One of my voice actors has turned in two files that sound fine to me but hit the “noise floor too low” error on ACX Check. I’m having trouble finding a fix. Every article I find is about reducing the noise floor, not increasing it.
I created a brownian noise track, but when I ran ACX Check, it paused halfway through to give me a FAIL then finished and gave me a second FAIL.
I figured it saw the noise track as a separate recording, but couldn’t figure out how to merge the noise directly into the voice recording. Instead, I threw out the right channel of the voice recording and made the noise track a new right channel.
I got one result from ACX Check, but it was still a failure.
There’s a better way, isn’t there? Is there a process I don’t know about? A set of steps written for a total neophyte?
I just want to create the faintest room tone so I can pass this friggin’ test!
This is the wrong approach. You need to find out why the noise floor is measuring too low, then fix that. Most likely it is just one, or a few tiny sections of the file that are faulty. Adding noise to the entire track will make it sound worse, and “what it sounds like” is far more important than measurements.
What is the full and exact message? How low does it say the noise floor is?
Most likely it is just one, or a few tiny sections of the file that are faulty.
ACX Check measures the whole show and then reports the quietest portion. So if you created an absolutely perfect book chapter but then jammed on a couple of seconds of Generate > Silence, the whole chapter will most likely fail Too Quiet. As an experiment, select the middle third of one of those “faulty” chapters and analyze that. A repair might be just getting Room Tone from the show and copy that on the ends of the chapter instead of Generate > Silence.
On the other hand, It’s possible to use extreme noise gating and other noise reduction techniques so the whole chapter has dead silence between words and sentences. That’s the one that the Brownian Noise was designed to “fix.”
ACX doesn’t like gating and extreme noise reduction because those tools and techniques can easily create voice sound damage. They’re hard to manage and ACX hates distractions.
Nobody should be delivering audiobooks in stereo. It’s wasteful, ACX recommends against it, and in some cases as above it can actually cause problems. Deliver in Mono.
@ Koz
A long time ago I suggested a more “advanced” version of ACX-Check that added labels where problems occurred. At the time you thought that was too complicated and wanted a single “Pass / Fail” response from the analyzer. What are your thoughts now that we have trialled ACX-Check.ny for a few years?
(perhaps better to start a new topic for this discussion).
For what it’s worth, I edit the audio in Zoom recordings and the background noise in them has been cut to a level of -155dB by Zoom.
So I have recorded a background noise track in a quiet room of my house. Then I use AutoDuck to duck the noise track I created based on the audio track (no noise when there is speaking and noise when there is quiet). Then I mix the ducked noise file with the original audio. In my case, this brings my noise floor up from -155dB to somewhere in the -70’s dB. But I’m not using my audio for any audio books, I’m just trying to get a consistent audio level in my files.
I have a Macro that does the ducking and mixing for me.
As an experiment, select the middle third of one of those “faulty” chapters and analyze that. A repair might be just getting Room Tone from the show and copy that on the ends of the chapter instead of Generate > Silence.
Okay, tried it on several, smaller sections. Except for a bit at the beginning, they all failed for having too low a noise floor. The value of the noise floor varied.
Nobody should be delivering audiobooks in stereo. It’s wasteful, ACX recommends against it, and in some cases as above it can actually cause problems. Deliver in Mono.
I thought that was what I was trying to do with adding brownian noise. ACX Check doesn’t like having third tracks, but as someone else said, I could export and reimport and check.
The effect is detecting a part of the selection that is “totally silent”.
The most likely reasons:
You have included some empty track space when applying ACX-Check (empty space will be detected as total silence, aka “-inf dB”).
You used the “Silence” effect, or generated some silence into the track during editing.
Some absolute silence was introduced into the track during editing by some other means.
There was a “dropout” (a gap) created in the recording due to a recording problem.
The first thing to do is to try and find where the total silence(s) occur. If it’s only one place then it should be easy to fix by editing.
See this post for how to detect where the silence(s) occur: https://forum.audacityteam.org/t/noise-floor-too-low/60410/5
I am currently attempting to check each silence by doing an ACX Check on every ten seconds of recording. Part of my problem is that ACX Check will tell me a section is a problem, but when I go back to check it again, it no longer says it’s a problem.
It seems that if you created some Brownian noise with an amplitude of .0005 (or less), that it would give you a noise floor in that track of -80 to -90dB. Then you could mix that track with a copy of your audio and see what that does for the overall ACX testing.
It might be a very good idea to work out why the problem occurred so that it can be avoided in future rather than just papering over the cracks.
It’s easy to fool the ACX robot algorithm, but much harder to fool the human review that comes later, so “faking” a high quality recording could be a big waste of time.
Blindly putting the puzzle pieces together: The artist has Skype or Zoom running in the background of their machine and their voice is going through Zoom processing. As you pointed out, it sounds OK, but just won’t pass technical tests.
