Pro-L 2 by FabFilter

[mutantdog]

Question regarding ISPs and ReplayGain then...

Say I have a lossless audio file with ISP overshoots of around 2dB, I have the exact same audio file, only rendered with the volume reduced by -2dB. After applying ReplayGain my media player (Foobar2000) will output them at the exact same volume. So in this example is there any real difference betwen the two? Surely for me they should sound exactly the same, right?

It depends on the DAC and the analogue design. Lets say that the analogue section does not have the headroom to accommodate the 2dB overs of the first file. Then, it will clip. If the 2dB gain is added to the second file then it will also clip because it has to pass through the same analogue section that doesn't have the headroom (whether or not the gain is added in the digital domain or the analogue domain).

There is no free lunch.

So if the replaygain is reducing the volume of the louder file by 2dB and I'm not clipping the output of my audio interface, then they should effectively be the same for my own personal purposes..?

[plexuss]

Question regarding ISPs and ReplayGain then...

Say I have a lossless audio file with ISP overshoots of around 2dB, I have the exact same audio file, only rendered with the volume reduced by -2dB. After applying ReplayGain my media player (Foobar2000) will output them at the exact same volume. So in this example is there any real difference betwen the two? Surely for me they should sound exactly the same, right?

It depends on the DAC and the analogue design. Lets say that the analogue section does not have the headroom to accommodate the 2dB overs of the first file. Then, it will clip. If the 2dB gain is added to the second file then it will also clip because it has to pass through the same analogue section that doesn't have the headroom (whether or not the gain is added in the digital domain or the analogue domain).

There is no free lunch.

So if the replaygain is reducing the volume of the louder file by 2dB and I'm not clipping the output of my audio interface, then they should effectively be the same for my own personal purposes..?

Yes. It should not clip. If by some odd chance it does clip, the clips would likely be short and sporadic and not be audible. It's hard to predict what will happen in the analogue domain because there are so many variations, but theoretically you shouldn't have any clipping in that scenario.

In fact this part of the EBU R128 spec which requires audio to be reduced in amplitude to prevent peaks from reaching 0dBFS so all clipping is mitigated. This is for broadcast. But for music, the R128 standard is rarely used because it defines a loudness target of -23 LUFS. I have never seen any audio mastered to that low a level and that includes a lot of high def digital audio taken straight off the master tapes in quieter geners like jazz, folk and classical. But the approach of the R128 spec is essentially what you are suggesting: reduce the gain to keep the peaks under the ceiling.

https://tech.ebu.ch/docs/r/r128.pdf

[mutantdog]

Thanks. I've been analyzing a bunch of random tracks from my music collection and noticed that much of the "louder" stuff has these sorts of overshoots. Since its compensated by the level reduction of ReplayGain, it's probably not much to worry about.

I think overcompression and heavy limiting has become such a staple part of the sound of modern music that even as we move towards a streamlined loudness normalisation standard, there will remain a desire for more traditionally loud masters. A little bit of clipping can make a flabby kick sound better for example. Also, there is little benefit in having a good dynamic range if it's just gonna get cranked up loud and distorted on weak audio systems (eg: your average car stereo or cheap hi-fi). As someone else mentioned, things like YouTube aren't exactly good for consistent loudness either, although Spotify is much better.

[plexuss]

Thanks. I've been analyzing a bunch of random tracks from my music collection and noticed that much of the "louder" stuff has these sorts of overshoots. Since its compensated by the level reduction of ReplayGain, it's probably not much to worry about.

I think overcompression and heavy limiting has become such a staple part of the sound of modern music that even as we move towards a streamlined loudness normalisation standard, there will remain a desire for more traditionally loud masters. A little bit of clipping can make a flabby kick sound better for example. Also, there is little benefit in having a good dynamic range if it's just gonna get cranked up loud and distorted on weak audio systems (eg: your average car stereo or cheap hi-fi). As someone else mentioned, things like YouTube aren't exactly good for consistent loudness either, although Spotify is much better.

You make very good and relevant points. And it's true. It simply comes down to one's own "audio aesthetic". For myself, I want the highest quality audio I can create. So this is why I concern myself with things like loudness and peaks and the detailed nuances of them. But others don't care and aim for the loudest audio they can create while still retaining a sense of the music - clipping, distortion and PLR are not a priority. Whatever floats your boat.

[mutantdog]

Don't get me wrong, I'd love to see a bit more dynamic range return to music. It immediately puts me off when i watch the preview trailer for Pro-L 2 and it seemed really focussed on showing how loud it can go. Compared to Barricade 4 which positively excels on more dynamic material and I know what I'd rather use.

[Ivan_C]

I think everybody here doesn't understand something very important about true peak limiters and ISP metering. The EBU standard involves 4x oversampling, and oversampling works always this way :

upsampling -> upsampling artefacts filtering -> processing (either the limiter or meters calculus) -> filtering for removing aliasing -> downsampling

As you may know, whatever the filter coefficients, filtering processing isn't transparent on the phase (IIR filters have often a nonlinear phase response, and linear phase FIR filters add a strict delay), which means filtering changes the audio waveform shape.

Because of that, we can say that the oversampling process itself changes the position and the number of ISP, whatever the processing embedded inside does.

Even worse, if you have a limiter which is supposed to catch these ISPs, you can only do so inside the oversampling process, which means you have no control over the additional ISPs which might happen in the last filtering process before downsampling. And then, if you have a meter to display where the remaining ISPs are, and that this meter is not inside the same plug-in, it will probably display something more or less wrong since the calculus involves more oversampling/filtering, so different ISP locations and numbers.

And if you think it's not enough, oversampling algorithms structure is not standardized either, you can find as many algorithms and filters types + coefficients inside as there are limiter / meters plug-ins, so every single one is going to change something there in a different way... I think there is a standard for these filters in one of the loudness metering test process, but since there is no reason to follow it in the process oversampling...

So it's not surprising at all to see some differences between the plug-ins. The only thing surprising me actually is that they are able to display more or less the same thing with a tolerance lower than 0.6 dB !

Anyway, I guess a few things are still not clear to me since I have never coded a professional limiter yet, but what I can say is that we shouldn't care too much about exact ISP detection because such a thing doesn't exist in my opinion. And I think Pro-L2 looks very promising

[Guenon]

For myself, I want the highest quality audio I can create. So this is why I concern myself with things like loudness and peaks and the detailed nuances of them. But others don't care and aim for the loudest audio they can create while still retaining a sense of the music - clipping, distortion and PLR are not a priority. Whatever floats your boat.

In my opinion it's also important to note that it's not an either/or thing. I get the feeling there's a rhetorical device at play when I read that someone wants "the highest quality audio" and says that "others don't care and aim for the loudest audio they can create" -- it's not like it has to be either highest quality possible or loudest possible.

There are realistic cases where you want to have a certain degree of loudness, preserving nicely detailed dynamics to taste but still noticeably restricting the overall dynamic variation. I don't mean squashing it into a block of sausage , I mean tastefully bringing the body of the music up in relation to the peaks. What this taste might be like depends on the producer and the genre.

To tell you the truth, what interests me most in this topic, these days, is the way ear specialists have warned music listeners about loud headphone use for years and years; especially teens who enjoy blasting high-energy genres while on the go. The problem with such headphone use is, in order to have the body of the music loud enough, they like to crank the volume way up so that they hear the bass and midrange very prominently while riding the bus or where ever. If the music is tastefully mastered, to genre, there are still nice dynamics and all, but at the same time their ears aren't being bombarded with extremely high volume peaks that, in genres like that, would occur all the time in super-dynamically mastered material listened with the same subjective loudness level and feel.

Anyone in this field can test it by doing a version of such a track to something like -16 LUFS, snappy high frequency percussive transients peaking high, and then doing a more genre-conscious version of that track with the body of the music brought up in level more prominently, by controlling the dynamics more. You can see how many dB more you are inflicting upon those ears at the upper range when going with the extremely dynamic setup. I wish someone with headphone measurement equipment would do a comparison of the real world difference this way, having test tracks that random young listeners would play on their own equipment, and then measuring the actual levels coming from the phones, especially the upper end of the spectrum.

[mutantdog]

That's a really good point there. I suspect my iPhone probably clips the output of extreme peaks, which is exactly what you get on a very dynamic track using ReplayGain.

As subjective as it is, i tend to find an integrated value around -10 LUFS ± 2 is a pretty good range to be in as a compromise. Of course it varies a great deal on the style and sound of the music itself. The limiter algorithm and whether or not clipping is used plays a great deal in enhancing the perception of dynamics too.

[plexuss]

Hey. I agree with everything you guys have said in the last few posts. The ISP issue is as you say, not an absolute both in terms of detecting them and mitigating them. I find that the few true peak limiters I have are useful because they all seem to take a different approach to mitigiating ISP clipping. So, they have a different sound mostly attributed to sound of transients. My workflow always employs at the end of the mastering chain, a true peak limiter, loudness meter and ISP velocity meter. Typically I will use ISL2 or Limitless, Mastercheck pro and RoundTripAAC. I target -16 LUFS and watch the PLR too to make sure its reading high (> 9 PLR on average, but ideally > 12) and absolutely no sample or ISP peaks on the rendered file. I might adjust loudness up or down depending on what I see for ISP clips on a simulated 128k stream but I mainly care about the rendered file. Then, for Soundcloud, I use Myriad for SRC and conversion to 44/16 which I upload to Soundcloud. I find this gives pretty good streaming quality and of course the rendered file is free of clipping. I will review the files on both an expensive DAC (Lavry) and cheaper DACs like in laptop and cellphone to make sure everything is working well.

Here's my soundcloud if you want to hear the results of this workflow. A track like Crims Kingdom or The Lighest Rose or IFOOMW are good example of very high PLR tracks. I then check the soundcloud stream with RoundTrip to see how bad the clipping is. they do clip a bit on transients even with a -16 LUFS target and all that attention to ISP etc. This is a comprimise I am willing to make.

I have a note on the page (on the right) about why my loundess is lower than others. I've had people ask me if there is a problem because my files are lower volume than what they are typically hearing on soundcloud.

If you want the 44/16 file of any of the tracks for comparison let me know by PM.

https://soundcloud.com/musicofplexus

[Guenon]

I have a note on the page (on the right) about why my loundess is lower than others. I've had people ask me if there is a problem because my files are lower volume than what they are typically hearing on soundcloud.

If you want the 44/16 file of any of the tracks for comparison let me know by PM.

Yeah, Soundcloud is problematic as it still plays everything as is... What you upload is what the listener gets. Btw, have you noticed a difference when uploading higher sample rate material on Soundcloud and letting them do the conversion, versus doing a final 44/16 version yourself and uploading that? Haven't really checked out the difference with their own conversion.

I think I should probably say something about Pro-L 2 at this point to keep this somewhat focused on that, too . Dang... Well, I checked out the video and new features and was pretty sold already. I'm quite sure I'm going to upgrade my license in the near future.

[plexuss]

[snip...]
I think I should probably say something about Pro-L 2 at this point to keep this somewhat focused on that, too . Dang... Well, I checked out the video and new features and was pretty sold already. I'm quite sure I'm going to upgrade my license in the near future.

Soundcloud: I took a leap of faith that Myriad would do a better job of SRC than Soundcloud. Myriad SRC is by Goodhertz which I know of and they do good audio work in other areas.

Por-L2: I will upgrade too at some point. I like it as a character limiter and I like the usability.

[10bd01]

Update from FabFilter:

Hi guys,

After reading the full KVR thread (see link above), we decided to dig as deep as possible, and explain possible differences between the AAC RoundTrip tool, Pro-L 2 and other plug-ins with TP metering. What we've found, was actually quite interesting:

Apple's AAC RoundTrip as used for Mastered for iTunes etc), simply seems to use the example detection filtering (4x oversampling) as described in the ITU-R BS.1770-4 specs. This method is generally considered to be sub-optimal, and can give quite significant under- and overreads. As the spec already implies, "Higher sampling rates and over-sampling ratios are preferred".

In Pro-L 2, we've chosen to use 8x oversampling and steeper filtering for our true peak detection, resulting in much more accurate true peak readings.

Some plug-ins seem to have implemented that same, simple but sub-optimal example filtering from the ITU spec, which explains why their readings would match Apple RoundTrip's levels. However, most plug-ins, like Pro-L 2, implement a better true peak detection, and because of this, you might see 'overs' in Apple's RoundTrip AAC tools.

Having said this, we did find a small issue with Pro-L 2's true peak detection when running at higher samples rates (96 kHz and up), which could cause the readings to be a bit less accurate than we intended. Of course, we'll fix that in an minor update soon!

Cheers,

Floris (FabFilter)

[plexuss]

[snip...]
Apple's AAC RoundTrip as used for Mastered for iTunes etc), simply seems to use the example detection filtering (4x oversampling) as described in the ITU-R BS.1770-4 specs.

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. Either way, its all just a guess. If RoundTrip throws more warnings of clipping than "it should" then it stands to reason that reducing levels to reduce RoundTrip clip warnings would ensure less risk of clipping over a meter that is less sensitive to estimating clipping, such as Pro-L2.

No worries FF, we know that ISP clip detection is an estimation and that no one can make an estimation for all cases (which is why its an estimation). The fact is with some ceiling in Pro-L2, it will suppress ISP clipping (as far as it can estimate them) and is a worth contender in the true peak limiter world.

[Magnut]

Update from FabFilter:

Hi guys,
...
Cheers,

Floris (FabFilter)

Great work, FabFilter nailed it again.

[10bd01]

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.