Tokyo Dawn Labs - Feedback Compressor II - beta 1

[FabienTDR]

The TDR Feedback Compressor II is a radical revamp of its critically acclaimed predecessor. The processor elegantly combines the self-adjusting properties of feed-back compression with an advanced, yet musically intuitive control scheme.

Beta just started!!
More info and download available here:
http://www.tokyodawn.net/tdr-feedback-compressor-2/

Looking forward to your feedback!!

[trapmoo]

Damn you ... Still fighting with Xcode crap or have you given up already?

[sinkmusic]

Thank you
It looks very good !

[sinkmusic]

It is a joy to use, and i like the GUI a lot.
Thank you for this new instrument.

I guess it could benefit from some optimization (i have read it is still in beta and that the CPU use will be improved), but the CPU is not up the roof so it is already good (around 2,5% cpu use here) ;

Could you elaborate a bit more about the dual release envelope (peak & rms) ?
What is the purpose, when to use which, why is the RMS greyed below 100ms, etc

Why is there both a make up gain and a output level ?

What is the key stereo diff ? I guess it it releated to side-chaining ("key"), but i don't get what it is doing...

[DuX]

I just woke up and saw your "We are excited to present you our latest plug-in" e-mail. The excitement is absolutely mutual, I can tell you that. And for a good reason it turned out after about half an hour of testing. I think this one is going to become my favourite compressor and replace TDR FB I, even though you said this one is quite different, and for a different purpose. Yes, it is rather different. Well, anyway, I'm talking about using it as track compressor. I think that FB I is a more transparent buss compressor, but I FB II can also really shine as a buss compressor if you want more "smack" and "mojo". hmmm it can be quite transparent, though, too. It's more versatile in my opinion than FB I.

GUI aside [it is PERFECT], I'm very excited and eager to start using it in projects.

I was wondering about what will "peak crest" do, and it reminds me of what the "attack", when you turn it anti-clockwise [-], does on SPL "transient designer", for example. Absolutely useful! Love it! I can say that this is what I have been missing on every compressor so far, but I didn't know what it was.

While I was testing it with the 996kHz sinewave looking for aliasing and undesirable artefacts which it passed with flying colours, I can point you out to one thing that I think could be a little bug. On the default preset, when you turn on the 3dB/oct HP filter on, and a threshold a bit down so it starts compressing, FB II starts to generate additional harmonics [1kHz,2,3,4...] and a little DC offset. It doesn't occur in any other mode of working, so far. But you know what? I like it how it sounds in this mode... It sounds a little bit more bright. which is logical, due to additional harmonics. Lots of compressors have that, even very expensive ones [that I've only used and tested in other studios since I'm on a budget most of the time ] If this behaviour is a bug, I think I would like to be able to turn it on or off. Together with "smooth" turned on, it sounds like music to my ears. I even wonder if this behaviour is intentional? But then it would probably behave the same way and generate harmonics and DC on 6dB and 9dB/oct sidechain HP filter, too. hmmm That's why I thought I should mention it to you and get to the bottom of it.

Speaking of optimisations and CPU consumption, yeah - it's not bad at all. On my PC it consumes 0.8% in both "smooth" and "precise' modes, which is comparable to other compressors that I use [Molot for example], just a tad higher. So yeah, I would say it is already good, too. Especially if you have to sacrifice quality to make it use less CPU, of course. If you don't have to sacrifice quality, then I'm all for more optimisations, yeah, so I could put it on every channel without thinking.

Thank you so much Fabien! Great work and thinking!

Cheers!

p.s. oh, and "Delta" absolutely ROCKS! I see me using it for some "creative purposes" also...

[Holophonic]

I'm on XP-SP3 and it's crashing on initial plugin scan on my hosts.

Reaper/EnergyXT 1 & 2

[Aiynzahev]

First of all GUI looks great. Secondly I am really pleased it's 64-bit already.

I can't wait to try it out!

[DuX]

Weird, Holophonic, I use XPSP3, and Reaper, and Energy XT 1, too, and I haven't experienced any crashes. So far.

Oh yeah, the GUI is instantly usable. Great colours, great GR display, all very useful, too, not just nice looking. Some people tend to think if a plugin doesn't look as a rack, and doesn't have fake screws on it, it cannot sound analog. Funny, people...

[Dean Aka Nekro]

Nice one, Looking forwards to trying and hopefully offering some feedback

Cheers

Dean

[FabienTDR]

Ok guys, I messed up several things yesterday during the final builds.

It seems I broke the 64bit version, so I removed the link from my page until I fix the problem.

Not enough, I packed an old 32bit into the zip, instead of using the latest!

So if you already downloaded the beta zip, please delete all FB compressor II dll's, re-download the zip and try it again. Press the "i" button on the plugin and check if the version-text really says: "Version: 2.0.0 beta 1".

Sorry for all the struggle, it was along day yesterday.

Thanks everyone for pointing me to that mess

Damn you ... Still fighting with Xcode crap or have you given up already?

No success i.e. enough time yet. But I don't give up that quickly!


(Will elaborate on the algorithm/control part in another post)

[DuX]

Roger, Fabien. Thank you!

Edit:

OK, this version doesn't exhibit the "3dB/oct-HP-SC-Filter bug", and it works and sounds smoother. I haven't found anything wrong with it, so far. It also uses quite a bit more CPU - 1.5% now on my PC, in "precise" mode, of course. 0.9-1.0% in "eco" mode. I don't really care when it sounds like it sounds, but I'd like to know the difference. Is it just oversampling? Because I work at 96k, maybe I could use it in "eco" mode all the time.

I kinda miss the peak/RMS meter now, but I can use my ears instead to adjust the peaks. For me it works best between 3-6dB. 12dB only if you really want to smash "things".

I also kinda miss that "3dB/oct-HP-SC-Filter bug" dirtiness, somehow... hmmm, but that sound could be possibly achieved with an EQ that introduces some nice saturation, and a bit of broad high shelving, just a dB or so, too. But I can't say I don't miss it, if you know what I mean. I rarely find a good sounding bug. That is in "COMP" mode... And this version also introduced the very interesting LIM/COM button. When I engage LIM, it does introduce some additional harmonics and it sounds more similar to the "old version". Coool.

♥ ♥ ♥ it! I'm going to start using it right away... doesn't matter if you change some things, I'll compensate for them later, no probs. It's great!

Thank you Fabien!

[Tp3]

The plug (v II) is not visible in Chainer

[FabienTDR]

@Dux: The old 32bit version I packed yesterday has a bug in the 3dB/Oct filter. Deja vu The latest version should work as expected.

I have to work the the 64bit bugs urgently, but I'll try to answer some sound/usage related questions from Dus and sinkmusic:

1. CPU performance: Except a nearly 50% performance boost over the time of the beta. It's hard and time consuming to debug highly optimized code, so I'm very glad to hear you guys are already more or less fine with it. I will optimize the code shortly before the final release.

2. About the dual release envelope:

Maybe this explains a few things:



As you can see, the compressor basically runs two side-chains in parallel, the one with the strongest gain reduction takes control.

It's only barely visible in the left flow "tree", but you can see that the peak crest control is in fact just controlling the threshold of the peak detector path relative to the main threshold (which directly controls the threshold of the RMS detector path). Peak crest allows to control the crest factor of the processed audio (within reasonable limits).

Push the crest high and the RMS detector will nearly take full control -> Smooth compression

Reduce the crest to 0dB the the Peak detector will take control most of the time -> Very fast and heavily controlled compression


Feel free to ask more, I know this isn't exactly vanilla comp design. The manual will discuss these things properly.


3. Make Up Gain vs Out Gain: The "Makeup" and "Dry Mix" controls basically form a 2-channel mixer, "Out Gain" is equivalent to the "master". Use it to fix gain staging issues without changing the balance between dry and wet.

4. Key Stereo Diff: Haha, strange naming eh? The thing is, this is not a stereo link control. The compressor is never allowed to shift the stereo center (which happens with dual mono / unlinked compressors). The default setting (100%) perfectly maintains the original stereo image in all circumstances (equivalent to max() linking). Full left (0%) allows the compressor to dynamically change the width of audio in certain cases, or in other words, ignore counter-phase information in the side-chain (equivalent to sum() linking).



Finally, a few specs for those who like them..

Signal path oversampled 2x
On top of that, the side-chain is oversampled 4x

That is, the processor actually runs at ~0.45Mhz with an internal bandwidth of ~200kHz. The shortest attack settings has a reaction speed far smaller than one sample.

[DuX]

Thank you, Fabien. Now I get it completely. Really great and versatile compressor design, man. ♥ ♥ ♥ "Peak Crest". It sounds quite a lot better [natural] than transient designer to my ears.

So "Key Stereo" is kind of a "3D" control... But preserving the stereo image. What can I say - brilliant. Much more useful than "link", since in most cases you do want to preserve the stereo image. great!

I must tell you that both FB I and II, I find them easy to work with. I don't need any manual. But it's good to be in the know about these new controls you introduced, and what they do. It helps.

Cheers!

[FabienTDR]

The plug (v II) is not visible in Chainer

Pls check your current version by pressing the "i" button. Does it say "Version 2.0.0 beta 1"?

After a quick check, it seems to work fine in vstanalyser, soundforge and reaper. Which host did you use?

@Dux: Really glad you enjoy the concept! I put lots of work and desperate tries to make it flexible, but easy to use - and sometimes lose track because I know too much about the inner details.. ..so, it makes me really happy to hear your feedback.

BTW, you can find similar concepts on the GML and Weiss comp. The fb comp actually combines what I like most about these two comps.

About the "Precise/Eco" switch: This option was carefully designed in such a way that it really only reduces the internal accuracy. As you supposed, this has to do with "oversampling". But the switch itself doesn't change any sound parameter or static sound properties (such as noise, bandwidth, etc) at all. The switch only selects the internal rate of the side-chain (either 4x or 2x), in both modes, the oversampling filters have practically the same shape. So, what you hear is really just the improved accuracy due to better aliasing suppression in the side-chain. IMHO it's important to not do the mistake to use this knob in a creative manner. There is absolutely nothing to gain (except cpu cycles of course), if "Eco" sound better, your mood and/or monitoring system is probably fooling you. "Precise" sounds better, period.

The oversampling is structured in a relatively clever manner . Let me explain. All dynamics processors basically multiply the original signal with a DC signal generated from the original input (or output which is the case here). This is a primary source of aliasing, since such a multiplication extends the bandwidth 2x in the worst case. In this compressor, this evil multiplication is upsampled and filtered to a bandwidth of at least 42kHz to kick these problems below -130/-140dB. This (very cpu demanding) stage is automatically switched off if the host reports a sample-rate equal or above 88.2kHz.

Note that this oversampling is bit-transparent, only the actual compression ("delta") gets is oversampled, not the original signal! The full original bandwidth is maintained!

Additionally, the side-chain itself always gets oversampled 2 or 4 times depending on the precise/eco setting (practically 4x or 8x because it runs "inside" the already oversampled compressor). This oversampling reduces aliasing inside the side-chain, IMHO the biggest qualitative bottle-neck of digital dynamics.