More natural = irregular? Playback predictability and imitating real musicians for the sake of it
Hi,
let me state upfront that I appreciate the work that people put into MuseSound, a free product, even if it may not be for me. :)
I just tried Musescore 4 and MuseSound again after a year, to see how it has improved. I love using Musescore 3 with oldschool soundfonts, and I cannot seem to wrap my head around MuseSound, and figure out what the overall design intention is.
Example:
I have 4 quarter notes in violin, pizzicato, at ff. When I play it back, the third one is a Bartók pizz, the others are much more quiet. Huh.
Is the goal to simulate the natural chaos that real musicians introduce? I was hoping to avoid that by staying on the computer. :D
This is meant as a more general question, I'm curious what people's goals are with playback. For me, it's most important that I know what I'll get when I type in some notes; reliably getting something that doesn't sound so good is preferable to erratically sometimes getting something amazing and sometimes a complete mess.
Another example: Very long held notes on a solo string instrument. A real musician will train very hard to make the change of bow as unnoticeable as possible. Meanwhile the software deliberately adds it, despite being technically capable of just holding indefinitely. I'm wondering ... why? Shouldn't we be glad that computers can do things that humans cannot?
I have never used high-end playback software like NotePerformer, so I don't know whether this approach is universal.
My best musical results so far have been with the "expressive" sf3 soundfonts that came with MuseScore 3. It's amazing what you can do by tinkering with the dynamics of a single note, splitting it up in many notes tied together and then adding > and < between them. Very time-expensive, but I feel like I have control over how it will sound, unlike with more advanced tech.
Comments
Interesting. I just tried to recreate your two examples. I got nothing like what you got. Several measures of ff solo violin pizz were all even. As was a 5 measure long note.
This is not to say there aren't quirks in Muse sounds.
NotePerformer is certainly not high end.
We each get to use the software that suits us.
In reply to Interesting. I just tried to… by bobjp
It depends on the pitch apparently. A3 pizz will have one Bartók pizz per 4 notes, as does D4; A4 will have one in four be 50 cent higher ... it seems like there are always 4 samples that are iterated through, and sometimes they are similar enough to not notice, sometimes they introduce intentional errors.
50+ cent error is significant, that's a different pitch. Odd choice for samples.
Additionally, since it is every 4 notes, you get a 4/4 rhythm no matter what the actual time signature says.
As for noticeable bowing changes - I can hear it for all longer held notes I tested, it's built in. It's not the most obvious, but there are always some small gaps. Usually not a problem, but it is a design choice to introduce a human limitation where there was none before.
I have no idea what current high-end is :D I can't afford it anyways.^^ But do you know whether they also work like that? Intentional errors to make it seem more human?
In reply to It depends on the pitch… by jundurg
I still can't reproduce this. Please post a score where this happens. So far I can't say intentional errors have been introduced in MuseScore. None in Sibelius. If you think about it such a different sound at a regular interval is hardly more human.
In reply to I still can't reproduce this… by bobjp
Alright.
First and fifth note is too high, then we get some notes with sharper (Bartók-) pizz; finally the held note has about 4 subtle bow changes.
(I like the bow changes actually in this context, it does sound more natural. But of course, it depends on the situation, and sometimes it might be just off.)
I'm surprised that you couldn't reproduce this. I've assumed from my testing that MuseStrings is just built like this, having 4 samples per note and switching between them. Sometimes the differences are more noticeable, sometimes very subtle.
In reply to Alright. First and fifth… by jundurg
Indeed, at that volume level Violin solo 1 has problems. What's worse is that if you remove the pizz, the D4 measure is played as a whole note.
But it is only the solo 1 voice that does this. At mf, it doesn't do it.
The difference between any of the 1 or 2 sounds is that 1 is more expressive than 2. It has nothing to do with desk seating. I don't use 1 sounds because they tend to be too sloppy. Your example just adds to the reasons to not use that sound.
I don't think it has anything to do with programming intentional human error. But it isn't quite right.
You can report this on Github if it bothers you. Me, I just don't use that sound.
In reply to Indeed, at that volume level… by bobjp
Ah, Violin solo 1 seems to be the odd one out indeed. My luck for starting my tests there. :D
I went through all MuseStrings with my example; the pattern of having 4 different samples per sound seems to hold up, however the amount of variation is far less extreme and sometimes I can barely hear it.
Tutti pizz sounds are deliberately a bit messy to be more realistic (as it is very hard for an orchestral section to all twang a string at exactly the same time) and there is a bit of variation there. With the solo strings, the changes are almost too small to notice. I don't remember which one, but I think one of them had one of the 4 pizz notes slightly higher (but I'd estimate it in the 15 cent range, so barely noticeable)
Longer held notes have some amount of error, but I was only able to hear a slight mistuned attack on the violas (admittedly I only checked a single note), and it sounds like on the solo instruments, avoiding the bow change seems to work by overlapping two sounds very slightly. Not something I'd notice normally. (And this one imo doesn't fall into the category of intentional variety.)
I hope we'll at some point hear more about the design philosophy behind MuseSounds, it's not like I disagree with the idea of introducing some natural variety (that is one of the main advantages to normal soundfonts after all).
In reply to Ah, Violin solo 1 seems to… by jundurg
Personally, I'd rather no attempts be made to humanize playback. Real players don't purposely make mistakes. Especially at random times. Muse sounds are always being up dated. I think the things you discovered are results of mistakes in the programming. When muse sounds were first introduced, they were different than they are now.
FWIW, I think the oboe could use some work. it is a bit harsh sounding. And the trumpet could use some improvement. It is very sloppy, and very thin as far as tone goes. I do like the horn