Is a Warning equivalent to a Fail?

Hi,

I’m using Windows 7, with Audacity 2.4.2

Have created a couple of different files. After analyzing them in ACX Check, I get warnings, but passes in the same result box. For instance, “Peak Level: -7.00 dB (Warning - may be too low or overly compressed or too quiet”). But then, it gives me a “pass” on RMS level.

Also, have been getting the dreaded warning for “Dead silence sounds unnatural.”

So, my questions are: 1. Will the warnings prevent me from passing the checks when I submit the recording? 2. How can I fix the Peak level issue? 3. How do I fix the dead silence issue (other than ask my dog to hum as I record)?

Thanks in advance.

Regards,
Jay

“Fail” means that the measurements do not comply with the specifications published by ACX.

“Warning” means that although the measurements do comply with those specifications, there may still be a problem.

That indicates that the plug-in detected either absolute silence, or “almost” absolute silence. Nobody has a home studio quiet enough to get a noise floor that low, so it begs the question, how did you get the noise floor so low?

Thanks, Steve.

Not sure how I got the noise floor so quiet.

I’m using the book, How to Make an Audiobook by Rob Branson, and am using the settings he recommends.

Regards,
Jay

What did it say the noise floor level was?
If it said “-inf”, then you have some “absolute silence” which may be due to including “white space” (empty track) in the selection that you were analyzing,


Did you use any noise reduction or noise gate effects?

Have you checked that Windows Sound “Enhancements” are all disabled? (see: FAQ:Recording - Troubleshooting - Audacity Manual)

And, although it passes the technical specs/limits it might still get rejected when a human listens…

Have created a couple of different files. After analyzing them in ACX Check, I get warnings, but passes in the same result box. For instance, “Peak Level: -7.00 dB (Warning - may be too low or overly compressed or too quiet”). But then, it gives me a “pass” on RMS level.

It means that the peak-to-RMS ratio (difference*) is unusually small and it may sound unnatural. Usually with natural voice that ratio is more than 20dB (too much for ACX before processing), so after getting the RMS level right you usually need limiting to bring the peaks down into spec.

The RMS level is a kind-of average and it’s an indication of “overall loudness”.

The peaks sort-of represent the “loud parts”, so the peak-to-RMS ratio is a rough indication of “dynamic range” of “loudness range”. But, the peaks don’t correlate well with perceived loudness because our ears don’t respond instantly to short-term peaks. You can slightly reduce (limit) the peaks with little or no effect on the overall sound character/quality. If the peaks go over 0dB** you’ll get clipping (distortion) and the peaks determine how high you can turn the volume control (without distortion).

  1. How can I fix the Peak level issue?

You can’t un-do compression or limiting but if you still have the original recording [u]this procedure[/u] will perfectly-nail the peak & RMS requirements every time. (That’s assuming you have a good-quality original without any clipping or “enhancements”.)

  1. How do I fix the dead silence issue (other than ask my dog to hum as I record)?

They want some natural constant very low-level room noise. The noise may not be audible unless you crank-up the volume with headphones but it should be measurable.

You can add white or pink noise but normally a “home studio” has excess noise so normally it’s a matter of not getting too aggressive with noise reduction (or noise gating).

I’m using the book, How to Make an Audiobook by Rob Branson, and am using the settings he recommends.

I don’t have that book so I don’t know what kind of “processing” you’re doing.




\

  • Decibels are logarithmic so a difference (subtraction) is a ratio.

** The ACX maximum peak of -3dB is an unusual requirement… For most applications your peaks can hit 0dB, or some people normalize (or limit) to close to 0dB (like -0.1 or -1dB). And, MP3 compression often changes the maximum peak so you can have a peak of 0dB and when you make an MP3 it can end-up with peaks over 0dB. For that reason, some people shoot for -1dB peaks if they are making MP3s.

Steve, thanks. The noise floor was -113.83 dB. Not sure why it’s so low. I do wait about ten seconds in silence before I start speaking into the mic, and then take a quiet piece of that “waiting time” to insert between sentences or paragraphs when my lips/tongue/whatever gets clicky/poppy, whatever.

DVDdoug, thanks for the tip and link. I’ll give it a try.

Regards,
Jay

-113.83 dB.

That’s crazy low. The best you can do outside of Audacity is -96dB. That’s where “regular” sound files numbering schemes run out. You only get -113dB when you do something magic—and naughty—inside Audacity. Audacity uses 32-bit floating internally, not 16-bit for regular sound files. That’s so when you apply effects, filters, and corrections, you don’t create permanent distortions by accident. But you lose that protection when you make a sound file.

For one example, After the Loudness step in Audiobook Mastering, it’s not unusual for a particularly expressive chapter to overload louder than 0dB. The next step, Soft Limiter, gently brings everything back down to normal and everything is perfect. if Audacity didn’t do that floating trick, anything louder than 0dB would be permanently and loudly damaged.

Your reading process is too aggressive. You can make intentional processing do that, Noise Gate or Noise Reduction, or you can have automatic corrections do it. Sometimes, Windows internal processing can do that or sometimes Zoom or Skype processing will hang around even after you close the session. Super Silence is a Zoom trademark. That’s the only way they can have 9 people on the same conference without the background noises driving everybody nuts.

ACX tends to not like processed sound. They think it’s distracting. That’s the ACX Check warnings. The reading didn’t “fail” but they’re not perfect, either.

It can be said, frequently, that the best way to produce good quality theatrical sound files is not use the computer. Many people have posted using good quality microphones plugged into stand-alone sound recorders such as the Zoom H4n and the H6. I used an H1n for my sound test. By itself with a roll of paper towels in a quiet room.

Boop. All the computer sound problems vanish.

If you’re still set up, read a ten second sound test and post it on the forum.

https://www.kozco.com/tech/audacity/TestClip/Record_A_Clip.html

Unless we can hear an actual sound example, we’re all guessing.

Koz

OK, that’s not good news. A dangerous link warning popped up when I tried to investigate the book.

How to Make an Audiobook by Rob Branson

Apparently it tried to use doubleclick.net to launch a tracker on my system.

Koz

Koz,

Thanks for the very thorough replies. I ordered the book on Amazon, so didn’t get any weird attempted hacks or other shenanigans.

For noise reduction, the book has me use, Effect>Noise Reduction, and then set Noise Reduction (dB) at 30, Sensitivity at 15.00, and Frequency Smoothing (bands) at 4.

For amplification is has me set the new peak amplitude to 0.

For compression, it directs me to set Threshold to, -30 dB, Noise Floor to -60 dB, Ratio to 3:1, Attack time to 0.10 seconds and Release Time to 1 second.

It has me set my Target RMS level to -20 dB.

It has me set Limter to, Type: Hard Limit, Input Gain (right and left) to 0 (I record in mono with a Blue Yeti), Limit to (dB): -7.00 db, Hold (ms) to 10.0 and Apply Make up Gain: No

Mind you, I have no idea what any of this means. I’m just a reasonably successful author trying to DIY those books of mine that the publisher(s) did not turn into audiobooks.

I’ll try to send a clip later.

Thanks!
Jay

Noise Reduction (dB) at 30

That’s very aggressive. We discourage noise reduction higher than 12. Many close readers make good use of noise reduction of the beast (6, 6, 6). Gently quiet down any background noises.

High noise reduction can make you sound like a bad cellphone or talking into a milk jug or wine glass. ACX’s goal is natural voices and listening to someone telling you a fascinating story over cups of tea, Not Zoom voice.

Can’t wait for that test clip.

We’re not the only process out there, but we have a pretty good record for getting people published. Also rescuing people who failed other processes.

Nobody is holding a gun. If you want to go with the published process and ACX Check can’t find anything actually broken, then go for it. Fair warning though. ACX stopped offering the full evaluation services for test files. That’s how I found out my reading sucked (but my files were great) without having to actually submit a book. I think the best they do now is the actual ACX version of ACX Check. Theirs is hiding behind a questionnaire and red tape.

So the first time you actually encounter Human Quality Control (and find your voice frightens people) is when you submit all fifty-six chapters of the book. Awkward…

Can I buy your book on Amazon? That’s a new requirement. It’s worth mentioning because if I can’t, that’s the end of the road. People have tried publishing both paper and voice at once and it’s not pretty.

Koz

As the creator of the Audacity Limiter effect, I would not recommend using “Hard Limit” for audiobook recordings. That setting is too aggressive in my opinion for spoken word.

I notice that the instructions that DVDdoug linked to (Missing features - Audacity Support) says to use the “Soft Limit” setting. In my opinion that is a much better option to use for spoken word.

Remember, the ten second test has no processing at all. Read it, cut it to length, export WAV, and post it. It doesn’t have to come out perfectly even on the word. We’re not submitting for an Academy Award.

Koz

Mind you, I have no idea what any of this means. I’m just a reasonably successful author trying to DIY those books of mine that the publisher(s) did not turn into audiobooks.

That’s not an unusual situation. If you are a famous author or celebrity, or if you have lots of extra money you can go into a pro studio with a recording engineer and maybe even a producer to help you with pronunciation and enunciation, etc. But a LOT of authors have to become recording engineers if they want an audiobook…

For noise reduction, the book has me use, Effect>Noise Reduction, and then set Noise Reduction (dB) at 30, Sensitivity at 15.00, and Frequency Smoothing (bands) at 4.

That’s why you’re getting a warning and it might be too aggressive. Nobody should be recommending noise reduction settings without knowing what kind of particular noise problems you have (if any). Even then, it often takes some trial-and-error.

The Audacity-recommended procedure has some noise reduction hints (if you fail noise) after everything else.

For amplification is has me set the new peak amplitude to 0.

If you are using compression it’s good to have a “starting point”. Otherwise, this is redundant because RMS Normalization is going to make another volume adjustment. But the Audacity recommendations don’t include compression.

For compression, it directs me to set Threshold to, -30 dB, Noise Floor to -60 dB, Ratio to 3:1, Attack time to 0.10 seconds and Release Time to 1 second.

Dynamic compression (unrelated to file compression like MP3) evens-out the volume by making loud parts quieter and/or by making quiet parts louder.

Limiting, which we do recommend, is a kind-of fast dynamic compression.

It has me set my Target RMS level to -20 dB.

Perfect! Again, RMS is an indication of loudness. At -20dB you’ll have the same loudness as other audiobooks and loud enough so the listener doesn’t have to crank the volume all they way up, etc.

That’s a simple-linear volume adjustment (up or down) for whatever is needed to hit the -20dB target. The change to the RMS, peak, and the noise floor is the same dB amount. Most often, it’s a volume boost which increases the noise level and usually pushes your peaks out-of-spec. That’s why it’s important to do everything in the correct sequence. (If you run RMA normalize a 2nd time with the same settings, nothing happens.

It has me set Limter to, Type: Hard Limit, Input Gain (right and left) to 0 (I record in mono with a Blue Yeti), Limit to (dB): -7.00 db, Hold (ms) to 10.0 and Apply Make up Gain: No

Too much limiting. After RMS normalization your peaks are usually too high. That’s just the nature of natural-unprocessed dialog or music… The difference between the peaks and RMS is often more than 20dB.

Limiting “pushes down” the peaks without messing-up the RMS level. The ACX spec says your peaks should be no-higher than -3dB. (Digital dB levels are usually negative so a smaller negative number is higher and a bigger negative number is lower.) The Audacity recommendation of -3.5dB (and soft limiting) gives you 1/2dB of safety margin so you’re not on the edge of the actual ACX spec.



P.S.
ACX mostly wants a natural sounding recording with predictable loudness/levels and without excess background noise. It shouldn’t sound “over processed” or “enhanced”. It should (almost) sound like you are there live reading the book to the listener.

And a side note. The ACX standards are really close to broadcast, so if you can pass those, you can post and submit just about anywhere.

Koz

Koz and friends,

Thank you so much for so many detailed answers–it’s very gratifying. It’s especially nice that the explanations are in language I can (sort of) understand.

I’ve attached two short clips. One has the processing that I described earlier. And one does not. I hope that they are useful.

In reply to an earlier question, Koz, yes, the book(s) that I am hoping to narrate myself were traditionally published and are available on Amazon as hardcovers, softcovers and Kindles.


Regards,
Jay

The unprocessed clip is broken. It’s too loud and it has several points of overload and clipping damage.

Open in Audacity and set View > Show Clipping. The time line will reveal red bands at the points that the digital track failed to follow the performance.

Screen Shot 2020-11-18 at 11.53.36 AM.png
There’s a tiny one at about 8.3 and three others at 16.8, 18.9 and 20.8. The 18.9 one is especially serious because it’s six overload points grouped together.

There are repair tools such as “clip fix”, but they have to be applied individually for each overload point and they only approximate the blue waves and not the sound.

I have to drop for a while.

Koz

This is the fuzzy goal.

Blue wave tips at about 50% and/or the bouncing sound meter just starting to turn yellow from green.

That’s the happy medium between too loud with risking overload and permanent sound damage, and too quiet where microphone system noises start to compete with the show.

You got really close with the other points and measurements. You did miss one one. You really need to hold your breath for between one and two seconds before you start speaking. We need that gap to measure the actual background noise/Room Tone.

And yes, if you post a mono track, you can go up to 20 seconds.

For the brief term of the actual speech, can you kill the air conditioner or other ventilation? As way up the message thread, the less processing and noise reduction the better. We designed the mastering system to sound exactly like you except meeting the audiobook goals.

Koz

Koz,

Thanks.

Yes, I did as you directed and the clipping shows the same.

So, it’s too loud. Do I move further from the mic, or…?

Yes, I can kill the air conditioner. I’ll try to post a new clip in the morning.

Thanks again for the help.

Regards,
Jay

So, it’s too loud. Do I move further from the mic, or…?

Don’t post anything new until we resolve this. This is serious. Can you see Audacity from your announcing position? You can get a good idea what you’re doing from that. Use my goal picture.

There’s tricks to this. I don’t see anywhere that you told us which microphone you were using. Ideally, it’s a type that allows you to connect headphones so you can listen to your live performance.

(Use real headphones, not Apple earpods)

That can go a very long way to self-limit volume drifting. That’s why everybody you see in a studio has headphones.

Having very high volume is really unusual. Home performers always come in too quiet and we have to talk them through boosting it up.

Do you have that tennis racket blast filter in front of the microphone? If you do, the spacing should be about a power fist or little further.

You will hear yourself distorting your own voice if you’re listening on headphones and get too close. If it’s a directional microphone (almost 100% likely) your voice will get boomy and ballsy if you get too close. I once played two different people by messing with the spacing. It’s a significant change. On the other hand, if you’re after the Radio Announcer Voice, this is the way to do it. “NBC Radio News On The Hour…”

Let’s see. You’re on Windows. There is a Windows setting to boost quiet microphones. You may not need that.

Something like this? I’m not a Windows elf. Check that.

What have you discovered?

Koz

Thanks Koz,

I have a Blue Yeti microphone (the ~$150 version) and have the gain turned essentially all the way down. Volume is down as well.

I actually have TWO pop filters in front of the mic. If one is good, two must be better…yes? :wink:

My face is about one fist–with pinkie and thumb extended–away from the mic.

To reduce reverb, I have a three-side, adjustable sound shield (egg crate(?) foam) surrounding the microphone. Am in a walk-in closet, well-hung with clothes.

I do wear full headphones, Sony. Voice sounds okay to me as I read.

I can see Audacity as I record. On the sound bar, it does occasionally bump into the yellow, with very rare and brief excursions into red.

As you can see, I’m getting “wave tips” up to 1.00 and beyond.

Is any of this helpful?

Holding off posting any more clips per your request.

Regards,
Jay