Why is my limiter responding to other people's music played at 0db?

[cron]

... and how to deal with it?

I'm currently comping a radio show which of course involves playing music back. I'm doing a wee bit of mixing and a few outrageous transitions on otherwise unmixable material, so I've got a limiter on the master to deal with that and reduce the burden of fader juggling.

I thought the limiter would only kick in during the transitions, but I'm seeing it during 'standard' playback too. No processing, no timestretching, no conversion from a lossy format, no nothing. Just the lossless tracks, purchased from an online store, played back at -0dB. Intuitively, I'd expect 'bit-perfect' reproduction under those circumstances, but I'm seeing 1dB or more of gain reduction at times.

I'm using Barricade CM, but I gave Limiter 6 (the old free version) a spin and saw the same thing. Barricade does it regardless of whether ISP is on or off.

Is this expected behaviour? I'd have thought there was nothing left to limit when you're just playing back an already mastered track. Especially with ISP switched off. Can anyone help me compute this in my brain.

The plan was to export at -0.3 or thereabouts to (at least somewhat) mitigate overs caused by the subsequent lossy conversion on Mixcloud and my station's player, but dropping down -1.5db or more to avoid these weird overs is getting into really noticeable territory. Obviously I'll do it if I have to. Keen to understand what's going on though.

[papawise]

Good question.
I expect an answer too. My limiters or different kind of plugins with a type of VU meter also shows red for mastered tracks with no effects, warp or edit applied.

[aciddose]

It depends upon how the peaks are measured. There is likely a filter which is likely an approximation which yields peak measurements up to 3 dB or more over the discrete (digital) peak level.

If you want to try Xhip Limiter it will do what you want because it measures the discrete peaks. One limitation of Xhip Limiter is that it can allow some leakage due to a minor bug in the peak detection algorithm (to be fixed soon). Most of the time however (if I hadn't mentioned it) it won't be noticed.

(If the GUI is too small, I've got a resizable alpha version available.)

[cron]

Many thanks Aciddose. I'll give your limiter a shot. I'll gladly take the odd clip during transitions over constant audible ducking artifacts. Cheers!

[Winstontaneous]

Interesting. I was using Adobe Audition CC2017's loudness analysis tool to compare some of my recent tracks to house-y references bought through iTunes. Except for one Deadmau5 song (he masters himself and ceiling was 0.3 dB with no overs/ISPs) all of the purchased, supposedly "high quality Mastered for iTunes" tracks had multiple peaks above 0dB (some tracks had 100s or 1000s of them).

aciddose - these were actual statistics computed by Audition on each audio file - is it safe to assume they are accurate, compared to the approximation filter used by OP's limiter? Thanks!

[aciddose]

Files encoded in formats other than integer PCM may have arbitrary levels.

For example .flac format uses either 8-bit, 16-bit or 24-bit integer PCM values and so it is impossible for the data itself to be greater than or equal to 0 dB.

Using 16-bit the largest sample is +32767 while 0 dB is either 32767 (the better option IMO, although usually 32768 = 0 dB) or 32768. Where 0 dB is 32768, 0 dB has no representation in the format at all. This is why a level = 0 dB is considered clipped. (I hate this in meters; it should always be OPTIONAL but almost never is.)

Where 32767 = 0 dB, the max level is 0 dB.
Where 32768 = 0 dB, the max level is -0.265 thousandths of a dB.

These files can carry a "gain correction factor" however that sets the total gain to +3 dB or similar which could then produce levels approximating +3 dB. If your files have a gain correction other than zero (do nothing) you should set it to disabled if possible so the results you get are accurate without a hidden gain adjustment happening that you're unaware of.

MP3 and other lossy formats do not use integer PCM and so may produce arbitrary levels. Floating-point PCM format may also include arbitrary levels like +9000.1 dB without even needing a gain correction factor to cause it.

It is also possible your "accurate" meters are actually measuring approximate inter-sample peak levels and so are technically "inaccurate" when we're talking about discrete peaks.

[aciddose]

Another possibly may be that your host applies panning to each track. Normally with an equal power pan law (-3 dB) the "center" will sit at -3 dB with far left/right at 0 dB. Often there is an option to normalize the panning law to get 0 dB at the center position with +3 dB at far left/right.

Another type of filter you may not have considered is the re-sampling (interpolating) filter used where the input is at a different sample rate than the output. You must use exactly identical input and output sample rates in order to ensure an interpolation is not being applied.

In addition to that there may be dithering applied to the signal.

It's hard to say but if Xhip Limiter reports peak levels beyond 0 dB you can be 100% certain the discrete input to the plug-in is beyond 0 dB. I'd be curious to hear whether that is the case. If so you'll need to look for where the filters/dither/panning/normalization are applied to your input.

The solution in the case of interpolation would be to re-sample the tracks yourself ahead of time using an interpolation that does not increase the discrete peak level. Such an interpolation would probably be quite audible if your target rate were below ~96 kHz and so this would be a trade-off between peak level, audible level and spectral content.

[niektb]

What you should know about Barricade CM (I use it too and I love it ) is that it's not just a limiter. The signal chain also features a compressor placed before the limiting stage. Maybe that's where the gain reduction is coming from? I think Barricade CM's manual has some more info about this...

[cron]

Many thanks for your additional thoughts all. I'd completely overlooked sample rate conversion as a possible source of clips and also hadn't taken into account the possibility of a 'ReplayGain' type correctional factor messing with things. I'm inclined to think Ableton Live applies this correction factor given that there's a tag which seems to fit the bill in the metadata, and toward the end of comping I stumbled across a few tracks I could actually boost by a dB or two without clipping. Thanks for helping me unravel that one.

Thanks also for the mention of Barricade containing a compressor in the signal path. Looking through the manual I think this only applies to version 4 of Barricade, whereas Barricade CM appears to be based on version 3. Good to know there's an update that makes springing for the full plug seem much more appealing though.

edit: actually, I'm seeing a few places report Barricade CM has both a compressor and peak limiter in there. There's no mention of a compressor in the v3 manual (couldn't find a manual for the CM version on my drive) but I wonder if that has anything to do with the algorithm mentioned in the v3 manual that can supposedly distinguish brief from sustained overs...