Advice on Chris's Dynamic Compression: use INSTEAD of RMS Normalize? When to do the De-Noise?

[EricMC]

Guys, thank you, thank you!

That is such good advice — and all in one thread.

(Meanwhile, I have got more acquainted with Chris’s Compress Dynamics, RMS Normalize, Audacity’s downloadable Noise Gate, and the very effective, built in Noise Reduction filter.) I have summarised all the answers as I understand them in a fuller essay further below. If you’ve time, I’d be grateful if you could correct me on any misunderstandings.

In a nutshell, though, the lightening answers to my many questions are:

1) Yes. Filter first, then edit. But, before anything, Export a copy of your raw recording.
2) No need to Compress; it might make your reading sound flat and can introduce noise problems.
3/4) Either use Noise Gate or Noise Reduction; perhaps no need for either;
[–]but, if you must Compress, do the Denoise first — not after any other processing.
5) Don’t bother to Declick unless someone tells you to. Mouth noises are natural
[–](no one answered this, so I’m inferring from Koz’s remarks on just keeping processing to a minimum)
6) LF Rolloff is optional — but you’d better be in a very quiet room and not using a USB mic.


This leads me to two further questions:

• Regarding these “USB Mics”? I use a Heil (practically the opposite to a condenser mic, it is very up-close and directional, enabling musicians to isolate their singing voice from the echo of the band — and me to reduce background noise and reflection without egg boxes and a duvet).
  It has an analogue cable requiring a preamp: I have that cable connected to a red Focusright Scarlett 2i2 box
  — This, in turn, plugs into my computer’s USB socket (from which the Scarlett also draws its power).
  — Does that make it a USB mic?
  — Or, by USB mic, did you just mean something that plugs straight into your computer without preamplicication?
  .
• One important final question: does RMS Normalize simply raise the volume of everything uniformly (thus keeping background noise absolutely constant)? It affects answers 2a and 3 of my longer summary below, which I may need to correct.
  
  
  So here is the detailed summary for the solutions to my queries above.

1. Best processing order:

• Export a backup of your fresh raw recording to .WAV
  • Run any effects — esp. those described in Koz’s AudioBook Mastering version 4
    (all tools and instructions on one page; just download and install the filters, and follow steps in the order listed).
  • Run an ACX check to see you’re in the ball park.
  • Export another backup of processed audio at this point
    (since still unedited, should match the raw .WAV file second for second)
  • THEN do your edits (deletes, cuts and pastes).
  • Save, run ACX check again, and Submit.

2. No need to Compress (sounds more natural without).

• (That said, my client takes my samples out for a spin in his car and gets agitated if there is too much variation between loud and soft parts — so I’m resigned to some degree of levelling)
  [–]So, if you wish to compress (especially with Chris’s Dynamics):

1. No harm in running RMS Normalize first, right?
   It’s simple and takes some of the pressure off the compressor
2. With Chris’s Dynamics, try “running on” for a sec or two at the end of a reading (speaking gobbledygook or “Cut this, cut this, cut this”) which you can clip out in the edit (hopefully stops that volume fade at the end of paragraphs; that said, I found setting the “hardness” control high enough may also reduce this).
3. I can find no one recently who uses the 1.2.7 beta, nor further improvements — nor, even,
4. recommendations for any newer compressor. My client loved the sound of what I’d sent through Levelator. This works brilliantly and comes highly recommended, but it is older and pretty basic: there are no controls; if you’re unlucky, just sometimes things can come out a tad fuzzy — take it or leave it; and, since no longer supported, it may eventually become obsolete.
5. The “Floor” and “Noise gate falloff settings” in Chris’s Compress Dynamics are handy — but less subtle than the downloadable Nyquist one (see top and 3 below); probably best left at 0 (certainly no more than 4); play around a little with the Floor setting so you don’t ever dampen your own speech. Better still, use the Nyquist one below
   If you are compressing with low-level background noise, then do some gentle de-noising beforehand, or there might be unacceptable variation in noise levels (harder to strip at the end). You can use NoiseGate or Noise Reduction:
   .
6. Noise Gate simply dampens sound quieter than a certain “floor” level, but leaves that sound otherwise unprocessed. This is pretty handy before you Normalize° or Compress°, both of which risk amplifying your background noise:
   __

• Normalise raises the volume of EVERYTHING by a uniform amount: determined either by the loudest “peaks”, or from taking an average volume across your recording (RMS is short for Root Mean Squared) and boosting that average to a uniform level (which might lead to clipping — then to be pared down by a separate peak “limiter”)
• Compress boosts the quiet bits and pares back the overly loud bits (Normalizer and Limiter in one package); as Koz points out, it could nullify some of the natural emotion in your reading as well as draw attention to background noise by “pumping” it up and down.
  I tried gently using NoiseGate before Compress dynamics 1.2.6 and (especially when there was background traffic hum) it actually worked much better.
  .

4. Noise Reduction… seeks specific steady background “noise” to strip out; it comes ready installed with Audacity (you might have to enable it), and does a great job on steady white hiss, traffic hum, fridge buzz, etc., but beware: although it applies itself more heavily in pauses, it may also gently strip some of those tones from your speech. ACX doesn’t like the voice to sound too electronic — also, after processing, your breaths may sound more intrusive — so just be wary of setting it too high; steer clear of double-digits. But, if used sparingly, on steady background noise levels, it’s damn handy.
   .
5. Declicking. From what I’ve read, I’m going to infer that this is not necessary. Sure, if you have a seriously loud problem with mouth noise, then you may want to run the filter just once (and live with collateral loss of clarity where 't’s and 'd’s get inadvertently damped). And, sure, if you’re Stephen Fry reading Winnie the Pooh or Harry Potter, then they’re going to throw thousands of pounds at some studio bod who will sit for a month and clean up every lip-smack whilst preserving your fine articulation. But ACX wants natural readings at a bargain price. So don’t get bogged down. Right Guys? Same for de-essing? Only if you whistle like the musk rat in Deputy Dawg, right?
   .
6. LF Rolloff. If I understood correctly, it sounds as though you can get away without the LF Rolloff Equalisation stage (but you’d better be far away from muffled traffic and not using a USB mic, or ACX’s automatic controls might reject you)
   Also, from what I can gather and intuit, it would make no difference whether you use it right at the start — or even after the final edit. Right guys?

[christianw]

But ACX wants natural readings at a bargain price. So don’t get bogged down.

Audible customers rate you publicly on performance. Reviews of sloppy processing or annoying sounds can be brutal. Audiobooks are expensive and intimate. Getting them right is usually a lot of work.

It sounds as if your performance has few problems with mouth sounds, breaths and so on. That makes it easier.

[kozikowski]

Pieces of that are wrong, but I can’t stick with this right now. Sometimes the Low Rolloff for Speech is needed to correct noise problems inside microphones and interfaces and has nothing to do with your voice. We called it that because that’s roughly what Hollywood and the studios call it and the most people will recognize it and what it does.

I gotta go.

Koz

[kozikowski]

I need to hit and run. Blitz-Posting.

Koz, is the complete instructions on one webpage — no guidebook .pdf to download?

It’s not even a web page. It’s a forum post. Posting it to the Audacity wiki manual is on my 2-Do list. There is one up-side to the post. It’s a snap to post corrections—not that it’s had that many. But it’s not easily searchable.

“USB Microphones” are understood to be the ones that plug directly into the computer with a USB cable. The G-Track is a USB microphone.

They come with baggage. They almost invariably come with low volume because high volume and overload in the hands of a New User is immediately and permanently fatal to a performance.

You can’t separate the performance location from the noisy computer by more than one USB cable. By contrast, you could put the noisy computer and XLR interface out in the garage.

Technically, that would work, although it’s a little rough to see the Audacity sound meters out there.

I need to go.

Koz

[EricMC]

More good advice. Thank you very much.

And thanks, for the offer to have a listen.

Please find a sample of my recording set up as a .WAV file.

It was recorded around 4am, when traffic quiet.

I may send another sample from later in the day - when you’ll hear more intrusive traffic noise.

[kozikowski]

I personally would not put punctuation marks in filenames. Upper case, lower case, numbers, -dash- and underscore are the only universally acceptable characters. Otherwise, you could submit paid work to someone running an older Windows machine and have some entertaining emails as to why they can’t open your work.

~~

That was easy.

I applied Mastering 4 and Noise Reduction of the beast, 6, 6, 6.

There is a technique to listen. Scroll forward and set your listening volume for normal voice. Then scroll back and listen to the whole thing. Don’t change any settings. If I did it right, you should hear little or no noise at the beginning.

That was the good news.

The awkward news is your sibilance. You have strident “SS” sounds in your words. Many new microphones do that because it sounds “professional.”

There is a “DeEsser” program available about which I know next to nothing and I have had reasonable luck with custom equalizer settings—about which more later.

https://forum.audacityteam.org/t/updated-de-clicker-and-new-de-esser-for-speech/34283/1

I know DeEsser has a very gentle effect after you master a work. I experimented changing the first value to -30dB from where it normally is. I understand that’s the value which sets the amount DeEsser squashes the harsh tones.

I could listen to a story in that voice. I could probably listen to you read the phone book.

Koz

[kozikowski]

Skip the DeEsser at -30dB. That makes you sound like a cartoon character. I got reasonable results on your sample at -20dB which I believe is the default.



Koz

[kozikowski]

“Glazing.” That would be “glass,” right?

I am concerned about this being the most quiet time for recording. Not that it didn’t work. It worked fabulously, but if it turns out this is the only time you will be able to work…

Microphone system hiss (fffffffff) was the reason I used Noise Reduction this time. And technically, you didn’t need it, but you passed by the thickness of a playing card, so that’s not repeatable or stable.

Traffic.

Effect > Noise Reduction doesn’t work on moving or changing noises. So if you can arrange for the same cars and buses to go by repeatedly, we may be able to help you. Noise Reduction is a bloodhound. You let it sniff the noise (profile) and it goes howling off looking for that exact same noise in the show. If the noises changes…

Prepare a new test under Difficult Conditions and see how it goes. This is a good deal less of an emergency now because you have a known, good, working technique.

Koz

[kozikowski]

There’s been a little bell ringing since I heard your test. You’re a professional presenter, right, and you decided to record either your own paper book as an audiobook, or record other people’s books unencumbered by the studio system.

Did I hit it?

Koz

[kozikowski]

Since you’re this close, you should start to worry about Computer Hygiene.

Export each chapter raw reading as a WAV (Microsoft) 16-bit sound file and copy it somewhere safe. This against the time the laptop fails during an edit and takes the track with it into the bin. You should never be in a position to need to read it again.

And after you finish your edit, make a new WAV (in spite of ACX requiring an MP3) and that becomes your archive. Then make the MP3. Then be able to point to two separate places that contain your work. This is against the time the dog eats your laptop.

That’s no problem, just get the backup thumb drives from the credenza. Cloud storage works here, but I’m leery of corporate shenanigans. “Doobly-Do Corporation declared a hacking intrusion that wiped their servers.”

Darn.

Koz

[EricMC]

Thanks again, Koz. All great advice.

No. Not a professional presenter so much as a humble actor with a marketable voice.
The author found me on ACX, because he liked my second rate impression of Lawrence Olivier narrating World at War.

I have done V/Os here and there, over the years and, until I tried this ACX malarky, I had no idea just how much work the poor engineer put in after I went home.

Very sad to hear my sibilance is de trop (I did wonder, but thought I could get away with it — on the grounds that every presenter under forty seems to lisp these days; but that’s actually a trick of the microphones, eh?).

I had been toning down extreme esses by hand, using parametric EQ. Easy to do in seconds, here and there — but a pain if every last ‘s’ actually needs attention. Thank you for the suggestion. I will look at that and various other packages.

I got caught late at work, today, and will send the worst-case noise scenario tomorrow.


ATB, Eric

[kozikowski]

a humble actor with a marketable voice.

I wondered. I spent a long time in Washington, DC and when Dulles airport went on line, they had the Tannoy/PA System announcements in your voice. “American Airlines flight 45 now boarding, gate 5.” It was glorious while it lasted.

As I get closer to actual experience with DeEsser, I found I can watch the filter working. DeEsser, when launched and set, produces a progress panel on the screen and if you can watch the blue waves in your timeline, tiny portions of each blue word-blob vanish or decrease in volume.

I worried that such a waveform change could affect ACX conformance, but it didn’t seem to. DeEsser certainly helps with the “ice pick in the ear” sound of harsh sibilance. This is such a wide-spread problem I’m learning how to spell “sibilance” without looking it up each time.

I found the best way to test for effectiveness is not so much flipping back and forth between two different timelines, but use one line and only correct half. Critically listen at the switch point and it’s possible to more clearly hear the cutting edge of each S sound vanish.

You might try the correction on a freshly mastered clip and overdo it with the -30dB setting just to hear the “cartoon” effect.

every presenter under forty seems to lisp these days; but that’s actually a trick of the microphones, eh?

It’s a bad trick. One producer claimed on the forum he DeEsses everybody and produces desirable results almost always. It’s visible. You can Analyze > Plot Spectrum and see a little “haystack” caused by the boosting of certain vocal tones. This led to analyzing the tones and writing a custom Effect > Equalization setting to suppress it. That’s effective, too, but it has to be done for each presenter/microphone. DeEsser has analysis and detection built-in.


What did you decide for theater corrections: breaths, tongue-ticking, and wet mouth noises? My Opinion is ignore the breathing unless you’re clinically asthmatic like one reader and concentrate on ticks and clicks.

And there our story turns to christianw and his vimeo presentation. I believe he addresses those corrections.

Making an Audiobook with Audacity

Making an Audiobook with Audacity

Note he uses other Mastering processes. We both get there using very different techniques. Theater corrections are addressed after that.

Koz

[kozikowski]

I’m going to vanish in about two hours. If you have any testing you want, you should hit it before then, or wait until tomorrow.

Koz

[EricMC]

Oops. Sorry. Just spotted that.

Thanks again. It should be a small matter, then, to de-ess the chapters I have already edited.

I am into work for an early call this morning.

I realised that, on a Sunday, there is not so much background traffic - so decided best to leave it until this afternoon or tomorrow morning (my Tuesday), when there will be the usual heavy trucks through the day.

I’m thinking perhaps I should send a 20sec sample of one of my finished chapters — to check that it’s not going to sound too processed for ACX.

All the best,


E

[kozikowski]

If your problem is low pitch rumble and walls shaking, most of that is going to vanish in the First Step : Equalization. That’s one of its jobs. If you can actually hear clashing of gears (does anybody clash gears any more?) then yes, that may be audible. Honking is pretty obvious.

I don’t remember mentioning I have a bad pavement patch in front of the house. If the Metrobus hits it just right, I get a 1.8 Richter in the front room, but I’ve never heard it in any recording. Fair warning if the rumble and vibration are enough to cause stuff in the house to vibrate, then the stuff may make it into the show.

There’s a Hollywood joke about the wine glass. You can’t photograph an earthquake, so you are required to show wine in a glass making rings. If it’s bad enough, the glass falls over.

Yes, post a finished clip.

Koz

[kozikowski]

I’m thinking perhaps I should send a 20sec sample of one of my finished chapters — to check that it’s not going to sound too processed for ACX.

It will be our opinion, but yes. We’re on the edge of our seats.

Koz

[EricMC]

Great advice. Thank you very much.

Yes, I live in a port town at present and there are trucks rumbling by all the time. The lorry gears do whine as they accelerate - and, when they trundle over a certain section of bridge there can be bangs and thumps. Though that wine glass would not register a ripple. My friend, who is a video editor, went over a couple of chapters for me and said I was worrying too much about what I overheard through the mic monitor — that the truck sounds were barely audible to him.

Unfortunately, I just learned that Osaka is on a three day bank holiday — so I won’t get normal daytime traffic until Thursday.

Instead, I will attach two samples, if that’s OK: the raw recording; and one I have just started to process (with LF Rolloff, Noise Gate, RMS Normalize to -23, and some de-clicking* — mostly only between spoken bits*). I experimented with making the LF Rolloff a little less steep, so as to preserve more lower harmonics.

I have not limited yet. I was wondering whether to use what had become my usual (Chris’s Compress Dynamics) or switch to your formula (Effect > Limiter: Soft Limit, 0, 0, -3.5dB, 10, No > OK). It also occurred to me that it might be OK just to run the limiter once at the end, after the theatrical edit?

[*De-clicking. I am restraining my usual urge carefully to remove every last pop, smack and mic “nit”, because I’d love an opinion as to what ACX finds passable. My edits take four times too long as I obsessively clean the spoken audio — eg., applying declicker only between plosives so as not to muffle 't’s and 'd’s. I need to streamline my approach, or hand the audio master to someone else]

[kozikowski]

Don’t change Low Rolloff for Speech point by point. that way lies insanity. Instead, change the “Length” slider to the left. From the original recipe, the length slider should be about 5000 for mastering. The filter will be less strict with a lower number. And yes, you would one of the victims of that filter affecting men’s announcing voices. If you have been using that filter with a 7177 length as in that illustration, just returning it to the recommended 5000 or slightly less might be enough.

Failing to use a filter is not recommended because of digital microphone oddities.

Here’s how to find out.

Select a goodly chunk of Room Tone (see how handy Room Tone is?) > Analyze > Plot Spectrum.

You can grab one corner of the display and make it wider to reveal more information. This is what the bottom of mine looks like.

Those are the settings. It will tell you if you didn’t select enough Room Tone.


I found an “evil” clip to use as a sample.

The display is low pitch sound on the left, high on the right. Louder is up. It looks at the whole clip you selected at once. It can’t tell time. In a perfectly recorded analog microphone system, there would be little or no activity to the left of 20Hz. Digital systems, in contrast have no real world restrictions.

Here’s a bad one. Note that there’s activity at 2Hz (reading on the bottom). Not only is that not audible, it’s not even sound. That’s an earthquake.

This is Plot Spectrum after Low Rolloff.

It starts the affect at 100Hz and suppresses everything below that. That number was picked as a trade-off between damaging your voice and getting of rid as much rumble as possible. Everybody knows what the 100Hz filter does. It actually appears as a switch on the sides of some microphones.

Instead of Effect > Equalizer, you might try a custom setting of one of the other tools.

Try Effect > High Pass Filter, 40Hz, 24dB.



Listen to what it did. Write it down if you like it and use it as the first step instead of Low Rolloff.

~~

If you’re happy with the Chris process you should stick with it. Limiter directly affects passing ACX Peak. If you pass that now, you don’t need Limiter.

You should try and avoid mixing and matching suites. Mastering 4 is a series of tools that produce predictable affects and then clean up after each other. If you use them out of order…

I obsessively clean the spoken audio

Yup. That’s a problem. It’s not a good idea to crank up your headset volume to “dive for noise.” You will be doing that one book for the rest of your life.

There is a technique to submit a test to ACX

How to send a test to ACX

Just sharing in case this helps others… As a newbie, I am trying to make sure everything is done as well as possible, so after benefitting from a 10 second test here (thank you!), I found that I could also send a longer test to ACX for feedback. Here are the instructions I received from ACX: “I understand you would like to submit a sample audio file to our Audio Team for review. This survey only accepts mp3s no longer than 10 minutes, at a 320kbps maximum bit rate. To begin this process, ple…

Koz

[EricMC]

This is terrific. Thank you.

It’s now nearly nine am here, and I have noticed that many truckers are not observing the bank holiday — so I did take a couple of samples of traffic at it’s worst.

Nb. I would usually close my blackout curtains — and my air vents — to keep the worst of it at bay.
To show the contrast, this is morning traffic with them all open:

And here is the same time of day with everything sealed:

This is an analysis of the heavy traffic room tone:

So, if I understand correctly, there is really no sound beneath 35Hz - and that bactrian camel hump is just an artefact of the mic?
Well worth removing.

[kozikowski]

to keep the worst of it at bay.

You could record that way, but I probably wouldn’t. Attached is the last clip with Mastering 4, Noise Reduction and DeEssing.

Note the noise is -62.5dB. That means even with noise reduction, you’re still within spittin’ distance of failing noise (-60dB). The next more powerful noise reduction is the last one before you start to experience voice distortion.

So now several unpleasant options present themselves. ACX puts great stock in matching chapters. So:

– Arrange to record with roughly the same traffic sound through a whole book.

– Record everything at night.

– Mix and Match times with stiffer Noise Reduction, hope nobody notices and be sure to apply the same corrections to everything whether the chapter needs it or not.

– Mess with the other technique and hope Noise Gate can pull you through. Remember, noise accommodation is very different between the two techniques. In Chris’s Technique, it must go at the front.

– Add physical noise suppression. That would be my first stop. One of the other posters who hasn’t checked back yet is trying a “portable studio” that shows promise.

I need to drop out and search for the posting.

Koz