I recorded a short audio clip and passed it through RMS Normalize at -20dB then passed it through Limiter with a Soft limit at -3dB. An ACX check produced a failing Noise floor value of -56.7dB. I subsequently, on a hunch, went through the audio clip and silenced any breath sounds between phrases and sentences. Passing the audio clip through an ACX check again resulted in a Noise floor pass of -157dB.
Is it the case that ACX check was considering the breath sounds as noise? Obviously, part of my vocal training will focus on reducing breath sounds to achieve ACX compliance because I’m the weakest link in the chain. My room doesn’t register on the recording level meters (for a minimum of -60dB) at my chosen 2i2 gain setting.
Audacity-2.3.2 (from developers’ PPA) on Linux Mint 19.1 Cinnamon w/ kernel-5.0.15
Scarlett 2i2 USB audio interface
R0de NT1 condenser microphone (L)
AKG D5 dynamic microphone (R)
The noise-floor measurement is; the quietest 0.5 seconds found in the selection.
If all of the silences include breaths, then it will not be a true measure of the noise floor.
When selecting the region to test, ensure that the selection includes at least on bit, longer than 0.5 seconds, that is true “room noise” (no breaths).
then passed it through Limiter with a Soft limit at -3dB.
I wouldn’t do that. The mastering recommendation is soft limit to -3.5dB. Nobody is going to miss that half dB, but the conversion between WAV and MP3 can cause errors that large. Wouldn’t you be upset if a quarter of your submissions failed because of Peak conformance?
Passing the audio clip through an ACX check again resulted in a Noise floor pass of -157dB.
You know none of that is real, right? The sound inside Audacity isn’t 16-bit. It’s 32-floating to allow for extended filtering, effects and corrections. The minute you get the show outside Audacity, you’re going to snap back to the “real” -96dB limit. And that’s not real, either. Nobody can hit that in normal production and ACX sometimes looks for numbers like that to bust you for “Excessive Processing.”
As Steve above, you can have a terrible chapter presentation and just put 3/4 second of pure silence in there anywhere to fake out ACX Check. That’s why after the actual ACX Robot passes your work, a human listens to it.
ACX Check is the shortcut. There is the classic/legacy measurement technique from the Bad Old Days.
And just given the question title, the noise measurement is not conditioned. This noise measurement will measure “stuff” in your show whether or not you can hear it. Earthquakes and bat love calls all count. That’s one of the reasons for the Equalization step at the beginning of ACX Mastering.
If you wanted to reference health noise violations in industrial applications, it’s still sound, but it’s measured completely differently. Sometimes Audacity users try to conflate those two and it’s not easy, convenient or cheap.
Actually, I’ve just checked the code and it is “conditioned” a little. It is passed through a 10 Hz high pass filter prior to measuring the noise level.
So the noise level will include at calls, but will largely ignore Earthquakes.
Can we assume that’s to keep it from responding to DC Offset?
I wonder if the ACX Robot is set to catch DC offset…? I think I would rather catch it here than to wait two weeks while the ACX Testing Service reports back. Even worse they might report “Something Wrong” with a recording.
Thinking about this requires extra coffee. DC would make clean editing a nightmare, but only if you edited outside of your show recordings.
Would ACX even care? We know they produce many different products, but how many of them require clean DC levels?
I would remove DC from every submission and just don’t tell anybody.
The minute you get the show outside Audacity, you’re going to snap back to the “real” -96dB limit.
Actually with integer formats (like 16-bit WAV), you can get “dead digital silence” which is minus infinity dB. I think you knew that, Koz. As soon as there is some non-zero signal, you start getting quantization noise (I think -96dB at 16-bits). And yeah, if you get digital silence (or -96 or -150dB) it’s obviously been “artificially processed”.
With MP3 I don’t know if there’s a limit. MP3 uses floating point but it doesn’t store the individual samples and it “tries” to throw-away stuff you can’t hear. But, “stuff you can’t hear” is mostly based on masking and I’m not sure if there’s an absolute level where it starts acting like a noise gate.
Testing with Audacity:
If dither is off (disabled), then like 16-bit integer formats, the audio is gated at -96 dB.
If dither is enabled, then like 16-bit integer, audio data is retained below the -96 dB barrier.
I asked the question because I assumed that ACX Check has to make a determination of what the noise floor is. So, I guess my clip didn’t have any 0.5s period of silence and one of the breath-containing intervals was taken instead. Silencing the breaths is, obviously, artificial and would present an absolute minimum noise floor. I did it to determine if the breath sounds were being interpreted as noise, not in an attempt to subvert the ACX submission requirements. I would aim to achieve those requirements by careful adjustment of my interface gain and applying appropriate RMS Normalization.
@koz The application of -3dB of Soft Limit was just an initial test value. In fact something below that value is required by ACX Check in order to achieve a -3dB peak level. I think I recall establishing -3.1dB at least but more headroom, as suggested, is desirable. I am not specifically intending to create audiobooks but considered the ACX requirements a good starting point for all of my audio experiments.