Pure Dynamics

[regorxxx]

https://hydrogenaudio.org/index.php/topic,128423.msg1068911.html#msg1068911

Very bold claims about the superiority of PD in the algo description. But it seems to fail to produce even a little bit consistent results. A phase shift, which doesn't change perceived audio at all, changes the PD result. Lossy compression, which is transparent to ear and of course doesn't change dynamics, changes the PD result. And the page itself mentions that resampling changes the result. For reference LUFS-I and LRA produces same numbers regardless of these tricks. Conclusion: PD doesn't work.

What are your thoughts or reply on this? @TT-ReBORN

The resampling part seems to be a misunderstanding of your docs, since you simply note that resampling artifacts may subtly affects results since you use frequency analysis. I don't see a problem here to be honest.
https://github.com/The-Wizardium/Audio-Wizard/blob/main/assets/docs/PD.md

What about the rest? Phase shift? Lossy compression? Have you performed tests?

[TT-ReBORN]

Hi mate,

I know about that HA thread — deliberately ignored it.
I’m still pissed at that one user who’s been getting on my nerves big time ( and you know, I have a lot of patience and tolerance ).
He started the thread without even asking me first, right at the exact moment I was releasing GR 64-bit and wrestling with the idiotic WinAPI to fix all the Windows UI bugs in UI Wizard. I was getting like 10 PMs a day for a whole month — absolutely swamped. Then I saw the thread and some of the responses… come on, you get a pro-grade free tool that rivals commercial plugins, and people post that kind of stuff? I decided not to waste my time. If you knew what that guy sent me privately on HA, he’d probably get banned. He’s asked other same questions on HA 4–5 times already and ignored my answers — my guess is he’s just trolling.

Anyway, back to your question.
I’m really proud of Pure Dynamics. It’s currently just one feature in Audio Wizard, but it deserves its own standalone app. I even bought www.pure-dynamics.com — site will launch early next year.
Pure Dynamics is imo superior to existing metrics like LRA, Crest Factor, or classic Dynamic Range. Why? Because it’s built on real psychoacoustic principles — the stuff pioneered by Hugo Fastl and Eberhard Zwicker — not just statistical tricks. You simply cannot fairly compare statistical metrics (LRA, DR, etc.) to a true psychoacoustic model. PD actually tries to model how the human ear and brain perceive dynamics. That’s a huge leap, and it gives far more accurate results in practice.

I’ve tested it across tons of genres and audiophile reference material, and the difference is night and day — it really reveals when a mix/master has great dynamics. I’ll post proper comparisons and screenshots on the new site when I have time.
As for classic DR being flawed: I remastered albums 10+ years ago and could easily manipulate the DR score by artificially boosting a few transient peaks. Everyone serious knows this — it’s why many pros avoid it. There’s even an AES paper exposing exactly that:

"Measuring Dynamics: Comparing and Contrasting Algorithms for the Computation of Dynamic Range" by Jon Boley et al. (AES 129th Convention, 2010). They tested various algorithms against listening tests and concluded none of the simple crest-factor methods accurately predict perceived dynamics — and highlighted how easily they can be manipulated.
Funny side note: the guy saying “PD doesn’t work” is the same one who develops and maintains the current DR plugin. So he’s actively promoting the very metric we both know has these flaws… I don’t think I need to say more, right? 😏
That’s exactly why I don’t bother replying on HA — not worth the energy with certain people.

PD isn’t perfect yet, no dynamic range metric is. The resampling artifacts are there because the FFT-based psychoacoustic processing picks up subtle signal shifts of the resampled track — I’m totally transparent about it in the docs. Next year I’ll refine it further (maybe even solve those issues). DSP gurus are welcome to jump in and help!

Cheers,
-TT

[regorxxx]

I was getting like 10 PMs a day for a whole month — absolutely swamped.

Sounds familiar. xd People seem to think they will get their requests/fixes if they write by PM

my guess is he’s just trolling.

Other devs just stopped sharing due to this same user, so sounds familiar too. Unfortunately I don't think it's trolling, since it has been years with the same attitude.

As for classic DR being flawed: I remastered albums 10+ years ago and could easily manipulate the DR score by artificially boosting a few transient peaks. Everyone serious knows this — it’s why many pros avoid it. There’s even an AES paper exposing exactly that:

Well I'm on a few music forums and it has been the norm for years to "remaster" albums by applying a low pass filter which produces phase shifts and increases DR by 3/4 points xd

Something similar is seen on vinyl, where it was claimed that brickwalling was impossible but I demonstrated that's false by comparing a known digital release vs the vinyl version, and surprise surprise I was able to match the digital release by applying an inverse low pass filter which reversed the phase changes introducing at vinyl cutting/ripping/riia. And brickwalling was there, literally, just shifted (so instead of being at the top of the waveform as an horizontal flat line, it was a flat diagonal line shifted a few ms). A lot of people didn't care at all about it, claiming +4 DR points sounded better... even they were just a byproduct of wrong processing without any audible change regarding dynamics.

Funny side note: the guy saying “PD doesn’t work” is the same one who develops and maintains the current DR plugin. So he’s actively promoting the very metric we both know has these flaws… I don’t think I need to say more, right? 😏

Well, to be "fair" on both sides ;) we have to take the comment in context and accept the reply was probably made to say your component "is not magic". And things "work" within a given context. DR "works" in a subset of tests and contexts, but is obviously flawed in many senses. So I can understand he develops it. Also probably the problem is you didn't create the topic by yourself and didn't give a corpus of examples and results showing the limits of the algorithm, when it works fine and when it doesn't. And how much deviates. In the end HA requires tests to prove things, so claims without them are discarded pretty fast.

I was more interested in how much the "flaws" of PD affected the results. i.e. how much resampling or transcoding affect the final scoring. If we are talking about less than 1 point I don't really see a problem here.

I’m really proud of Pure Dynamics. It’s currently just one feature in Audio Wizard, but it deserves its own standalone app. I even bought www.pure-dynamics.com — site will launch early next year.

Looking forward to it and your results. Meanwhile, since the plugin seems to work fine now, I will try it by myself, maybe posting the results at HA at such topic.