Is ACX Check Lying to Me?

Hey all,

I have been recording for Librivox for a couple of months now, and decided that I’d take a stab at ACX. I was very skeptical that my setup would be capable of meeting their noise floor requirements as I live in an apartment (ew), and own an old Blue Yeti (also ew). But today I ran ACX Check over some of my raw recordings, and it is telling me that they pass the technical requirements, as pictured below:

NoProcessing (2).png
Out of curiosity, I then ran Mosquito Killer over the recording (to banish the Blue Yeti curse) and equalized it with a lightly modified version of Low Rolloff for Speech. This also passed, even more clearly this time:

WithProcessing (2).png
Here are my specifications:

  • Audacity Version: 2.3.2
    Exported on Windows 10 Pro w/ Audacity
    Recorded w/ Blue Yeti USB (Not Pro)
    Recorded onto an Android Tablet (Lenovo Tab 8) via an OTG Cable + Audio Evolution Mobile Suite
    Recorded in a walk-in closet with a homemade mic box

Could I get a second opinion on the recordings? Do they actually meet the noise floor requirements? I am baffled as to how I am passing when I am using low end equipment. I don’t want to start trying to do ACX only to have to revamp my entire setup partway through because ACX Check had a slight bug or something.

If you actually read this monstrosity, thank you and I appreciate all of the hard work done on these forums.
~Buswheels

I then ran Mosquito Killer over the recording (to banish the Blue Yeti curse)

The Yeti Curse (frying mosquitoes) is a bad interaction between the Yeti (typically) and the computer. The computer maker says: “We’ll cut a few corners on the USB system because it’s cheap and, hey, who’s going to notice?” Meanwhile, the microphone maker is going: “We won’t bother to filter, process or clean the USB system because that’s expensive and, hey, whose going to notice.”

Only everybody who gets both of those together. You are listening to the digital data inside the USB cable leaking into the analog microphone amplifier inside the Yeti.

I don’t remember who first put it together that the annoying buzz was the same pitch that a conventional USB system uses to transfer data. Once we got it that far, one of the senior elves wrote a tool that suppressed it, since the process is published and usually dead predictable.

I don’t remember who it was and they didn’t sign the program.

I was very skeptical that my setup would be capable of meeting their noise floor requirements

It’s my opinion that anyone careful with their recording technique in a quiet room can meet ACX. The specifications are remarkably similar to Broadcast Proof Of Performance. They’re strict but not magic.

I’m having a hard time believing you performed directly into ACX. Are you sure you didn’t tune the volume a little after you read the work? It is possible to read that way, but it’s super unlikely.

Let’s assume you’re not funnin’ us and you did actually do that. That announcing technique almost guarantees a good noise performance. The object is to get your voice loud enough so it swamps the noise but not so loud that it overloads the sound channel. You win.

Low Rolloff for Speech is designed to mimic the rumble filters available on many if not all field sound mixers. In our case, not only does it remove wind noises and some P-Popping, but also the ultra low pitch trash that some microphones put in the sound by accident. After all, nobody can hear it and who’s going to know?

The answer this time is somebody trying to meet ACX. The ACX specification for noise is to measure everything in the background whether you can hear it or not. So yes, you can mess with the curve, but there really should be one there getting rid of most stuff lower than about 100Hz. If you have a ballsy male voice, then you can hear it working and a lower rolloff value is indicated.

ACX Check Lying to Me

Maybe. ACX Check needs at least 3/4 second of clean, pure background noise to get a good measurement. If you don’t put such a segment into your performance, then ACX Check will measure whatever it can find. Usually, that’s you breathing or shuffling in your seat. Ewwww.

only to have to revamp my entire setup partway through

It’s good that you checked. It’s a very new user mistake to read a whole book and only then check if everything is OK.

ACX will happily evaluate a sound test. It takes between one and two weeks for them to get back to you.

Last I checked they want between one and five minutes and you need to conform to their submission requirements.

I need to look that one up. Back shortly.

Koz

Screen Shot 2019-05-15 at 20.06.25.png
Screen Shot 2019-05-15 at 20.06.47.png
Those were taken from a posting by someone who wrote back to us what the evaluation experience was like.

https://forum.audacityteam.org/t/how-to-send-a-test-to-acx/49588/1

The MP3 goal is 192 Constant Bitrate minimum, Mono performance, and that’s the goal most people shoot for. Yes it is silly overkill, but ACX is going to resample the works for various products and services and your voice has to survive all those gymnastics. The quality will drop a little every time they do that.


Screen Shot 2019-05-15 at 20.19.40.png
You have it all over most people reading for the first time because you’re not reading for the first time. You already have the story-telling thing nailed from your other experiences. The ACX version of ACX Check I fondly call “the robot” is the only first test. After that the performance goes to Human Quality Control and that’s where they do theatrical evaluation. Can you read out loud? Do you have clinical asthma?

Good luck.

Koz

One more piece.

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

Koz

a homemade mic box

How did you do that? I just published a workup on a Popular Mechanics version of a mic box.

After I built that, I drank some beer, plowed the south forty and did a ring job on my '57 Chevy.

Koz

There are several things working in your favour that cause problems for others:

  1. You appear to be recording in a very quiet, substantially echo free environment.
    This is the BIG one.
    If you record in a kitchen, surrounded by hard surfaces, with computer fans, a refrigerator, air conditioning, traffic outside the window … then the excessive noise floor kills the show.
    The importance of a good recording space cannot be over emphasized.

  2. You’ve got your mic position and recording level nailed. The recording level is high enough to avoid excessive microphone noise, but you’re not so close that it causes wind blast or distortion problems.

  3. With the type of material that you are reading, and your delivery style, the volume level is pretty constant. Achieving a constant recording level becomes more difficult when the delivery is more “dynamic” (such as dialog that includes whispers and shouting).

Achieving good quality voice recordings does not require very expensive equipment, but with less expensive equipment it can become more difficult and you have to get everything else right. Mid-range equipment (around $100 - $200) should be adequate for good voice recordings, provided that you take care with the set up and recording environment.

High end equipment can be a bit more forgiving and provide a bit more wiggle room for correcting problems, but you still need to take reasonable care over the set up and environment. It is still perfectly possible to produce trash recordings with equipment costing thousands of dollars.

Do write back when you get your evaluation. Quote the post.

Everybody with a pulse is trying to read for audiobooks and ACX’s rapid evaluation can be misleading. For one recent example, they complained about excessive compression. What actually happened was the reader was overloading their sound channel and the resulting distortion sounded like excessive compression.

Koz

Thank you for the detailed replies! If I miss replying to something let me know, I’ll try to be thorough.

I did not alter the unprocessed version at all after recording. For full clarification, the recording suite on my tablet has the option to control the input level of any input channel. I have it set to reduce dB from the Yeti by -2. This was after my earliest recordings had nasty distortion, despite the fact they weren’t really even close to true peaking. A symptom of USB microphones and beginner mic technique I can only assume.

Here is a different recording sample which doesn’t meet ACX specifications due to RMS:


I cannot honestly remember what input level that recording was at, but it was at the end of a recording session and too quiet and monotone. I doubt that my accurate RMS and Peak values in the original post will be easily attainable across a narration, as these are just poetry readings without a lot of dynamics.

I’ve wanted to do narration and voice-over for a while, but after listening to some rather poor readings and seeing the technical requirements I figured it would be best to practice in a sandbox first. Librivox is wonderful for that for sure. Hopefully I’ll be able to meet ACX quality control standards and not develop asthma.


It’s just a plastic storage tub from Walmart lined with mattress foam like this:

I can’t remember the adhesive I used, or I’d list it. The old carpet in the mic box is there temporarily for height as I’m still saving for a proper microphone arm so that reading longer texts will become feasible. If I made it again I’d make it taller and build something in it to hold a reader of some kind. I do rather like your version Koz, I may steal it someday…

Thank you for the list Steve, that’s actually really helpful to know what to prioritize. Would upgrading my equipment be a good idea before attempting paid work, or is it better to wait for some amount of progress before further investing?

I tried to find where to submit a sample for review, but wasn’t able to find it on the ACX website itself. Maybe I missed it. Do I need to send them an email asking about it explicitly? I thought they’d discontinued listening to plebeians who didn’t have recording contracts already.

Thank you again for the feedback,
~Buswheels

https://audible.az1.qualtrics.com/jfe/form/SV_265YEkshVJXHqPX?c=

I do rather like your version Koz, I may steal it someday…

Now is good. The only good effective sound padding in that picture is the rug. Packing and shipping foam’s goal is to take up room and not weigh anything. It’s too fluffy to do much good for sound.

If you threw a block of that foam to anybody who wasn’t expecting it, it would bounce off, they would be upset at you and that’s that. If you hit them with a moving blanket, you could put them in the emergency room. Lift two gallons of water. That’s one moving blanket.

If you picked up a piece of commercial Acousti-Foam, you would be surprised how dense and heavy it is. And expensive.

Can you get any more carpeting like that? That counts. Drive around back of a carpeting store after hours and see if they have remnants in the dumpster. Some of the studios at NBC Chicago have floor carpeting that doesn’t stop at the corners but continues up the walls.

As Steve above, after you get done beating up the room, I don’t see any reason not to continue with the microphone you have.

You know how to get rid of the mosquitoes, your voice is loud enough and the noise is fine.

Both can be easily fixed if you drift off standard.

Just as an exercise, apply the three mastering tools to one of your other works. Pick one that’s slightly low. Use your custom Low Rolloff if you feel like it.

Low Rolloff (rumble filter)
– Select the whole clip or show by clicking just right of the up arrow.
– Effect > Equalization: Low rolloff for speech, Length about 5000 > OK

RMS Normalize (set Loudness between -18dB and -23dB)
– Effect > RMS Normalize: -20.00 > OK

Limiter (no peak louder than -3dB)
– Effect > Limiter: Soft Limit 0, 0, -3.5, 10, No > OK.

There is one not so obvious goal in that mastering protocol. With the exception of Low Rolloff, if the tool isn’t needed, it doesn’t do anything.

I need to change that wording. Audacity 2.3.2 has a formal Track Select button now. No more “Click in the tiny blank spot just to the right…yadda…yadda.”

Koz

Carpeting stores also have that flooring foam that looks like a multi-colored trash pile compressed into sheets. That stuff might work. It’s job is to be dense and stiff.

I didn’t use Home Depot for the pads. I walked into a moving company in Gritty South-East Los Angeles (I’m required to say “gritty”) and said I wanted ten moving pads. I was making a walk-in studio.

http://www.kozco.com/pictures/boothFinished/laptop-mic.jpg

Each one of those walls is double thickness. Note there’s one on the floor.

Koz

I followed the link and this happened when I tried to use the survey:


I’ll email ACX this week and ask them directly about QA, I’m sure I’m not the only person who has been confused by this process.

Well, luckily I live within walking distance of a flooring and carpet store! I’ll have to ask around to see what I can find, or raid their dumpster as you suggested.

Here’s the sample from last post with those mastering tools applied:

Appreciate the feedback on the micbox that I made, as it’s better to learn what I’m doing wrong now rather than later, when some hypothetical author is upset at me. Will update when I’ve actually made some progress.

~Buswheels

I followed the link and this happened when I tried to use the survey:

It’s highly animated and I wouldn’t be shocked if some browsers can’t deal with it. Do you have cookies enabled? Try a different browser. You may also find you can’t “back up” in the posting process. If you find something in step five that causes you to rethink your answer in step three, tough.

I’ll email ACX this week and ask them directly about QA

They are swamped with millions of requests. I was not able to get email response.

There’s another interesting observation. We hosted one of the ACX representatives on the forum for a while. It didn’t go well. We were trying to resolve “talking into the microphone” problems and he wanted to discuss publicity management, rights assignment and contract negotiations with clients. That’s why we developed production tools on our own.

There was another disconnect, too. They did resolve the background noise and echo problems. They recommended a fully qualified walk-in sound studio in your apartment. They got a lot better at this instruction and training thing since we started, but they still insist, as do we, that you have to start with a quiet, well behaved voice recording. You can’t fix it later with software.

I live within walking distance of a flooring and carpet store!

That’s the desperation method—zero cost option, but you do have to be careful to get thick, heavy carpeting, not thin, indoor-outdoor stuff. Do you have room to put a script or tablet in there with the microphone? How do you do that? That’s one reason the kitchen table studio is as big as it is.

ACX check said it was too low but now it passes perfectly.

ACX Mastering guarantees Peak and RMS leaving Noise as the only technical problem.

Though, I did forget to swat the mosquitoes.

Anything that’s not voice is going to throw Mastering off. To bring this around, that’s why Low Rolloff is first. That gets rid of non-sound trash before the critical loudness setting steps.

Koz

invictus_test_master4.wav

Works for me. Again, technical conformance is the first step and you still have to pass theatrical and vocal quality tests. We can do some of that, but that part is mostly up to them.

Koz

I downloaded the ACX Check plug in after my first few novels submitted to Audible failed miserably. I couldn’t understand what I was doing wrong. What’s more? I have 20-years of commercial radio experience, using digital editing programs at most stops!

How could I possibly fail the Audible tests?

Well, I was failing. I wasn’t too old to go back to school, which is essentially what I did after downloading the ACX Check plug in and testing and retesting every file from a 28-Chapter monster of an audio book that I filled with production elements (it’s about submarine warfare in WW II).

I knew the files were good. I knew my mixing skills were good. But how do I get from good to passing Audible’s requirements. That is where I went back to school.

I cannot explain to you why I take the additional steps to use the Amplfication tool to reduce amplification by 2.5 or more. Nor can I tell you why I run that file through another production step, use of the Hard Limiter. The only thing I can tell you is those two additional steps produced the types of files that Audible wants. When I ran each chapter through the ACX Check plug in, they passed. So, I held my breath and submitted everything to Audible for approval.

Imagine my shock when I received a notification 12-days later that “your audiobook is now for sale.” No problems. No bumps in the road. Just clear, smooth sailing.

I had to mess around with the Amplification and Hard Limiter settings quite a bit before I finally hit upon the right solution. Again, I don’t know the WHY of what I’m doing (which bugs me because I am a former reporter). I just know it works. And it works, because the ACX Check Plug In works.

That is odd, but congratulations.

Did you go through our Audiobook Mastering Suite before ACX rejected you? You never said those words. I’m guessing no because ACX Check is part of that protocol. If you only started using ACX Check in a way later step, then we could have saved you a lot of work.

You kept their exact failure notice, right? Can you post it?

ACX Check is our product based on ACX’s explicit publications and actual experience. They had no hand in it.

I don’t know the WHY of what I’m doing

If you use our mastering suite, I can bore you to death with details of what you’re doing. If you scattershot it like you did, it’s a lot harder to figure out what happened. Write it down. If you’re not clear what you did, then you have to do it exactly the same way every time.

I’ve been clear multiple times, our mastering suite is not the only way to do this. I think it’s the minimum work and the least sound damage to get to ACX acceptance.

Koz

Hey all, finally back with an update.

I sent a sample to ACX for review and I received a response in about 12 business days (they claim they respond in 7 days which seems ambitious). The sample I submitted was mastered using the Mastering Suite 4, and ACX Check said the noise floor was -63. I didn’t use any noise reduction outside of killing the obnoxious Blue Yeti mosquitoes since the minimum requirement is -60 noise floor.

Below is the response, minus the copy-paste greeting:


The engineer did not elaborate any more on what was unacceptable about the noise floor, but I took their advice and checked for bad cables. Turns out the cable I was using was 6ft, which–as far as I know–is too long for cleanly transferring data over USB. I hadn’t even thought about it as it came with my microphone, but I’ve now replaced it with a much shorter cord (1 ft).


There doesn’t seem to be any difference in noise floor after changing the cable, though. So I’ve used noise reduction and low rolloff EQ to combat the noise and it appears to work well (dropping the floor from -63 to -77). The frequency of the noise floor never goes above 130 Hz so the EQ seems effective. (I wonder if the noise is self-noise from the Yeti, because if I amplify the noise floor it sounds like white noise or TV static. No background sounds like cars or air conditioning or anything of the sort.)

I haven’t had time to review the example recordings that were attached in the email yet, but if anyone wants to see them let me know.

Upgraded my setup with 6 moving blankets that are quite heavy! I managed to get them directly from a manufacturer so they were half the price of getting them from somewhere like Home Depot. They have not actually reduced the noise floor very much, but the tone is noticeably more pleasant and the reverb is reduced.

Thank you very much for taking the time to help me out, Koz and Steve. Things appear to be good to go so far, thanks to you guys. If I manage to land any work I’ll be sure to hit that donate button.

Please let me know if there’s any way I can help out around the forum!

~Buswheels

ACX may be using a different technique to measure noise. We have had several people recently do everything right and still fail their noise tests.

There is one item which can cause problems. You can have special noises such as the Yeti Curse or Frying Mosquitoes which can cause your ears to bleed even if the background noise measures as conforming to standards. A close cousin to this is baby screaming on a jet. You can’t un-hear that.

The ideal background noise sounds like spring rain in the trees ffffffffffff or shshshshshshsh sound. That’s typical of the analog noise made by condenser microphones or external microphone interfaces. They do that normally. This is why you can’t record from across the room. Your voice has to be close and louder than the natural microphone noise.

I’d like to get into this more, but it’s getting toward bedtime in Los Angeles—the last, western-most support time zone that I know of.

Koz