6, 6.
Stiffer reduction: 12, 6, 6
Maximum. Possible sound damage: 18, 6, 6. Beyond here dwell monsters…and reshoots.

You have a horror-reading if the hum has some of the characteristics of the voice in the same performance. In that case there is no good Profile or reduction setting.

A common error is to get a little of the performance in the drag-select Profile by accident. In that case, certainly, Noise Reduction will try to remove the voice (!!). The Profile step must be noise (or hum) only. No contamination.

As to the hum coming and going, I can generate a scenario for that past the obvious fans going on and off. There is a microphone shielding or wall socket grounding error where touching or gripping the cable can cause or cure live performance hum.

“No, I don’t know. But I touched the rubber outside of my mic cable once and the hum in my headphones went away. So that’s how I record now.”

Koz

I’m actually not kidding. There is a sister hardware error where you can touch your computer and the hum will suddenly vanish. So you can see if you play your cards exactly right, it’s possible to lean the right way in your seat and cause problems.

“Y’know you’re right. Every time I shifted to my left butt cheek the hum disappeared. Who would have guessed?”

Koz

ERMAGERD I have been using noise reduction wrong this whole freaking time. When I use the specs you just gave me, it works so well!!! :smiley: :smiley: :smiley:

Open a copy in Audacity , duplicate it by pressing “Ctrl”+“D”. So you then have two identical in-sync copies. On one copy filter-out everything below 200Hz with the equalizer, or a high-pass filter. On the other copy, filter-out everything above 200Hz with the equalizer, or low-pass filter. Then apply a [dynamic range] expander to the bassy sub-200Hz track only, which will attenuate the bass frequencies when the man is not speaking , ( i.e. when there is silence, or when female is speaking). Then create a mix of the two tracks.

If you post some audio where the speaker changes from male to female and back again, I can see if what I’ve suggested above can be done transparently using the free expander plugin called FloorFish : i.e. where you can’t hear when the expander goes off/on when the man starts/stops speaking.

Trebor,

Thank you for the offer. Unfortunately (fortunately??) I wasn’t able to find a place where the hum is consistent for a long enough period of time to include male/female dialogue as well as “silent” hummy space. I think it has something to do with the mic - that maybe we were inadvertently bumping it (or grabbing onto the stand! Dammit; I think we would often grab the mic stand) as we leaned in to deliver our lines [two humans/one mic, because neither of us had a mixer. Probably definitely going to invest in one of those for our next book]). And then other sounds that I’ve been lumping in as the hum have actually turned out to be various vehicles going by on the street outside. It is hella tedious, but I’ve been doing noise reduction on each of those sections separately (getting a different noise reduction profile for the dump truck that went by v. the Fiat, etc.). Some of the sections are too far gone and we just need to re-record them, but in many of them, the sound is faint enough that noise reduction does the trick nicely. I still plan to try your duplicating and equalizing fix.

One more question! I got the de-clicker and the de-esser, but the default settings on the de-clicker don’t seem to be making much of a difference at all. If you wouldn’t mind listening to the attached clip and letting me know what de-clicker settings you would use to fix it, I’d be super grateful. Thank you!!

Just spent an hour trying to get Floorfish to show up in my Effects to no avail. :cry: Downloaded it like a gazillion times, using the same process I’ve used in the past to download plugins that have then shown up in my Effects lists after re-starting Audacity. Any similar effects I could try? Thank you!

noise reduction profile for the dump truck that went by

I know you don’t need any more cold water poured on this project, but traffic and other constantly changing noises don’t work well with Noise Reduction.

As you’re finding, it’s required that you constantly change the profile to keep up with the noise for a certain number of words. Repeat for more words.

Even worse, Noise Reduction can get ratty if you can’t find enough profile. The solution is carefully copy the small sample you can find and paste/duplicate it multiple times on its own track and use the expanded track as the profile.

For example at my house, I can clearly tell which direction traffic is moving because in half of the travel (the east half), it ducks behind a hedge and the sound changes. That’s poison to the Profile.

we just need to re-record them

…which will dump you into sound matching problems. Nothing like having your voice change when there’s no story reason for the change. One very common Audacity failure is: “Can you make me sound like…?” Unless you’re trying for a game special effect like the Voice of Doom in Mega Zorroid III, we’re not going to do a very good job matching studios, particularly if one of them has a little reverb or echo.

While you’re in your well-behaved studio re-shooting the damaged parts, why don’t you go ahead and re-shoot the whole thing?

Koz

If you are trying to manually-remove occasional [truck] noises , Audacity’s spectral edit tool is a more precise implement than the blunt-instrument that is “noise reduction” …
Demo of Audacity's spectral edit tool.gif

Decreasing the “Step size” & Increasing the “Number of frequency bands” does a better job , but involves more computation , so takes more time …
suggested DeClicker settings.png
Faster, stand-alone, DeClick $oftware is available : this is the least-expensive I’ve seen … http://www.clickrepair.net/
[ In the past, a 2-3 week free-trial was available on “clickrepair” software ].