FabFilter

Mp3 decoding often generates overs, especially if the mp3 file was encoded quite hot, that's what you're seeing. It's generally worse the lower the bit-rate and it depends on the specific source material, encoder, etc., but even at 320 kbps sometimes I've had to limit down to -1 or -1.5 dBFS before encoding to be able to produce an mp3 that wouldn't cause any overs on playback.

As an aside, this is not to be confused with the more or less "standard" -0.3 dBFS that you use when mastering to avoid ISPs (inter-sample peaks). I've seen quite a few discussions where both issues are talked about as if they were the same thing, but they aren't. The mp3 encoding-decoding issue is due to the fact that the process introduces distortion, and even if your original waveform doesn't go over 0 dBFS even considering ISPs, that distortion may well change its shape in a way that will make it go over, some times quite a lot, when it's decoded.

Sorry, one more thing: you say "even when the file isn't going past 0 dBFS". I don't know which DAW you're using but generally the way the waveform of an imported file is shown is one thing and the way the file is processed internally is another. If what's shown is what the decoded WAV would look like, it's never going to be over 0 dBFS by definition, but if the file is converted to 32 or 64 bits floating point for internal processing, then the sky is the limit! :-)

Cheers,

Sorry again, I just realized I didn't actually answer your question, I need more coffee! Depending on what you're doing with that track, this isn't particularly worrying since your DAW is using floating point internally, but if you do want to avoid it for gain staging reasons or whatever, you could simply put a gain plug-in before Pro-L, reduce it until Pro-L isn't showing any gain reduction at all and then you can safely remove Pro-L.

Cabirio,

Thanks for your elaborate multi-part response! You certainly cleared up two suspicions I had for awhile now. First, that DAWs use internal processing that can be different from what is seen. Second, the practical reason we reduce the overall output. I already export as WAV with -0.3/-1 and IPS when preparing for online distribution, but I never knew the real consequences of not doing so until now.

Do you think that when I convert the 320 MP3 to WAV and save it as a 32-bit float it also keeps the overs stored? Should I use 24/16 instead?

The reason I'm asking about this is because I create podcast mixes and I want to preserve the original track qualities as best as possible while keeping to my workflow. The decision now is do I want to decrease all the tracks uniformly, decrease the overall output to prevent possible future decoding distortion, or do both, but that will make the audio file too soft compared to my other ones.

Yes, I assume that saving them as float will keep the overs, but the potential problem with saving them as 24/16 WAVs is that you will inevitably have clipping, since those WAVs can't store anything above 0 dBFS, so in a way you're hiding the problem rather than solving it.

On the other hand I say "potential problem" because how much of a real problem that clipping is depends very much on the specific material. I use clipping (sometimes soft, sometimes hard) routinely on individual tracks when mixing, but also occasionally when mastering, before the limiter, just to shave off the highest peaks and make the limiter's job easier. Some times this sounds better to me than just limiting, some times not. As with most things audio, trial-and-error and let-your-ears-be-the-judge are your friends.

So, you may find that those 16/24 WAVs sound perfectly ok to you despite the clipping, but if not, what I would do if I wanted to avoid it at all costs but keep a reasonable volume would be to reduce the overall output level a bit (maybe up to 1 dB or so, whatever is acceptable to you), which will already get rid of a bunch of overs, and then use Pro-L to do the final hard-limiting on those that still manage to get through.

It's a compromise, so play with how much you reduce the level vs. how much you're hearing Pro-L kicking in (I stress "hearing" rather than "seeing it apply gain reduction on the meters") till you get a balance that works for you. You may well find that you don't have to reduce the level at all: the amount of gain reduction you can get with Pro-L before you start hearing it is simply amazing.

For music, I've found the "Hard Rock - Transient Priority bM" preset absolutely wonderful at keeping things nice and punchy, and not just for hard rock. I've also had good results with "Almost Clipping bM" and recently on a big band-type piece "Jazzy Pop - Dynamic, Spacious and Loud" did wonders. If it's mostly speech you're dealing with, you may want to try the "Vocal - Transparent MTK" one.

I have to say that the Pro-L presets are really, really good, and don't be misled by the names, many are much more generally useful than the name might suggest, so experiment and you may get very pleasant surprises. You can also tweak them of course, but more often than not, once I've found one that gives me the sound I'm after, I can simply adjust the input gain to taste and I'm good to go.

Cheers,

Thank you! That's awesome!