De-Revereration made possible? Not only that …
We will test it. We’ll find out … Has anybody tried it, already?
Also for VST 2.4, still 32 bit but 64 bit is in the pipe…
De-Revereration made possible? Not only that …
I’ve been using the Beta.
On some things it’s amazing. Had some loose dialog, recorded with boom in a practical office. Unveil did a much better job than my previous standby: carefully tuned parametric to kill room resonances, then an expander.
I’ve also played with using it to enhance natural reverb as the camera angle changed. The knob goes both ways.
Just be aware it can’t perform miracles. Nothing can make a camera mic ever sound like a lav.
Also: spend the time to learn the knobs, using their very good docs. This is a new way of doing things, and the controls are both interactive and non-intuitive.
THe 8-channel version EATS CPU alive.
My 8-core Xeon E5 chip will not run even a single instance of Unveil 8-channel…
Give them 8 bowls of hearty stew for lunch … lol …
Good point! If it is such a CPU beast, the developers should go back to the drawingboard.
We will see, what the next update brings.
What is truly strange is they reckon it works fine on an i7, yet not on my monster Xeon.
Quite right. I have mailed them and will post their statement here.
If this Unveil is fully usable, it is a rather powerful gadget for post.
Have you seen their other gadget? For composers this could be fun…
Oh, it is fully usable. Neil was talking about 5.1/7.1 instances - which means running 6 or 8 channels on one core. Mono or stereo instances are fine, also multiple instances. I’m currently running 6 stereo instances on an elderly 2.8gHz Mac Pro.
Indeed, on our i7 3.4gHz Imac, we can run a 5.1 instance at 44.1khz SR in Nuendo 5.5 (even running Win 7 on Parallels). 8 channels on one core is too much for that machine too. We’re rather surprised that the E5 performs less well than an i7, but then the advantage about server CPUs is in handling lots of processes quickly (as opposed to handling one calculation intensive process), so there may be some different optimizations in the E5 and similar processors that generally enhance the performance for a variety of scenarios - but not for UNVEIL.
Essentially it is an issue with the threading/load balancing though — the CPU overload is essentially one core overloading with all the others idling. This is down to the way that pretty much all DAWs out there handle rendering of plug-ins: one plug-in is always run on one core (actually, most hosts will also render all plug-ins that are on one track or within one routing context on the same core). There are a handful of hosts that allow instantiating multiple mono instances of a plug-in into a channel while giving you access to all of the individual channels’ settings separately, that will then actually split the channels across multiple cores for rendering.
I may add that you’ll be fine running several mono and/or stereo instances, Neil was talking about 5.1/7.1 instances.
As to the drawing board…well, let me say that - while we do of course understand the inconvenience of the high CPU demands UNVEIL has - you should please take in to account what this plug-in is doing! Removing reverb from any signal type without significantly interfering with the useful signal is quite literally NASA-level mathematics. It’s been quite a lot of research and work to have this possible in real-time at all (and as you’ll notice there’s nobody else capable of this particular trick in this quality, not even as an offline process…), so suggesting we “go back to the drawing boards” is sort of missing the nature of this level of engineering. It’s really not a case of us not having “done our homework”, the nature of the process is just really, really demanding in terms of calculations per second.
In a nutshell, we’re running a pattern recognition process, a perceptive model, a de-mixing stage as well as some other bits and pieces to make this possible, so it’s not like a gate or FFT-based process that can be made extremely CPU-efficient because it needs to solve only relatively simple equations. This is more like half an AI (artificial intelligence). Well, not half, but you get the idea.
All that being said, there are some routes for CPU and/or threading optimizations that we’re exploring for future versions.
Denis (CEO @zynaptiq)
How does it work in VE Pro? It might be that the load balancing is much better than using it direct in the DAW. It’s worth a try anyway, I would have thought.
There are always new ways and routings to be explored for a more slick way of processing and safeing CPU resources. So, the drawing board is kept not too away far from the developers bench, I reckon. And this is just fine.
But don’t mind me saying: if Neil can’t use Unveil’s full potential on his fat rig then you can’t sit back and say “Well, …”
If it is supposed to work with 7.1 it must be good to go.
Nuendo itself is an “eierlegende Wollmilchsau”, meaning a rocket science multi-purpose hell of a software.
So are others like Melodyne’s DNR or Algorythmix “ReNovator”, etc, pipapo…
It is safe to claim that UNVEIL’s high level of engineering ingenuity is well recognised and appreciated by 99 % of the Nuendo forum members.
You at zynaptiq have produced a true jewel of a program with UNVEIL. But don’t claim it runs 8 channels, depending on CPU power, when a whopper processor like Neils 8-core can’t pull it off. Make it fully run on all commonly existing, strong Studio computers and you’ll be getting a bunch of awards for it!
You get my vote…
it must be good to go.
We’re not claiming it runs in 7.1 on any system, we’re explicitly saying that if your CPU can handle it, the plugin allows you to run in any format up to 8 channels wide. We’re pretty clear about that, and encourage people to evaluate how well their systems performs in terms of multi-channel CPU use by downloading the demo. I agree that it’d be more elegant if the plug-in could run all the channels it technically supports on a broad number of systems, but this is where we’re at for the moment.
We could have decided to leave out the multichannel VST, but that would have meant witholding the option to run multichannel if the CPU allows from our users. I personally prefer having options to not having options 10 times out of 10, especially as CPUs get faster and faster, so we went and provided the 8ch VST build. We don’t believe in making decisions for people. Options are great, though.
The fact that Neil’s E5 doesn’t seem to be be noticeably better for running UNVEIL (or even less good) than an i7 is rather interesting actually, and I can only assume that being a server processor it is optimized for running many processes instead of running fewer but heavier processes…a desktop user is more likely to do single, heavy stuff like rendering HD video filters etc, while a server is more likely to be processing hundreds or thousands of parallel requests (which would generally make it great for running many tracks and plug-ins that are typically light-weight in terms of CPU), so I’d just assume that this is part of the design specs for CPUs.
But did you actually TRY the demo on your system, BigK? If not, I strongly suggest you do. Clean up some dialog or SFX, tighten up some wallah, dry up some drums. Then come back and tell me we didn’t do our home work OK grin .
All that being said, of course we’re working on improving the CPU side of things, as I already stated earlier. Nobody is “sitting back”.
It’s simply that there are limits to how far the number of calculations per second required can be reduced because of the nature of the beast. It already is heavily optimized code. It is actually amazing that this is possible in real-time AT ALL. To give you an idea of the level of optimization (math nerd warning!): we’re solving an “n multiplied by m” problem in an “n plus m” amount of time, where n and m are not small.
For example one thing we’re investigating is whether and how it could be done reliably to bypass host CPU load distribution / threading (which is really the bottleneck here - the fact that all channels are being rendered on one CPU core). If we could distribute rendering our channels to the cores independently of the host’s scheme that’d be great. That’s pretty much uncharted territory compared to sticking to the host’s methods though, so something that takes a long time to test and make rock-solid. Another thought we have is providing a simple stand-alone app for multi-channel use, as that way we could devise our own threading pretty easily & render channels on separate cores. The latter is not unlikely to make an appearance at some point.
And i’d like to reiterate that Neil would presumably be happily running 7.1 if Nuendo allowed using multi-mono instances with full control of the individual channels with VST 2.4 plugs, and rendering individual channels on separate cores <cough, cough> One channel of UNVEIL should use about 16-17% of one core.
By the way it’s looking like we’ll be done with the x64 build rather soon.
As to awards…we just got the Electronic Musician Editors’ Choice 2013 Award
For the non-German users: the correct meaning of “eierlegende Wollmilchsau” is along the lines of “one thing that is great because it combines many essential functions that you’d normally get from using multiple things, but which is deemed to be to good to be possible” (and NOT “something that is so difficult that only few people have the knowledge to do it” like the term “rocket science” implies ). The direct translation is “a pig that also gives wool and milk, as well as laying eggs”. A one-animal farm, essentially, and something that would make any farmer really, really happy.
But leaving semantics and British humor aside for the moment…don’t get me wrong but based on the comparisons you’re making, it would appear that while you state that you do generally appreciate the level of engineering in UNVEIL, you may still not be quite clear on the actual difference in mathematical/computational complexity between Fourier transforms as used in the products you mention and UNVEIL’s pattern recognition based blind de-reverberation using source signal separation and perceptive modeling. If you read up on those topics, you’ll notice that you’re comparing apples to oranges here. I’m not making this point to state how much of a genius our DSP guy is (which he actually is), I’m making this point to explain why the CPU load is as high as it is and why it is not merely a question of “going back to the drawing boards”.
Also, you’re comparing off-line editing applications using a pre-analysis step and requiring the user to make the decisions (even if DNA has a plug-in as GUI frontend to that) to a real-time plug-in with automatic/machine decision making. That’s comparing apples to something…erm… non-fruit. Or are you running a version of ReNovator that works on a real-time audio stream and automatically detects coughs, mobile phone ringtones and sirens that should be removed & does that all on its own? No, you’re not, you are taking minutes to find the sounds that should be removed (after the application did the FFT transformation for you), select the data that represents them using a variety of tools, and then the application takes a little while to interpolate the data. Chances are you won’t be running that sort of thing automatically for another 5-10 years, either (unless of course zynaptiq releases something like that…who knows? grin).
But UNVEIL is that sort of process. It detects the reverb components by itself (although you do help it a little by “telling” it how long the reverb is with the ADAPTATION parameter - but it does a decent job even without that additional information), selects the parts of the data that represent the reverb, and separates it out automatically. Similarly compared to DNA…if you could tell Melodyne to “quantize all the guitars to the rhythm they we’re supposed to have in the first place, but don’t touch the Rhodes or vocals”, and have that happen in real-time without pre-analysis or manually “telling” it what notes are guitars and which aren’t (and of course using a stereo mix of all the elements as input), then we’d be talking about a comparable level of mathematical complexity. There’s an entire additional layer of complexity involved in the machine interpreting the data on it’s own, and desktop CPUs have only very recently become powerful enough to be doing that sort of thing in real-time…barely. All the editing applications you mention are brilliantly engineered and great products which I use a lot for my sound work (though I personally prefer RX Advanced to ReNovator TBH), but again, it’s apples and oranges.
All the best,
Just a question. What kinda use you guys had in mind providing an 8 channel Unveil?
Instinctlively, I would say that this plugin is a pure dialog thing, no!?
Great plugin BTW. I bought it as soon as the VST verison was out.
I just showed the thread to a friend and he ment :
" He sucks…"
and I said:
" He’s got a problem and it is not 8 channels…"
Dialog is one of the main applications, yes, but there’s obviously also applications like making the room components in ambience less pronounced, modifying the amount of reverb in music recordings and up-mixing. 90% of the real-world usage scenarios are mono or stereo though - but we wanted to have the multi-channel option available.
I really don’t get you. I think I’ve actually been pretty polite in reacting to your statements of UNVEIL “not being good to go”, us “sitting back”, “needing to go back to the drawing board” and “making claims” and other pieces of wording that come across pretty polemic and are unjustified from my perspective - especially as you don’t appear to have actually tested the software yourself. It may well be that this is just a case of “language barrier”, but your last post surely isn’t.
Whatever, if that’s what you want to make of it, fine. Otherwise you’re welcome to our yearly single malt at the Musikmesse
To what degree do I have to insult you to be invited too?
You’re welcome figure something out while drinking the malt, actually hehe. Hall 5.1, booth A91, right next to the cool guys of Elysia, Metric Halo, Vertigo etc.
Same goes for everybody else, there’s plenty & we’ll also be showing something new that you might find very useful if you think UNVEIL is cool.
Last year we had some nice Laphroaig Quartercask, Lagavulin 16 and Ardbeg 10. Actually about 6 bottles of each…Tom @Elysia brought some unbelievably good Japanese malt of which I unfortunately forgot the name, and I think some Bowmore was also spotted.
I’m losing faith in humanity via Fredo’s use of emoticons…
But getting it back via Denis’ geniality and offers of Scotch. Not at NAMM?
We’ll be at NAMM, Hall A booth #6224 IIRC, but I have yet to find out how to get hold of good scotch in Anaheim. And whether we can actually hand out that type of drink at the show. Musikmesse is pretty relaxed about that kind of thing, I have yet to find out whether NAMM is, too.
At any rate, in the evenings after the show there’s plenty of opportunity to grab some drinks – I hear there are some good bars in te area
Well… there are bars… but it’s still Anaheim… on the other hand, there’s probably a shadow industry down there to help people recover from hours upon hours at Disneyland…
Certainly Laphroaig can be found. Certainly.