• Register
  • Help
Results 1 to 5 of 5

Topic: Getting a better rendering

  1. #1

    Getting a better rendering

    Last night I attended a so-called "Listening course" where two cello players of our local orchestra (The Residentie Orchestra) explained the orchestral practice. We had a lively discussion about various aspects of "being a full-time musician". The following surprised me the most.


    1. The conductor is the dictator and overlooks everything. He is only conducting moods, tempi, rhythms and some other minor but important aspects of the performance;
    2. The concert master is his lieutenant, but only for the string section;
    3. The sections masters are the sergeants and they set out the rhythm, bowing speed, slurs, legato lines, bowing direction in cooperation with the concertmaster and conductor and dictates them to the rest of the cello section.


    1. The section master follows the concertmaster and conductor;
    2. The first row of cello players follow the section master;
    3. The second row follow the first row!!!!! etc....
    4. In instrument wise "crowded" pieces they don't hear all cello players play, they just listen to the man on the left or right, sometimes they don't hear themselves (depends on the acoustical environment); it is more visual then auditive.

    Nobody is playing exactly on the beat, except maybe the section master. All others are a tiny bit later. It is like a wave from front to back.

    We all knew that before, that when we render the string section, we shift back and forth the start and ending of notes for the benefit of "humanizing" the sound. But now against the sample libraries where we have e.g. 16 1st violins as one sample, etc..... playing back that one note isn't good enough. We should have the possibility to assign some irregularity within this sample, getting that "from front to back of the section" time delay.

    It is rather stupid assigning 15 different solo-cellos for one staff/track. Multiply that by x, depending on the number of instruments you'd planned to use and multiply that by y for the different voicing (articulations) and we'll end up with numerous linked computers.

    Since we have to do with the samples as presented (in GPO I think the first violins are made up of 12 players - in EWQL I can choose between 18,11 or 4 players) we must make a strategic decision, e.g.:

    for violins 1 and 2 just "double" them 4 times
    for violas 3 times
    for celli 4 times
    for basses 2 times

    When rendering we now can shift the notes per section (some ticks or milliseconds) giving us that irregularity as in "real" life. This is very demanding, I know, but I think it will benefit your outcome a lot.

    I am about to try this in the next project, seeing how it goes and what the result will be.
    If this is complete nonsense, pray for me. If you think it has some value, experiment freely.


  2. #2

    Re: Getting a better rendering

    Interesting information, thanks for that.

    While the quality of the samples themselves play a huge role in getting the best rendering, anything we can do enhance the 'human' feel should be attempted. We're talking fractions of a second here.

    I play almost every line of music in all my pieces, and keep deleting and recording again until its 'in time' - but this has an already natural human feel to it. If instruments are playing in unison, I play each line anew, rather than just copy paste the notes. Small adjustments later ensure its still on time to the listener.

    If I'm inputting notes manually, I zoom right in and start shifting notes back or forward by tiny fractions so that they aren't all exactly on the beat.

    Never occured to me to do it in a front of orchestra to back formation. I'm not convinced of the difference it would make over just ensuring most instruments are fractionally out of step - but its worth a try.

    It also means breaking the ensemble patches up into smaller chunks, and depending again on the samples, this can be a negative result as they probably don't blend well. An ensemble patch (all cellos for example) might have been recorded fully and so probably have that slightly out of step feel already present. Unless it was an ensemble built from layering.
    YouTube Music:
    My Channel

  3. #3

    Re: Getting a better rendering

    What a great post, Raymond.

    I think the basic concept of not having our various MIDI tracks playing absolutely, robotically spot-on together is the main issue when we're recording/rendering something that's meant to emulate a live orchestra.

    Both you and Plowking have talked about shifting notes so they aren't perfect, and that's basically what many people do to keep up the "fuzziness" that we're looking for.

    You've described how musicians work together in concert, groups following section leaders, so there's somewhat of a wave of delayed notes from front to back - But I don't think that trying to meticulously emulate that in a virtual orchestra recording would necessarily create any more of an authentic sound than a more random note-shifting approach. Even when slightly different amounts of reverb are used for various sections, there's a limit to how much of a three dimensional sound can be achieved. Shifting notes according to how far they are from the conductor, I don't think, would achieve any more of a three dimensional effect either.

    What you're talking about is off-setting the timing of MIDI tracks which are duplicates of other tracks. You use Sonar, and in that software you can set that kind of timing off-set in the track header, rather than physically moving the data:

    --In the track header, look for the clock icon. That's the time off-set control. You can shift note playback for that track to come in earlier or later. If you want to experiment with different values, you can do that easily by changing the amount - no fussing with moving large chunks of data around.

    The thing with using an off-set is that of course it's constantly at one value when applied to an entire track. As the Sonar Help files says, using off-set is one way to easily achieve a "slap back" echo - making a copy of a track and then shifting it against the timing of the original track.

    You're not wanting a slap-back with your orchestra recordings, you're wanting an easy way to achieve what you've described in your post, a blurring of attacks. Used in small amounts, I still think using that time off-set control would get you at least part way to your goal. For a more complete effect, I think you would still need to do some hand shifting of notes throughout a piece.

    Note - I think it's quite possible that having a timing off-set control could eventually be available in Aria.


  4. #4

    Re: Getting a better rendering

    Very, very interesting topic.

    I tend to agree with Plowking. useless to play around if you are limited by the samples themselves. You can double a section to play around, but you'll just get the same sound doubled, not very naturally feeling, with risks of phasing and so on. Basically, the two sections will sound pretty much the same, at an higher volume. Plus, some natural offsetting is already recorded in the sample itself, if recorded as a whole, but also if achieved by layering. Unless you layer X times the very same sample of the same instrument.
    Even dirty tricks commonly used on newer libraries won't work. They record divisi sections, but when you play a divisi part the player will trigger, instead of the whole section sample, two samples of ... the same sub-section. Good in some cases, but not really convincing, plus it sucks all the cpu power.

    Some offsetting, though, is definitely worth to be looked after. So, not offsetting the violins of a violin track, but offsetting the different sections. 1st and 2nd violins, Cellos and Violas... strings and brass and so on.
    I have found a very interesting sfz specification which is the "delay_random" parameter. It is worth some experimentation.
    Just open an instrument sfz file with your notepad (usually, in Windows, Program files>Garritan>Personal... >Instruments. Scroll down to the "global" section and add among the other parameters a line like: "delay_random= N" (without quotes, of course). Where N is a floating point number from 0 to 100. Never write 100, those are seconds
    0.1 and probably much less would be fine. Now load GPO, load the patch you have modified and it will have now a random behaviour with concern to the note on parameter, that's when a note gets invoked for playing. Very useful, and works.
    It is really only time devouring to modify every single note by hand...

    I've been experimenting with these kinds of behaviour as well, in small ensembles. The beauty of a "human" ensemble is not only the random human behaviour, but the fact that, take a small violins section... those are 5 different violins with 5 different frequencies ranges. Those sound very different, they are not "Violin player1, 2 and 3". In the real world, frequency spectrum in output is much wider and richer than what you usually get with samples.
    You can buy the most expensive library, but still you'll get one solo Cello. In this, GPO rules as it gives a few more
    Listen to this. A small extract of Vivaldi ("my string testing device" ).


    These are 4 different violins, 3 violas, 3 cellos and 1 Bass. All different, plus each patch is programmed with a random delay (and more random controllers). One set is Philarmonia (free samples), one GPO's and one to another solo strings library I own, plus one violin from a Cakewalk expansion pack.
    At least to my ears, it sounds pretty rich and convincing. During the violin runs, you can "feel" the different violins and the way they add up in the texture, and you can hear that there are two at the beginning of the run, and 4 on the last scale.

    Just experimenting. Always worth....


  5. #5

    Re: Getting a better rendering

    Add up to previous post.

    Sfz, free standard.... that means it is easy to play with it.
    So, we were talking about that "delay random" parameter. I went to check after writing about it, and found an even better one. If you want to play with this kind of stuff, follow me.
    Let's say we want to add a permanent controller of the "start of sample" delay, adjustable and mappable to your DAW/keyboard.
    We will do it for the Gagliano, Violin 1 patch. So, in windows, go to program files>garritan>personal orchestra 4(or whatever version you have, I suppose)>Instruments>solo strings. We are going to make some changes on the "Violi Gagli solo KS".
    BACK UP the orginal file! I didn't, but you better not take the risk. Just copy the file somewhere else. Backed up? Ok, let's move on. Open the file with notepad. Find the "CONTROL" section, right at the beginning. See all those "set" lines? Well, add another one:


    Don't think cc31 is already assigned for something.

    Then, move slightly further down, "global section", right before Master - sustain section. At the end of it add the following line:


    Save the file and close it.

    Now, go back to Program files>garritan>personal orchestra 4 and move into the "GUI" folder.
    Back up the file "Violin1", then open it with notepad.
    Here we have lots of XML blocks starting with the tag "<StaticText".
    Scroll at the end of the file, right before the closing tag "</GUI>", and copy paste this other block:

    <StaticText x="420" y="215" w="75" h="18" text="Offset (CC31)" color_text="#CCCCCCFF" />
    <Knob param="31" x="440" y="233" image="knob_controls.png" frames="120" inverse="1" />
    <Label param="31" x="440" y="275" w="50" h="18" format="%3.f%%" offset="0" factor="100" color_back="#FFFFFF00" color_border="#FFFFFF00" />

    Save the file and close it.

    Now open GPO, load the solo violin 1 patch and you have random offset for it, adjustable via the newly appeared knob (I named it "Offset") on the control section and usable as you would use cc1, cc23 and anything else.
    Hey, do not try if this scares you.
    If anything goes wrong, just copy the two backed up files back in place, overwriting the modified ones.


    Ops... UPDATE. With this last method you'll get just adjustable delay. Not random. For random, refer to the first parameter quoted.

Go Back to forum


Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts