A good example of passing ACXCheck, but still producing Junk

In case anyone is wondering if passing the ACXCheck Analyser in Audacity is the ultimate target - please have a look at this, and you will quickly see that there is more to this than meets the eye.

I enclose a sound file here. 8 seconds of recording, and the image that proves that it passes the ACXCheck. However - if yo listen to the file, you will agree that if I were to send this into to ACX - the engineers would fall dead on the floor in amazement…

Sorry about the image size, it’s a screen dump of the Windows 10 - x64 build I’m working on, built with VS2015, and I was doing some testing of it. I tested the resulting audio file - checking that the plugins work etc, and was somewhat surprised. Is it actually working I ask myself :slight_smile:

Yes it’s working, but this is a good example of the limitations of this “ACX test” plug-in.
The actual test that ACX use is a lot more sophisticated than this simple Nyquist plug-in, and would not be fooled by your “SampleAudioPassed-Fail.wav”.

The reason that the plug-in measures the noise floor so low is because the sample starts with a very low noise floor for the first half second. The plug-in is not sophisticated enough to notice that the noise floor for the rest of the track is over 30 dB higher. Also, the plug-in has no concept of “sound quality”, it just dumbly measures amplitudes.

As you rightly say, there’s a lot more to producing a professional quality audiobook recording than just hitting the numbers. The numbers just give a bit of a guide for interpreting ACX’s technical specification.

Thanks Steve, I did check it on my “good” version on the Mac. Just in case … dreadful sound quality but nice figures. The question “Is it actually working” was rhetorical … :slight_smile:

I just put it up as a sample to others struggling to get to the magic numbers…

I don’t think the ACX Robot was ever intended as a final arbiter of acceptance. Just the first one.

Your submission fails basic sound quality standards —> help web pages and videos.

It would be a corporate nightmare if Human Quality Control had to manually dig through every submission, and I suspect it catches the vast majority of newbies with dreadful sound. Given the popularity of audiobook reading, they must get thousands and thousands of submissions.

I do wonder if they send all submissions through the robot, or only newbies. Do you still go through if you’re on your fifth book?

The actual test that ACX use is a lot more sophisticated than this simple Nyquist plug-in, and would not be fooled by your “SampleAudioPassed-Fail.wav”.

You know this from your conversations with them? What does theirs do?


Yes, from conversations with “them”.
Their test break up the audio into “chunks” at detected silences. It looks at the leading silence (part of the requirement) and at the silences between words/sentences. It expects to find a similar noise floor throughout. It checks the peak and rms level for each chunk, and flags a warning if there is excessive variation from one chunk to the next. “Warnings” may not mean automatic “fail”, but it alerts their QA of potential problems which they can then look at / listen to. “Some” measurements are automatic “fail”, such as being way outside the basic published ACX numbers. Rather than just giving a single pass/fail, if the test has not failed outright, then it generates a summary of what the dynamics are doing through the duration of the chapter or submission. This summary is “probably” the basis of the feedback that they send back to the audiobook producer.

Rather than just giving a single pass/fail, if the test has not failed outright, then it generates a summary of what the dynamics are doing through the duration of the chapter or submission.

Whomever wrote that is probably resting comfortably in a home somewhere. They would have had the same problems we do. Only it’s paid work.

Step 1: How bad is it?

…the basis of the feedback that they send back to the audiobook producer.

Carefully formatted to sound like the comments of a concerned reviewer.

“We’re just crushed some of your sound quality is a little off…”

Actually, ACX is four people in a second floor office in Palo Alto (BritSpeak, first floor) with a massive server farm somewhere.

The nightmare is a highly respected and famous author producing trash. OK, now what do you do? Suggest, respectfully and deferentially they get somebody else to read it? Commission somebody to read for them? That might work.


Actually, that’s not a dreadful addendum to ACX-Check. Instead of misleading test values such as: You don’t pass RMS, it reads: The volume of the whole clip is too low.

How about: All three values are too low, even though technically, that’s not true.

This in the face of the distressingly common problem of submitting a show almost, but not quite a straight blue line.

Where’s Will?


The limitations of acx-check are clearly stated both in the wiki page Wiki page and again in the comments in the the file itself.

I don’t know what Koz means about “all three numbers too low”. As I read the "requirements, they want the RMS to be no lower than -23 dBFS, if that requirement is met then the clip is probably loud enough.

Well, who could have guessed that my simple attempt to show people - aimed at newcomers to audiobook production - that the quality of their recording is what really counts the most. A picture I thought is worth a thousand words, because at leat a thousand words have been spoken on this very subject, seemingly to no avail, because people still ask the same question. It seems no one has heard the expression RTFM …

So, I produced, quite accidentally, the worlds worst recorded clip, yet pictorially it shows as “Passing with flying colours” - which of course it doesn’t. That’s not the fault of the plugin, ACXCheck, which is a brilliant little bit of software. It’s proof positive, that there are no shortcuts to a good clean recording.

In a nut shell then.
Make your recording in as clean an environment as is humanly possible. Your recording should be pleasant to listen to. It should have a flow of it’s own. You should be able to listen to a couple of hours of it.
You should not be able to hear the fridge turn on. Nor the car door slamming across the road. Nor the aircraft passing overhead. Nor your computer fan screaming at the cpu overload.
A good rule of thumb. If you can hear it, so can the mic. Even sounds you no longer notice, like your jaw clicking - the mic will notice.

You should be able to record your audio session on a good hand held recorder, and it should be almost production ready.
You then use Audacity to get the intro and outro blanks right. Modify chapter segments, and edit out the odd pin dropping that can’t be helped. Then run ACXCheck and see how close you are. If you aren’t even close, then you have got something wrong.
If I can do a segment on my iPhone with iRig recorder, and it’s within a few clicks of being spot on, anyone can do it.

… and there’s nearly another 1000 words on the subject :slight_smile:

Robert (OP)

That’s a very good point, and to take it further, if you aren’t even close then it’s highly unlikely that any amount of processing will fix it.

Personally I think it would be useful to have a “pre-processing test”. Running the p-p test on the raw unprocessed, unedited recording would show your peak level and noise floor in a label track in multiple places. Unfortunately we are not able to change the colours of labels, otherwise it would be nice to show warnings in red labels. If the recording is way out at this stage, cut your losses, undo the recording and fix the problem(s) with the recording set-up before starting again.

if you aren’t even close then it’s highly unlikely that any amount of processing will fix it.

If you aren’t even close, ACX-Check will tell you your RMS is low like it just did to me on a dead-flat silent clip.

Do I need the Photoshop blur to show you what all those numbers look like to a newbie?

Bla, bla, bla, bla, bla, clip fails ACX because of low RMS.

What is this manual thing of which you speak?


I think it would be useful to have a “pre-processing test”.

Yes, but internal to ACX-Check against the condition of conflicting results using different tools.

I think more can be done with Peak. Regardless of the specification, Peak has a lower limit. A peak value of -25 or -30 could trigger an ‘overall clip volume too low’ warning instead of (or in addition to) the misleading RMS message. -30 is where meaningful blue waves vanish.

As I posted in the email to flynwill, ‘The Hindenburg failed to maintain buoyancy’, while technically correct, is misleading.


I have an irig lavalier, and either Pro Audio To Go, or iRig Recorder on an iPhone. would that work? Samson C01u into a Macbook Air doesn’t give me enough volume not to fail RMS on ACX Check?

Yes, the iRig Lav + iRig Recorder on an iPhone should work. Record it, then send the recording to Audacity to clean up.

I use both a Shure 55SH Dynamic Mic ( phantom power off )and sometimes a BTSKY Condenser mic ( phantom power on ) + iRig Pre → iPhone + iRig Recorder. Both produce excellent quality sound on the iPhone SE and the 5C, and the iPad for that matter. Audacity 2.1.1 to clean it all up.

Your Samson C01u mic is actually a condenser mic, and may or may not have phantom power built in - (48v) so I’m thinking that this is the reason for the low output. It is advertised as being suitable
http://www.samsontech.com/samson/products/microphones/usb-microphones/c01u/ Even has a photo using a Mac in the web page … so I’m surprised that it’s not delivering the goods. Have you got sound levels etc set properly in System Preferences - > Sound? I guess you have.

However, it looks like a good mic, so maybe needs bit more experimentation. How much is it failing the RMS level by? This could be tweeked up by a little judicious ‘effect’ tweeking.
I only ever use Limiter, and Normalize. If I can’t get it with those two, I go back to square one.

but in short - the iRig stuff is good.

A lavalier is a special purpose microphone. It’s job is to hide. Secondary jobs are good quality sound at good volume. There are classes how to get reasonable quality sound and still not be able to see the microphone. Some lavaliers have custom sound to make up for hiding, so unless you’re shooting a video or movie, that may not be the best choice.

Nobody is excited about carefully watching sound levels during recording. Too much volume and overload is immediately and permanently fatal to the performance. So almost all consumer level microphone systems feature minimal or no volume indicators and low volume. If you play your cards right, you might have a flashing light that’s green for “go” and turns orange for warning and red for “too late, start over.”

Maybe not even that.

Couple that with a soft-spoken performer and you can get a low volume performance that’s non-recoverable. Those performers fall outside the microphone’s design center. I don’t know of a simple solution. My voice goes through sound-proof walls and and is rated a non-lethal weapon by the State of California, so I have no trouble cutting a sound track with any microphone—sometimes not even in the same room.

I once had to capture a voice track from a shy Asian woman who wasn’t all that excited about performing in the first place. She did it as a favor to the Producer. It took a portable sound-proof studio, good quality studio microphone and a pro field sound mixer to do it. iRig need not apply.

So there is no simple “buy this microphone and all your problems are over,” although all the suppliers would like you to think there is. Sign here.

There is one universal. Whole worlds open up for you if you start from a quiet, echo-free room. Trying to boost the volume of a noisy recording is a career move, and we should remember “Overprocessing” is an ACX failure. Yes, the show passes ACX Check, but it sounds like a swimmy cellphone.

Couple a quiet room with good recording levels and you should be good to go. Audacity provides two different sound indicators: the bouncing light sound meter and the blue waves. Neither should go all the way up during the performance. This is the rough goal:

Maybe occasional orange is good. Stay out of the red meters. You can make the sound meters large by dragging the graphic. It’s malleable. The meters in that graphic are expanded so they fill the Audacity window. Good to keep an eye on those meters.


Thanks Rachal, actually I tried this, and I still end up with the RMS problem when I import from the iRig recorder app on the phone into Audacity on the Macbook. In fact, it’s a little bit noisier even than the Samson. I have a feeling my room is the problem- I can never find a noise floor lower than around 51-54 I fished around the room with the mic, as mr. koz recommends, but to no avail. Using, LFRollof, Compress then Normalise, I can pass ACX Check, but perhaps that’s not the best way to go about things.

I have the sound levels set to maximum in System Preferences and also in Audio Midi Setup. In the latter one call also set Output volume, which is surprising for a microphone, but I think it is the level of the monitoring that comes through the headphone jack on the body of the mic. Without Normalising, it fails RMS by at least 10. So it might be 33 db for example before Normalising.



I have got my “Victorian padded room as quiet as I can” and have tried your fishing around with the microphone trick to find a quieter spot and have failed. Maybe I need corner bass traps?

(sorry for the wrongly formatted post, I’m new at this)

(sorry for the wrongly formatted post, I’m new at this)

Just don’t post in all upper case letters. We come after you with sticks if you do that.

Post A Sound Sample.



This format produces almost everything we need to analyze your world. Nothing quite like analyzing a system across multiple time zones and not be able to hear it.

Do not do anything to it. Record it, Stop and Export. The only “production” you need to do is cut it if it’s too long. The forum will refuse to accept it if it’s over about 10 seconds stereo or 20 seconds mono. We’re not producing “Girl On The Train” here. You don’t even have to be in English.


haha, I wouldn’t dare :slight_smile:

I posted one a little earlier, and being afraid to post it in the wrong place, Gale kindly made a new posting for it. It is here https://forum.audacityteam.org/t/posting-acx-test-sample/43383/1

I have also attached it here in case this is a more appropriate place for it, but I am happy to delete one version of it, in order not to clutter up the forum?

Thank you so much for looking at this, sir.