Pro-L 2 by FabFilter

[plexuss]

Another confirmation that RoundTrip adheres to the ITU spec. Whether its sub-optimal or not is a matter of opinion.

Is it? Or is it a matter of testing?

I think it's also worth asking at this point: are you hearing the clipping or just seeing it on your meter? Because if the meter is using a sub-optimal specification you could be viewing clipping with your eyes that isn't actually happening to you ears.

Pretty much everyone I've consulted that has tested and uses RoundTripAAC has confirmed it is legit and useful. This include professional mastering engineers through the mastering engineers and loudness facebook groups. For example Ian Shepard and Bob Katz are on those pages. As well as numerous other professionals.

Remember that the whole issue is trying to determine what will happen in the analogue domain. This isn't an issue if your peaks are lower than 0dBFS. Not AT 0 dBFS but lower than it. But in order to get the loudness that still many people crave, the gain of the whole track has to rise pushing the peaks at or over 0 dBFS. this is where clipping occurs and why we want truly true peak meters, a good sounding approach to managing those clips and some way to estimate if there will still be any clipping in the analogue domain.

It's almost rocket science.

Since its an estimation of the risk of clipping in analogue and there are so many DAC approaches, there won't be one way to do it.

I think RoundTripAAC is the best tool for the job for the reasons I cited: it appears to be based on the ITU spec for peak and clip detection and it measures clip velocity. No other meter I know of does these two things. But if a better meter comes out I will use it if I can. What would make it better: confirming that it uses the ITU spec at the very least and if it uses a "better" algorythm, then a description of the rationale for why it's better. Maybe a better GUI. Not much else. RoundTripAAC does the right stuff if you are trying to reducing clipping in the analogue domain.

In terms of subjective evaluation: this is why clip velocity is so important because one single clip is not going to be (very) audible (I always have to make room for people who believe they can hear one sample of clipping - that's the nature of audio). A stream of clips in a short duration become more audible. The problem with subjective evaluation is that audbile clipping imparts a kind of broadband "haze" to the audio and it can difficult to hear it as a problem. Until you AB with one without clipping. Just because a track may SOUND good without comparing it to a version that sounds better, doesn't mean its the best version. Audio is subjective and our brains often react in the opposite way we expect so AB'ing and metering for problems is the way to do it.

And, as has been discussed here, you don't HAVE TO eliminate ISP clipping: some genres actually sound good with some clipping. Clipping causes yet another kind of distortion. There are plugins just to create good sounding clipping. Just do what you think sounds best.

For my work I prefer not to have clipping like that. I do not like the sound of it in my music and audio work. So this topic and these tools and considerations are imporant to me. That's why I've spent so much time to learn about all this. YMMV

[mutantdog]

Surely the best solution for FabFilter would be to have a choice of true peak detection algorithms. By the sounds of what they describe, their own one should be better but the ITU standard exists for a reason. Anyone working in the sort of industry where strict rules are applied will most probably benefit more the ITU spec. This is a high-end product with a professional price tag at the end of the day.

[10bd01]

Thanks for the response, plexuss.

I appreciate the effort you've put into researching this, posting the video, and answering in intelligible thread responses.

[camsr]

A funny thing about ISPs is they get less prevalent at higher sample rates, when the signal is bandlimited. At 96khz, you may experience less ISPs with a lowpass on the masterbus at 20khz.

They have never seemed that problematic in the first place IME. I can think of a few circumstances where ISP detection is better than none, one being a synth waveform with a lot of ringing (gibbs phenom) that spikes the sample peak level at seemingly random (truly aliased) intervals. In that case the ISP detection would offer a more coherent peak level for the limiter to act upon. I don't think i've ever heard ISP clipping from the anti-aliasing filter of a DAC before. In fact any good DAC would have 6dB of headroom (with the possible compromise of a 6dB increase in noise floor) for the peaks the anti-aliasing filter does make. Simply because those peaks are louder than the noise, and more appearant.

[dangayle]

Another confirmation that RoundTrip adheres to the ITU spec. Whether its sub-optimal or not is a matter of opinion. The industry spec is the ITU spec. If FF has a "better way" that's fine. but it may or may not be to spec.

From the spec:

recommends... that when an indication of true-peak level of a digital audio signal is required, the measurement method should be based on the guidelines shown in Annex 2, or on a method that gives similar or superior results.
...

Annex 2
The 4x over-sampling filter increases the sampling rate of the signal from 48 kHz to 192 kHz. This higher sample rate version of the signal more accurately indicates the actual waveform that is represented by the audio samples. Higher sampling rates and over-sampling ratios are preferred (see Appendix 1 to this Annex).

[10bd01]

...

Annex 2
The 4x over-sampling filter increases the sampling rate of the signal from 48 kHz to 192 kHz.

Wait...does this mean if I'm at 44.1 and use oversampling it will over-sample to 192? Doesn't this add an element of distortion upsampling it and then downsampling it back to 44.1?

[plexuss]

...

Annex 2
The 4x over-sampling filter increases the sampling rate of the signal from 48 kHz to 192 kHz.

Wait...does this mean if I'm at 44.1 and use oversampling it will over-sample to 192? Doesn't this add an element of distortion upsampling it and then downsampling it back to 44.1?

That would be quantization distortion. The SRC would be done in the ISP detection "circuit" and not be imparted on the audio that the is being processed on the output (we hope).

[10bd01]

That would be quantization distortion. The SRC would be done in the ISP detection "circuit" and not be imparted on the audio that the is being processed on the output (we hope).

Thanks! Makes sense. So basically it uses the conversion for calculation but not in a manner that touches the original signal (my understanding/translation).

[OzoneJunkie]

I ended up upgrading to L 2. Is there any reason, other than old projects, to keep L 1? Has 2 removed any features or added any latency that 1 doesn't exhibit?

Thanks for any/all replies.

[plexuss]

I ended up upgrading to L 2. Is there any reason, other than old projects, to keep L 1? Has 2 removed any features or added any latency that 1 doesn't exhibit?

Thanks for any/all replies.

So far I am seeing no reason to keep L1 except as you said for backward compatibility. But I haven't dont any comarison between the two.

[e@rs]

The old algos have the same latency in the new version with True Peak Limiting disabled.

[rocolin]

I've been doing some tests with different limiters, kind of technical tests. I mixed two or three sines of different frequencies, one goes from zero to 32767(0 dBfs) @ 22050Hz, to see how the limiters react. I looked at the resulting waveforms (bounced) and I noticed that Limiter6 GE and Pro-L2 seem to keep the upsampled-downsampled signal, not the scaled (TP -limited) original. You can see how the signal extrapolated to a few samples before the original signal.

I think (IMO) it's not a big deal, up sampling the signal 8x just extrapolates a few samples before it begins (seven or eight?). I think it's harmless. It's probably better if the original signal has discontinuities at the start, since it takes a zero and interpolates with the first sample; creating a smoother signal, just a guess.

Limitless, Tracklimit, and Barricade4 seemed to keep the original signal.

[OzoneJunkie]

So far I am seeing no reason to keep L1 except as you said for backward compatibility. But I haven't dont any comarison between the two.

The old algos have the same latency in the new version with True Peak Limiting disabled.

Ah, roger that. Thanks, both of you. That helps

[mutantdog]

I've been doing some tests with different limiters, kind of technical tests. I mixed two or three sines of different frequencies, one goes from zero to 32767(0 dBfs) @ 22050Hz, to see how the limiters react. I looked at the resulting waveforms (bounced) and I noticed that Limiter6 GE and Pro-L2 seem to keep the upsampled-downsampled signal, not the scaled (TP -limited) original. You can see how the signal extrapolated to a few samples before the original signal.

I think (IMO) it's not a big deal, up sampling the signal 8x just extrapolates a few samples before it begins (seven or eight?). I think it's harmless. It's probably better if the original signal has discontinuities at the start, since it takes a zero and interpolates with the first sample; creating a smoother signal, just a guess.

Limitless, Tracklimit, and Barricade4 seemed to keep the original signal.

That's interesting. I've run audio (pre-master mixdowns) through both Barricade 4 and Limiter6 GE, in both cases with the audio at a low enough level to not trigger any gain reduction. Without doing any serious analysis, the rendered waveform using B4 was visibly different to the original while L6GE appeared to be identical. To my ears L6GE sounds more transparent in general use while B4 has a more distinctive character.

[10bd01]

rocolin, MutantDog,

If either of you are so inclined to include screenshots, etc. it would be helpful. I was wondering how Tokyo Dawn figured into this scenario.

So according to you, rocolin, plexuss' hopes are dashed? It upsamples, downsamples, and distorts the signal via the sampling?