Need help interpreting mastering feedback from ACX

Recorded my audiobook then ran the first mastering step (as I understood it), normalization, and the file sounded worse, not better. Someone at ACX’s quality control very kindly listened to a sample chapter and gave me the following feedback:

“The main issue with your audio file is that it was recorded at too low a level. ACX requires that all submitted files have an RMS level between -23dB and -18dB, and a maximum peak level of -3dB. The submitted file measures at -40dB RMS, and -15dB Peak. When an audio signal is recorded at too low a volume, it produces a poor signal-to-noise ratio. Thus, when you peak normalized the file, it brought out the level of the noise floor which was not previously audible.”

First question: How do I even know what level I am recording at? I am sure this is a dreadfully basic question, but I don’t know where this information hides. Where can I view the RMS level??

Second question: Must I redo the recording, or can I fix the volume level through the compressor, limiter, or other settings?

Third question: Is there an “Audacity audiobook cheat sheet” somewhere that will tell me what steps to take in which order, and settings to try? I am completely baffled. This, despite having successfully recorded a novella a few years back! Can’t find my notes and have no idea how I did it!!

Oh, the trials of being “talent” … I actually used to do voiceover work in Hollywood, but back in the days when all I had to do was show up and read!

Many thanks for any words of wisdom you can give me.

Diane the Distressed

How do I even know what level I am recording at? I am sure this is a dreadfully basic question

Basic, but not obvious. This AudioBook thing not a version of talking into your phone to your sister is Schenectady. You are replacing a recording engineer and studio. Which can be done, don’t let me scare you, but there are some rules.

Audacity has tools and testing to produce good voice work and also make sure you’re going to pass ACX testing without actually submitting anything. I’m thinking as I write that, we can give you a good idea of how to pass technical standards. If you can’t read out loud, ACX may reject you for voice quality issues and that’s kind of out of our hands.

The submitted file measures at -40dB RMS

Chances are good we will not be able to rescue the work, or be able to rescue it in a manner ACX will be happy with. Did you read a whole book?

Since we’ll be doing this across multiple time zones, it may take a while. It’s not unusual for someone to want us to fix the show by tomorrow because the client is waiting. Probably not, but we do have a pretty good track record.

Describe your studio. If it’s your kitchen table, say so. Which microphone have you got, which computer, and which version of Audacity? In the back of your mind, pretend I’m trying to buy what you have. Don’t abbreviate numbers. I’m using Audacity 2.1.2.

If you’re following an on-line publication, point to it.

Koz

I bet you’re asking yourself “why isn’t there a video or publication about reading for audiobooks?” There is. ACX has them. Step one: Figure out which soundproof booth you’re going to install.

https://blog.acx.com/2014/06/26/how-to-succeed-at-audiobook-production-part-1/

http://www.acx.com/help/setup/202008260

Their goals are slightly different from yours. They want you to record perfectly with a top quality microphone in a studio because if you don’t, it’s expensive for them to reject you. I know that’s a little mercenary, but it’s true.

There are other resources as well. Transom has good instructions for basic sound recording.

http://transom.org/

This started out as an effort by people at National Public Radio.

I really should make a collection of those. I’ve tried several times to organize a How-To here, but there are just too many variables. The microphone suppliers want you to buy a simple microphone, record audiobooks and retire to a nice island in the Mediterranean. Majorca is good.

It’s a little more involved than that.

Koz

Now you have some idea what the person behind the glass wall was doing. We can give you a good push when we find out where you’re starting from.

It would be good to get a sample of work you already did, but that can wait.

Koz

Koz, thank you so very much!! Still digesting the information and need to check the links you posted, but very quickly … I’m using a Snowball mic in a foam Portabooth. Recorded a novella in my bathroom a few years back, thinking it was farthest from street noise, but although it was accepted I definitely could tell it was recorded in a bathroom -!!- So no more tile floors for me. Now I record in my carpeted office with a blanket under my laptop to absorb any fan noise, and just pause when a car goes by. And yes, I have recorded an entire book, and it was a lot of work, and I haven’t yet checked to see if all the chapters are low volume. Ai yi yi. They are likely to be similar, I’m afraid, because I tried to be careful to make my work consistent.

I am in a rather unique position as an actor and former voiceover artist who now writes novels. I should be able to do this-! They did compliment my narration, but if it has insurmountable technical issues I will simply have to start over. Since I’ve been working on the thing since August I am really, REALLY hoping I don’t have to do that.

OK, I’m going back to read your posts again.

uploading a sample per instructions on one of the posts you referred me to …

I’m using your posting as an excuse to generate better graphics. The modern sound meters are miles (or km) better than the older ones, but you have to know what they’re telling you, and my settings are different from the factory defaults.

Reading your post.

Koz

[Note to self: Get Diane to read all my forum posts.]

I applied two effects to your clip. Effect > Normalize which is a fancy way of saying I made it louder (under very controlled conditions).
Then I applied the most gentle possible Noise Reduction.
See sentence about 2/3 down.
Screen Shot 2016-11-12 at 19.38.01.png
That panel is from ACX-Check that flynwill developed from existing parts that mimics the technical acceptance robot at ACX.
There is still a little rain-in-the-trees fffffff in there and now I have it together enough to critically listen.

As we go.

Koz

Open the clip in Audacity.
Effect > Normalize: [X]Remove DC, [X] Normalize to -3.5dB > OK
Effect > Equalization: LF-Rolloff for speech > OK. (this is a special download)

Listen to the first two seconds with the volume advanced a bit. Do you hear the little pencil drop at 1/4 second? This hold your breath “silent” thing is harder than it seems. That 2-sec segment is called Room Tone.

Drag-select from 3/4 second to 2-1/4 seconds (avoiding the pencil drop)
Effect > Noise Reduction > Get Profile > OK.
Select the whole clip by clicking just above MUTE
Effect > Noise Reduction: 9, 6, 6 > OK.

Analyze > ACX-Check [read it] > OK. (this is a special download)

I need to remember where the downloads are.

This is where I stand back and let the golden ears tune for tonal balance, sibilance suppression, etc. I think you could probably submit just like that. Does that sound like you?

Koz
Screen Shot 2016-11-12 at 20.27.31.png

I know seeing all that written out is daunting, but contrast a printed recipe with watching your mum cook.

Koz

http://wiki.audacityteam.org/wiki/Nyquist_Analyze_Plug-ins#ACX_Check

That’s where to find ACX-Check analysis tool. This is how to apply custom software to Audacity.

http://wiki.audacityteam.org/wiki/Download_Nyquist_Plug-ins#download

LF Rolloff is a bit of an odd duck since it’s not a stand-alone tool.

You can get it from me. Unzip it to the .ny file by clicking.

http://www.kozco.com/tech/audacity/Documents/LF_rolloff_for_speech.xml.zip

Adding Audacity Equalization Curves
– Select something on the timeline.
– Effect > Equalization > Save/Manage Curves > Import
– Select LF_rolloff_for_speech.xml > OK. (it won’t open the ZIP. You have to decompress it)
– LF rolloff for speech now appears in the equalization preset curve list.

I expect it to look like this. Click the graphic if needed. The green line is dead flat to the right.
Screen Shot 2016-11-12 at 20.39.36.png
Even with all that, you’re really close to being not quite loud enough. The quiet limit is -23dB and you’re at -22.6dB. Just a bit off and you’ll fail RMS (Loudness). In this case, you can apply Effect > Normalize again at the above values. If you read through a whole chapter, one or two other tools may be needed to force loudness, but once you’re this close, the rest should be relatively easy. I didn’t hear any room echo or other problems in there, so whatever you did for soundproofing seems to work.

Koz

I don’t remember saying what you did for the 20 second posting test is probably optimal for your microphone and studio. Do that more.

Occasional sound meter peaks up to -6dB is recommended roughly like this:

http://www.kozco.com/tech/audacity/pix/AudacityRecording-650w.jpg

Live recording volume and posting to ACX are slightly different so that gives you my Effect > Normalize recommendation.

– Select the whole clip or show by clicking just above MUTE.
– Effect > Normalize: [X]Remove DC, [X]Normalize to -3.5 > OK

Under perfect conditions, I can create an ACX compliant work with Announce > Normalize > Submit.

I’m not doing that with a Snowball and I have a very good room, but it can be done. ACX compliance is not magic or unattainable.

Koz

Success! Yes, it still sounds like me. My screenshot of the equalization looks different from the graph you posted, but I can’t figure out how to give you a screenshot. Grrr.

Your “recipe” format for instructions is exactly what I need! I am extremely grateful! I’m still not sure if I understand how to figure out the levels I am recording at, or the levels I end up with. After following your recipe, I end up with (via ACX Check) Peak Level -4.1 dB, RMS Level -22.5 dB, Noise Floor -68.4 dB, RMS (A) -30.2 dB, RMS Level (A) -75.4 dB, DC Offset 0.000006%. Naturally I have no idea what all that means, but it says the clip meets ACX requirements. I am going to re-read your posts and see if I can figure out where the levels are displayed. It seems to have something to do with the line of numbers at the left of the timeline, yes? But those numbers don’t seem to correspond to the numbers in the ACX check. The lowest line looks like it’s touching just above -0.4 and the highest peak is around 0.5.

Are there some settings (other than mono) I should/could set before I record?

Okay, back to studying your extremely illuminating posts.

I can’t figure out how to give you a screenshot.

In Days Gone By, you would press the PrtScrn (print screen) button and that would take a picture of your desktop. Then you open Windows Paint and paste the picture in. Use Paint to Export or Save a nice JPEG picture. I do this stuff to my desktop. Your mileage may vary.

From a forum text window, scroll down to Upload attachment…etc.

I’m sure nothing works like this any more.

I end up with (via ACX Check) Peak Level -4.1 dB, RMS Level -22.5 dB, Noise Floor -68.4 dB, RMS (A) -30.2 dB, RMS Level (A) -75.4 dB, DC Offset 0.000006%. Naturally I have no idea what all that means.

It means it was developed by an engineer for whom there is no such thing as too much information (TMI).

By now you know 0dB is maximum loudness and volume goes down with the negative number. -10dB is louder than -20dB. These numbers only appear on the sound meters even though the meters and the blue waves are measuring the same thing. The blue waves are in percent. -6dB is 50%. That’s why when you announce with your sound meters poking up to about -6dB, the blue waves only go half-way up. Those two are the same number.

“But wait,” I can hear you doing the sums in your head. “That doesn’t fit.” Right. The blue waves (unless you change them) only measure about the loudest 30dB of the show and then they run out of poop (technical term). That’s the Audacity default and I like to edit that way because normally, a theatrical presentation is carried in the loudest 30dB. It’s like zooming in to the important stuff.

Only the top three numbers are significant.
Peak Level -4.1 dBCan’t get higher than -3dB.
RMS Level -22.5 dBThat’s overall loudness. You have to hit a range between -18dB and -23dB.
Noise Floor -68.4 dBNo background sound (room tone) louder than -60dB.

If you have a noisy microphone, room or studio, you may find you can hit two numbers but not all three. It’s not unusual for a new user to post with the laundry list of effects, filters, patches and adjustments they had to make in order to force everything to work out. And remember, they still have to pass the Human Quality Control for whom “Overprocessing” is a valid failure.

So a quiet studio is a reeeely big deal. I would not like to read a volume of quality text this way, but I did manage to record a conforming clip with my laptop built-in microphone—in my dead quiet room.

I need to go play real life now.

Koz

Isn’t part of the ACX submission process to post a short chapter or some test clip or something?

I think this link is still the valid requirements.

http://www.acx.com/help/acx-audio-submission-requirements/201456300

If nobody posts that they hate your voice or some other error, I think you could probably read their test requirement, submit it and see what they say. Whatever you did for the forum posting is fine. It’s good to report back what they say if they find something wrong. Was this a phone call? Can you record them?

If you have a chapter that fails compliance, post the “bad” ACX-Check panel and we may be able to suggest a gentle goose (technical term) to bring it into line.

For example, in your test clip, you had good, valid “elbow room” left in Peak and Noise, but you were right on the edge for RMS (Loudness). It’s possible to “cheat” the whole thing slightly louder and more gracefully pass everything. Tiny adjustments like that are not obviously audible and will not violate the ACX “chapter matching” goal.

Koz

You are intended to more or less watch the meters and read at the same time. You are replacing the recording engineer, but the job didn’t go away. I random announce while gazing at the meters to get an idea what it feels like, and then launch into the reading, glancing over every so often to make sure everything is OK. I’m not a presenter. YMMV.

From your own experience, you know you can’t produce an almost straight blue wave. We can’t rescue those. That test you submitted was remarkably close to ideal.

You might suggest this is a terrific place for recording automation. Those exist, it but they’re remarkably difficult to get right and the ACX inspector is waiting for evidence of odd noise such as “pumping” or noise breathing. Yet once again, if you’re presenting into a studio console, you might get away with that, but not a Snowball or Yeti.

Home techniques tend to be self limiting.

Koz

Okay, I performed your steps with my Chapter 1 and passed the ACX check. Now I am wrestling with Chapter 2. Attaching a screenshot. Looks to me like the overall volume is low, and the reading is too dynamic (since I have at least one loud peak). Am I reading the tea leaves correctly? If so, will some sort of compression or limiter effect fix my issues? Should I try “amplify” on my files?

I hope you are done with real life now and ready to address something truly important, like my audiobook.
audacity screenshot.jpg

Am I reading the tea leaves correctly?

[Puzzled puppy look]

I’m not sure how you got to that ACX-Check reading from my brief list of tools. It should be impossible.

There is a sister post to yours concerning what happens when somebody harrumphs very loudly, clears their throat several times, coughs once and then starts to read. Those throat noises, although perfectly natural and expected, will almost certainly overload and doom the reading if they’re not cut out before processing. Processing and filters have no idea what’s show and what’s not.

Audacity and its internal software does not understand content. Sometimes they will try to make the Harrumph perfect and to heck with everything else.

Similarly, my silly joke of a gunshot in the middle of a perfect voice recording will doom the show. Even though it’s only there for a tiny fraction of a second, it will take over some of the tools most of whom are really stupid about decisions like this.

So. Did you Export a perfect quality, WAV file of the raw, unprocessed reading? That’s highly recommended. We can’t take effects out of a show and the instant you close Audacity, UNDO goes away.

Do you have DropBox or other file sharing service or account? I would want to see the whole raw recording and 20 minutes will not fit on the forum even if you compress it to MP3 or other small sound file format.


I’m just reading back through that. It really is a college semester isn’t it?

Koz

This is, without question, the worse possible time to mention that the longest forum thread (39 chapters) was produced by Ian, who ‘only’ wanted to read audiobooks with a Blue Snowball from his apartment in Hollywood.

Koz

Run a close second by Bruno who was “only” trying to record his acoustic guitar … :sunglasses: