Music Composers Unite!
Don't get me wrong, NotePerformer has many excellent qualities and when it works right, it produces beautiful demos. But in certain contexts, its design causes a serious problem: loss of ensemble (synchrony) between solo instruments, especially solo strings in a chamber setting.
This is not a problem with the host notation software that NotePerformer is running under: Wallander Instruments says it is a side-effect of their building in random timing delays in their playback routines for strings, because individual players in a string section are always slightly out of sync with each other and that contributes to the realism of the sound. But unfortunately they do not special-case solo strings to make the timing more precise, even though that degree of asynchrony would be unacceptable in chamber music.
I've attached an mp3 of the beginning of something I'm working on, created by NotePerformer running under Sibelius. You can hear clearly that the viola is lagging the 2nd violin by at least 1/10 of a second in some places. This is horrible in my opinion.
I wonder if anyone else has encountered this problem and maybe found a hack to suppress it?
Could you post some score so we can experiment with this, possibly compare results?
I tried several times to post an xml file, and then the .sib file, but each time the site gave me an error "we're having an issue with that page". So it seems it's going to have to be a pdf, as inconvenient as that is to get into notation software. Sorry about that.
No problem, probably best to start from scratch anyway. I'm pretty busy today, perhaps other NP users might want to give it a try. Probably only need to do a fragment to see some results I'm thinking. I've never done a string quartet in NP so it will be interesting. I'm also wondering what effect different tempos might have.
Thanks Ingo. It will probably be enough to just copy the first fugal entrance of the subject in the violin, with the counter-subject in the viola.
Gerd Prengel has posted at least a couple of his string quartet works here, and he has said that he uses NP under Finale. You can hear the asynchrony from time to time but it isn't nearly as bad as what I struggle with under Sibelius. I wonder if a random number generator is initialized differently under the two notation softwares, or something like that.
And: tempo makes no difference - I've heard it happen, very noticeably, in the fastest parts of the string quartet I posted a couple of months back. That demo was patched together from several playbacks using Audacity. Even the Beethoven sketch and first variation needed some work in Audacity to make a listenable rendering.
Ingo Lee said:
So Liz are you exporting mp3's of individual instruments from Sibelius into Audacity and then slip editing one track to align vertically with the others and then rendering the mix for a final output? As a work around in other words?
No... though that's certainly something one could try! What I'm doing is pasting together excerpts from complete renderings of the piece (with the full ensemble) that each has no or minimal timing errors, There's only so much you can do that way unless you're okay with unwanted noise - the edit points always have to be at a silence or there will be an audible (often quite loud) pop or crackle.
Ingo Lee said:
In Sibelius when I choose String Quartet from the template page the score defaults to Violin I and II which is what your pdf score is showing. Have you tried switching to Violin 1 or 2 or Violin (solo) or Viola (solo)? And while we are at it I'm assuming you have switched Sibelius from its default library to NotePerformer?
In my String Quartet I did choose Violin (solo) etc., and it made no difference for this problem. And of course it's NotePerformer... the default library just sounds like MIDI "instruments".
And BTW, it's instructive to play the score back using the Basic configuration - the timing is VERY precise. Of course it sounds quite awful.
Ingo Lee said:
Here is a clip from your quartet Liz and I have added four measures of 1/8 notes that don't fit (sorry) to highlight timing differences. I set the reverb to zero and the panning knob to zero. This is mecchanico and I ran the tenuto plugin. When I look at this clip in a DAW the 1/8 notes seem to line up better than when the reverb and pan controls are on. It is difficult to see the vertical alignment in your quartet so I added the 1/8th notes. Listen and see what you think, your ears are better than mine. One thing I notice is that the viola notes have a slower attack than the violin notes which could cause a difference also.
Thanks for your work on this. I agree that the timing sounds very precise in that clip of yours. So I did some experimentation of my own. What I found is that the setting that makes the biggest difference (by far) is meccanico. I'm really surprised at myself that I never tried that before as it is an obvious thing to try. I had thought that setting Rubato to meccanico would be the only thing that might affect it. So, thank you for that!
Removing reverb helps a little bit, but I'm not sure that's anything but an electronic effect anyway. Setting pan to zero does seem to reduce the problem even more, but (1) I really want spatial separation in my demos and (2) I think that's really an aural illusion: it's easier for the ear to detect timing discrepancies when the sounds are spatially separated.
Also: your added notes don't really tell us much (sorry). The timing issue almost never occurs on bow strokes played together. Here is how Arne explained it:
"Yes, it's likely that you will experience greater timing discrepancies
with legato notes. String players have a much greater timing accuracy
with notes that are triggered by bow strokes, as opposed to notes that
are triggered through fingering."
So you can see that the problem is really a side-effect of their trying to simulate human performance factors as closely as possible!
Even with meccanico, I still notice little timing "flubs" once in a while, but nowhere to the same degree. And they don't always happen in the same place in the score because they're random - also because of trying to simulate the human element. (Also, I find both the viola and the cello tend to be the "laggers" for some reason - but not always, as it's random behavior and one or both are sometimes ahead of the others.)
I wonder why the volume level in your clip is so much higher than in my demos? Maybe there is another Sibelius setting I'm missing?
Ingo Lee said:
Okay I'm going to take back most of what I said in my last post... it looks like this is all random chance. I just produced a clip, even with all the settings you suggested, that is still full of timing errors. I will bet that if you play back your score several times you will eventually hear one instrument lagging behind again. I've attached the new clip.
But here's what I really can't figure out: one of the Sibelius project files is the finale of the Elgar String Quartet. It is full of fast legato passages. When I play that score back I NEVER hear even the slightest timing issues, it all sounds perfectly precise. I figured it had to be some setting saved in the file. So I tried deleting the music from the project file and replacing it with my own, without touching any of the settings... but once again, I hear the same timing errors. Not all the time, and not always in the same places. But roughly as frequently and to the same degree as with the settings in my original .sib file.
I'm going to ask Arne what the reason could be. I've a sneaking suspicion they "trained" their AI algorithms for string quartet on that piece so that it plays back correctly. But that doesn't gibe with what he told me that they use the same timing algorithms for solo strings as for a string section. Another possibility is hidden directives in the score - when I imported my piece into the bare staves, at one point there was a sudden tempo change without any visible marking in the score. Replacing the bars themselves fixed the problem.
Something about this is very weird.