Nebula VST Plug-in Tips – Switching To TIMED Kernels 18


If you haven’t yet checked out Acustica Audio’s Nebula VST plug-in, i highly recommend it. Read my comprehensive Nebula review.

For those of you that are using the plug-in, and would like to get the very best sound quality from it, the good news is that there are various tweaks that can be applied to do just that. The best tweak to my knowledge is altering the way that the plugin plays its various kernels for each individual program.

What Is A Kernel In Nebula VST, And Why Would I Want To Alter How It Plays?

If you have used a convolution reverb you will be familiar with the term ‘impulse’. In the case of a reverb program it is a static snap shot of a room’s reverb taken by making a recording of the room in the real world. You can think of Nebula’s kernels as impulses, but there are a lot more of them and they record much more data about the sampled hardware.

These sampled kernels must be processed along with your audio to produce the final sound. Nebula’s engine has 3 different modes that can be used to perform this:

  • 1. FREQD mode

This uses the same standard convolution engine that programs like Altiverb and Space Designer use, FFT (fast Fourier transform). In the Nebula support forum the programs’ designer himself, giancarlo, said:

“Freqd is the classic fft algo used by best plugins around, so it’s not so bad”

  • 2. TIMED mode

This is also known as ‘direct convolution’. It is very CPU intensive, but more accurate and produces better sound quality. How much better we will discuss later.

  • 3. SPLIT mode

This is a hybrid of the 2 modes. It is assumed that the main advantages of using TIMED mode is in the effect on the transients of the sound, the very first portion. It is in this first part that the listener hears many subconscious cues about all sorts of interesting aspects of the sound. It is also the part responsible for the punch of percussive instruments.

SPLIT mode lets you process the first part of the sound in TIMED mode (ie the transients), and the second part (the tail) in FREQD mode

OK, What Button Do I Press To Increase The Sound Quality!?

Sorry, this is Nebula we are dealing with πŸ™‚ Nothing is easy or straight forward. We need to jump through a few hoops first, but don’t worry, all the info you need is right here. We need to make a global settings change, and then alter each program one at a time.

Step 1 – Modify The XML File

Nebula VST by default is installed in your DAW plug-in directory as 2 versions. Nebula and Nebula Reverb. Nebula Reverb is the same as Nebula but with higher quality, higher CPU use defaults. We of course use this version as a starting point.

The .DLL file that is the actual program has a corresponding .XML file with the same name. This contains the settings, and this is what we will be modifying in a text editor.

It is possible to change the names and copy the files of the .DLL and the .XML files if you want to make different versions of Nebula with different settings for latency and quality. This is what i have done, but don’t let it bother you. My main Nebula files live in C:Portable Apps/Reaper(x64)/Plugins/VST/acustica and are called:

Nebula3CoreII64 Reverb.dll
Nebula3CoreII64 Reverb.xml

Yours may differ, they could be in Program Files/Steinberg/Cubase Studio 4/VSTPlugins/acustica… i am sure you can find them in whatever plug-in directory they live.

I am using windows, so regret i can’t comment on the mac version, but i am sure any text editor will do. In windows open the XML file in notepad by right-clicking the file and selecting open with notepad:

Nebula-VST-Notepad

 

You will now have the following open in your text editor, i have highlighted the only line you need to change:

Nebula-VST-NebulaXML

The parameter is changed to <LTIMED> 100000 </LTIMED> . This lets us use up to 100ms of a TIMED kernel, any more and your CPU may explode, and it’s not needed. Hit save, boot up your DAW and we will go straight to the video to explain how to change the settings inside Nebula:

http://youtu.be/6F9mBvhg_l0

So now you can alter your own preamp programs to run on TIMED kernels! Hit the save button to make the change permanent, or better still make a copy of the program first so you don’t mess anything up by accident.

How much better is the sound quality in TIMED mode?

There has been quite a lot of talk on the Nebula forum about this subject, the main related thread is:

http://www.acustica-audio.com/phpBB3/viewtopic.php?f=11&t=5251&hilit=transient

I would like to thank all the forum members who contributed to this thread as they were the inspiration for this article.

I can detect different things in different audio. I don’t want to put ideas in your head of what to listen for, but a subtle ‘tightening’ and little more punch is what i am mainly picking up. Its not massive, but if you are into sound quality and have a decent CPU well worth it for the 5 minutes it takes to edit the program.

Download An Example Of TIMED vs FREQD

I have rendered for you 2 versions of an interesting old stereo percussive bongo loop (16bit / 44.1 / FLAC) through the Analog In The Box ‘Fate’ compressor preamp using both TIMED and FREQD:

Nebula VST Plug-in Tips – Switching To TIMED Kernels AUDIO SAMPLE

Please let me know in the comments below what differences you can detect, it would be very interesting.

If you want to get deeper into this subject I highly recommend you check out my Nebula Explained Udemy course. There is a section devoted entirely to the Kernel page.

If you found this article useful please consider joining the mailing list and/or clicking ‘like’ on the facebook box on the top right hand side of this blog. THANKS.

Facebooktwitterredditlinkedinmail

Leave a comment

Your email address will not be published. Required fields are marked *

18 thoughts on “Nebula VST Plug-in Tips – Switching To TIMED Kernels

  • dagovitsj

    Hi!
    Thanks for the tutorial, really appreciate it!
    However – when I downloaded the zip-file and listened to the two files, I couldn’t hear a difference, not with my headphones either.

    So I imported the two files into Audacity the ‘Nebula_Clean_FREQD_Kernel’ on the first track and the other’Nebula_Clean_TIMED_Kernel’ on the second track. Then I inverted the ‘Nebula_Clean_TIMED_Kernel’ (Effect > Invert) and pushed the play-button – no sound at all. Ergo they null sum, or in other words – as far as I know, the two files in the zip-file are identical.

    I think that by accident you have put two identical files in the zip-file, and I’m interested in hearing the difference between the FREQD version and the TIMED version.

    I also did the same test on the original file (which I think you posted at the Nebula forum?) with the FREQD file – and they didn’t null sum, ergo they are different.

    Conclusion: I think there are two identical files in the zip-file, not two different. But I don’t know if it’s the FREQD or TIMED version?!

    Can you please check on this?

    PS: Unfortunately, I haven’t had time to tweak my own Nebula as you have explained above, but I sure will – I have read much about this tweaking at the Nebula forum πŸ˜‰

    Cheers!

  • David Post author

    Hi dagovitsj,

    Thanks for writing, i am very happy you like the tutorial!

    The files in the .zip are not identical, you can confirm this with foobar2000 Binary Comparator:

    http://www.foobar2000.org/components/view/foo_bitcompare

    I am not familiar with Audacity, but understand exactly what you are trying to do to measure the difference in the files.

    I suggest you checkout this cool little utility. It is a VST .DLL, but comes with SAVIHost. It can run standalone or in your DAW:

    http://www.ohl.to/about-audio/audio-softwares/hear-the-difference
    http://www.ohl.to/audio/downloads/hear_the_difference.zip

    Load the 2 files to compare (as .WAV’s) into the Input A and Input B slots and select the A-B option. Now you can see where the difference is in frequencies and hear it.

    Are you convinced? πŸ™‚

    • dagovitsj

      Hi, again David!
      Sorry for answering so late, been very busy etc. But, yes, I’m convinced – but I find it very hard to hear any difference, probably because the perc sample isn’t the best recording (a lot of reverb in it). I did a file comparison in Wavelab7, and it pointed out a significant difference. I have to do more tests with my own audio files when I have installed all my software on my new SSD-disk (my PC crashed, so it takes a lot of time to reinstall everything, and yes, now I’m making an image of my C-disk!)

      I’ll follow your blog, much good info here. Keep up the good work πŸ˜‰

      Cheers

  • John Lardinois

    This article rocks! Thanks!!!

    I’ve been putting Timed on all my programs.

    However I am perfectly able to use both Even and Odd on every program.

    Also, the DSP increase is extremely minimal, at least on my machine.

    I can typically run about 15-20 instances of Nebula in my sessions, and with Timed I can still run about 14-18. My computer is not overly powerful either.

    1st Gen i7, Quad Core, dual threaded (multi threading actually slows down DAWs), 2.8GHz
    6GB Ram DDR3 (I forget the speed, but it never really stood out to me, so it’s not crazy high)
    64-bit system
    and everything is running off the same hard drive, so its not exactly optimized for speed (running samples and software off a dedicated HD, for example)

    The specs of my computer are actually far below even a modern laptop, so if the Timed kernels barely makes a dent in my CPU on this machine, I doubt it will make a difference on most other modern machines.

    I set my timed to Clean, Even, and Odd, maximum time, maximum kernels, my Rate CNV is maxed out, my L Timed and FREQD are maxed out, and my Quality setting is set to the maximum of 5, my sample rate is 96kHz, and my bit depth is 32-bit floating point.

    I should also let you know I am running Nebula 2, as I do not own Nebula 3. So if I can max out all those setting on Nebula 2 with a mediocre computer and still handle 14-18 instances, I really don’t think DSP is an issue.

    Also, my computer is not overclocked, but I plan on doing so sometime in the next few weeks. I’m hoping I will be able to run 30-ish instances.

    I will say however, if I am running a lot of preamps, the count goes to 10-14. For some reason, the preamps have a much heavier load on my computer than the other programs. That’s why I always get my preamps setup and bounce right away to take them off the CPU, then I do my regular mixing with the other programs (EQ, Comp, Timed, etc)

  • David Post author

    I am very happy you like the article! It seems very popular.

    Regarding your settings, i don’t recommend having Even AND Odd to TIMED due to a documented error. Its explained:

    http://acustica-audio.com/forum/index.php?f=11&t=5251&hilit=quality%20setting&rb_v=viewtopic&start=120

    The ‘quality’ setting is bad news! It does not work properly and can create clicks in mixdowns. I used it myself for a while and found out the hard way, took ages to track down the problem. Turn if off.

    Your PC is kicking ass! I am looking at getting an Ivy Bridge i5/i7 upgrade myself soon, looking forward to working at 96k πŸ™‚

  • Stuart

    What would be really nice is if nebula had an option to switch to the highest quality settings automatically at mixdown. That would allow us to mix at lower quality settings with reasonable demands on cpu and then use the high quality settings for the final render. I dont mind letting things churn overnight once im happy with my mix.

    Stuart

    • David Post author

      Thanks for your comment foxman, I agree the difference is small, but would disagree that the range below -60 dB is inaudible. Depends on the noise floor in your studio and quality of monitoring.

      I am reading forum posts by a lot of people saying that they are getting improved ‘punch’ using TIMED. It is possible that this is a strange cognitive bias brought on by reading other people are hearing it, but unlikely.

      Everyone must do their own ABX to decide for themselves. My bongo loop example is just one preamp on one sample. It does prove at least there is a difference, so if people can hear it or not is another matter.

  • Jules

    A great article collating lots of useful information..

    I do think its important however that you credit you sources. This is not info you have discovered. Other Acustica Forum users (Timp and a few others) collected this knowledge and I can imagine if you continue to write articles without giving credit where credit is due then the real pioneers of this amazing software will keep new knowledge to themselves.

    That aside, I think a real in depth manual for Nebula should be developed (you could be the man for it) and the database work you have down is excellent.

    Cheers.

    • David Post author

      Hi, glad you like the article and the database.

      Regarding giving credit, I am not quite sure who I would credit for ‘discovering’ this. I do link directly to the original forum post that brought up the subject of TIMED kernels. I take your point, I have added just below the forum link ‘I would like to thank all the forum members who contributed to this thread as they were the inspiration for this article.’

      As for a Nebula manual, there is something even better being seriously discussed, but don’t hold your breath πŸ™‚

  • Elucidation

    Whoaaaaaaaaaaaaaaaaaaaaaaaa!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Anyone who cannot hear the VERY CONSPICUOUS difference this makes, I advise you to invest in a good monitoring situation (room treatment and good monitors). This is ASTONISHING. One thing that may become a problem because of this switch are the existing sessions you may have with Nebula already on them (and un-bounced). You may have a problem because these awesome settings change the sound of the files almost fundamentally (in terms of character); with my AlexB 4KD, for example, this switch softened the hell out of the highs, made the transients punchier (and tighter), and to top things off, it made everything sit SO MUCH BETTER in its own space. Now that’s just with this AlexB compressor on the master fader alone. Imagine all my other disparate types of libraries! WHAT WILL THESE SETTINGS DO TO THEM!?!?!? Holy HELL! Thanks so much for this! I’ve been using the FREQD this entire time!

  • Matt

    I agree with Elucidation. I have mastering grade monitors (ADAM A7’s) and mastering grade converters (Metric Halo) and i to can hear the difference. Timed mode is more clear and open. Freq mode is more crunchy.

  • jim

    is another problem with this VST, I use Cubase and when using the VST procesoru is full, I can else has used, so that you can try a demo to figure out if it’s worth buying , I think that soft and the Library’s not worth the money , it costs a lot of money and you can not use it because you work full procesoru and have problems from the system, without having to open or insert of other VST ins or virtual instruments, I have an i7 with 16 gigs fell to memory so ……….shame it’s not designed properly, it sound great and work even better, as are actually professional VST……………..I’m curious who designed a muzicican or an engineer? ))))) big difference between them what you think ?

  • Ben

    Hi David,

    Thank you very much for the tutorial, clear and very helpful!

    I have decided to just change the Clean kernal to TIMED mode, leaving Even and Odd in FREQD mode. I have a question though…

    I edited the .xml file then opened up Nebula and increased the Clean TIMED kernal to its maximum, which is 73.5ms in this case.. (I guess this is the maximum for this particular library, alexb’s Vintage Blue Console, correct?)

    My question is, do I need to increase the FREQD millisecond values to their maximums too, or just leave them at the default 50ms?

    It is interesting because the maximum FREQD value of the Even kernal is also 73.5ms, but the Odd FREQD value goes up to 75.0ms –

    So how do you recommend I set these values?:

    Clean (TIMED): 73.5ms
    Even (FREQD): 50.0ms (default)
    Odd (FREQD): 50.0ms (default)

    or all matching..:
    Clean (TIMED): 73.5ms
    Even (FREQD): 73.5ms
    Odd (FREQD): 73.5ms

    or all at their maximums..:
    Clean (TIMED): 73.5ms
    Even (FREQD): 73.5ms
    Odd (FREQD): 75.0ms

    Many thanks in advance!

    Ben

    • David Post author

      >> I edited the .xml file then opened up Nebula and increased the Clean TIMED kernal to its maximum, which is 73.5ms in this case.. (I guess this is the maximum for this particular library, alexb’s Vintage Blue Console, correct?)

      I don’t have this program, but this will be the maximum if that is as high as it goes.

      >> My question is, do I need to increase the FREQD millisecond values to their maximums too, or just leave them at the default 50ms?

      If you are running a kernel in full TIMED mode (not SPLIT mode) then it doesn’t matter what the FREQD setting is for that kernel, it won’t be doing anything. I would leave it otherwise.

      >> It is interesting because the maximum FREQD value of the Even kernal is also 73.5ms, but the Odd FREQD value goes up to 75.0ms – see picture

      This will just be a tiny error in sampling or in Nebula, I often see these small number differences, can’t do anything about it, best just ignore it πŸ™‚

      If i were you i would set it at:

      Clean (TIMED): 73.5ms
      Even (FREQD): 50.0ms (default)
      Odd (FREQD): 50.0ms (default)

      But it’s easy to get obsessive about this stuff, if you can’t hear a difference then there is no difference (for you!)

      The clean one is the main one to think about, some developers are now releasing HQ versions that start at 30ms on the clean TIMED. As a rule of thumb you could go for 50ms, all the low bass should be accurate at 50ms, if you are running out of CPU then:

      Clean (TIMED): 50.0ms
      Even (FREQD): 50.0ms (default)
      Odd (FREQD): 50.0ms (default)

      might be a good choice. I am glad you like my video, thanks!