MS4.1.1 Playback peculiarities (Muse string sounds)

• Jul 27, 2023 - 18:13

From a string orchestra piece I am working on, I've cobbled together the attached samples of problems with the playback. The unwritten (and unwanted) dynamic changes particularly are vexing. I don't want my wonderful notation program of choice try (and fail) to be an interpreter as well.

Regards, SK.

Attachment Size
Playback_examples.mscz 30.24 KB


Yes, the score as you present it, has those problems.
However, when I C+P lines into a new score, most of them don't happen.

The cello delay to the low G doesn't happen. Swells over bar lines don't happen.

Measure 20, 2nd violin. When I C+P this phrase into a new score (and change it to solo violin) the p dynamic in you score is louder than in Mine. Also your solo violin sounds more like Basic than Muse sounds.

The cello cresc starting in measure 27 does happen in a new score. Perhaps MuseScore can't correct for that high range the same way a real player would. No idea.

The strange note over the Ab does happen. But not if the note before it is anything but a C. Should it happen? No. But such is life with recorded sounds. They don't always play nicely together.

I'm not making excuses for MU4. Just reporting my findings. And this is 4.1.1.

Some of what you describe I can hear, some I cannot, and some seems expected to me. Overall, an important thing to keep in mind is that Muse Sounds in particular does aim for realism - which means natural variations and interpretation. if you wish to hear less realistic playback, you can simply switch to MS Basic or another soundfont or a VST instrument. but some of what you are hearing do appear to be actual bugs not just subjective differences in interpretation.

1) The "wooden clacking noise" you refer is, as far as I can tell, the natural sound of the string snapping against the fingerboard when playing with the hard pizzicato attack that you asked for via the dynamic and articulation you indicated. This seems appropriate and very realistic.

2) The "flanging" seems to be some sort of faint high frequency overtone. Hard to say if that's a natural effect of the slow bowing or a background sound in that particular sample. Either I recommend opening an issue on GitHub to report that so the developers can look at cleaning it up.

3) I'm not convinced the swells are anything more than natural variation when you ask real musicians to play a long note, which is of course exactly what these sounds represent. Again, if you want less realistic playback, switching to a soundfont or VST should achieve that.

4) The cello glitch going from E3 to G2 at the "p" dynamic is definitely an issue, something is apparently wrong with the sample for that specific pair of notes at that specific dynamic level (note changing either pitch or the dynamic makes it go away). Definitely report on GitHub.

In reply to by Marc Sabatella

Mr Sabatella,

Thanks for the informative response, even if it was not what I was hoping to hear.

re: points 2 & 4: What is Github? And why the need for duplicate reporting? Isn't this the forum to report bugs to developers? If not, then what is its purpose?

re: points 1 & 3: I cannot disagree more, and find it a very troubling shift in the development of Musescore/MuseSounds if it starts attempting to be a bad interpreter of music notation than a faithful presentation of a score. If Iwanted swells or crescendos I'd mark them. It's faintly ridiculous to have to fight a notation program over "subjective difference in interpretation". There is no subjective. There is simply the score. Is there really no work-around to stop MuseSounds trying (and failing) to determine dynamics or other unwanted effects other than to employ different soundfonts?

Regards, KS.

In reply to by KSaul

GitHub is the place where you report issues to developers once they have been confirmed by fellow community members here in the forums to be a) actual bugs, b) reproducible given the steps provided, and c) not already reported. The idea is to keep the developers focused on development, and let the community work cooperatively to make sure only actual confirmed bugs are passed along to them.

Regarding the subjective interpretation that's fine if you prefer less of a "human" and more of a "synthesized" reading of the material. As I said, you can get that easily, by switching to MS Basic or another soundfont, or VST instruments. So it's the best of both worlds - those wanting absolute precision with no variance whatsoever can have that, and those who prefer a more lifelike interpretation can have that. No need to fight anything at all - just select the MS Basic in View / Playback setup. and set it as the default for new scores.

In reply to by Marc Sabatella

Some observations.
The clicking sound in the pizz cello does need to be reported. It only happens on the upper A note. Remove it or change it to a different note and the sound is not there. If it were somehow an accurate recording of that note played that way (seriously?) then you would hear it on every note.

Muse violin sections tend to swell on half notes at mf. This is not subjective. So much so that I have to use the viola sound. Sorry, but real players do not play that way. It isn't human. It's un-musical. Muse Sounds are higher quality recordings than Muse Basic. But far from perfect. We have to use Muse Sounds If playback is important. And, yes, there can be a struggle to get the desired result.

In reply to by bobjp

Just to add to this: I tried the solo violin from Musehub. The sound includes vibrato and it is varied (as a good player would do). Two problems with this: At every change of pitch the vibrato does not fit (a sound sample can't know which pitch will follow it) plus the fairly substantial variation in vibrato without any musical sense is the opposite of realistic and simply awful. I far prefer the muse basic fiddle. No vibrato, not realistic, but bearable.

In reply to by azumbrunn

I assume you mean the violin 1 sound? This was designed to be the more "expressive" sound. For a straighter sound, use violin 2. Dn't think of these as literally only meaning first & second parts in an ensemble; they could just as well have been called "more expressive" and "less expressive".

But also - as far as I know Muse Sounds does know the pitch of each pair of notes and in fact sampled not just individual notes but also pairs. So in theory things should work well here. If you have a specific score with a specific pair of notes where you hear an audible artifact (as there is with the E3 to G2 pair at "p" dynamic in the cello here), best to report it to GitHub so the developers can investigate.

In reply to by Marc Sabatella

Hello again.

"Regarding the subjective interpretation that's fine if you prefer less of a "human" and more of a "synthesized" reading of the material."

I cannot get my head around this assertion. The problem is that the soundset presents a single preset interpretation of events contrary to markings. There's plenty of variety in "human" interpretations. By dictating a single, intrusive, overriding approach MuseSounds becomes **more* synthesized, not less. It's also a bad faith approach for the core (and highly promoted) SoundSet of a notation program to disregard the notation! What on earth are the developers thinking?

"I assume you mean the violin 1 sound? This was designed to be the more "expressive" sound. For a straighter sound, use violin 2. Dn't think of these as literally only meaning first & second parts in an ensemble; they could just as well have been called "more expressive" and "less expressive"."

This is super useful to know. Is it pointed out anywhere else?

Sincerely, KS.

In reply to by KSaul

I'm not sure what makes you assume there is only a single interpretation built into Muse Sounds. It can vary according to many factors.

To be clear, though: no one is "disregarding" the notation. Just trying to play it in a realistic fashion, which means some natural variation in amplitude etc. Is there room for tweaks to improve here and there? Sure, and that is why opening GitHub issues is useful - to report specific opportunities for improvement to the developers.

But the basic goal of achieving realistic / human-sounding playback is absolutely sound; it's just a matter of tweaking how that goal is achieved.

As for the violin 1 vs 2 difference - it's been discussed in various other threads here and/or on other channels based on statements from its developer to that effect made ion GitHub etc. I don't think there is any specific official documentation on this, though.

In reply to by Marc Sabatella

I think the debate is about the meaning of "realistic". To me "realistic" is a target that can not be hit at this time. At some point in the future, using immense AI resources, it may be possible to produce sound like a good recording by human players. For now we have to put up with what is possible.

For now my main concern is about dynamic markings that can not be modified, e.g. "sf", which can not even be set to "don't play" in the properties dialogue. I am still hoping that a future version will bring back the flexibility that Musescore 3 offered. I used to put up scores for IMSLP on but with the present sound quality I can't. A sforzato may induce a heart attack in some individuals.

About violin 1 and 2: I admit that, once informed, I thought I should have figured this out myself. Generally we should not assume that developers of sound fonts are deeply immersed in the classical music culture.

In reply to by azumbrunn

"Realistic" is of course subjective. And that's the crux of it I guess, different people find different things realistic. To me hearing subtle variations in volume etc is quite natural and very welcome, and obviously the professional musicians who designed this and the folks who have written such glowing reviews in the press about the realism of Muse Sounds agree. And these are people very stereped in orhcestral music. Again, it's subjective., and you and others have every right to have different subjective preferences. But implementing something that meets the needs of an incredibly large number of musicians is not a bad thing, even if it means some might prefer something different.

To some extent you can get that effect even within Muse Sounds by applying a compressor effectk so you could consider that.

Not sure what you mean about sf, but indeed, future versions of MsueScore will have far more sophisticated controls than ever before over the dynamic shaping of your score, with DAW-style automation lanes etc. As for posting elsewhere, please try to keep in mind that not everyone will share your subjective impressions, so just because you don't like a particular playback effect, is not a good reason to deprive the IMSLP and community of the score. Some in that community will love the effect that you don't (again, it's all subjective), and a great many more simply won't care because they are after something other than just playback when they download the score (I use these for want to use it for educational and arranging purposes all the time, for instance). And of course, you can always simply use MS Basic.

In reply to by Marc Sabatella

>>> "Realistic" is of course subjective.

Then considering there are multiple performance traditions and interpretative styles, it would be wiser if the baseline sound presentation did not attempt to reproduce only a single (or a few, as you claim) style. Enforcing a specific form of stylistic playback devalues the utility of the program.

>>> "To be clear, though: no one is "disregarding" the notation."

This is incorrect, and it is at the very heart of the problem.
I didn't notate crescendo-diminuendos. Yet the playback added them. The notation was very clearly disregarded.

And I realise, as you've been at pains to make clear, there are other soundfont options that can be employed. But when a flagship component of the suite approach (MuseHub! MuseSounds! MuseScore!) is fundamentally flawed, that's an awful shame. Not to mention, a terrible waste of resources. What a shame.

Sincerely, KS

In reply to by KSaul

As I keep saying, it's all subjective. I get that you personally find something about the realism of Muse Sounds objectionable. So simply use a different sound set, and let the millions of others enjoy the amazing results of the hard work of the Muse Sounds team - nothing for anyone to be ashamed of.

In reply to by Marc Sabatella

>> It's all subjective.

Notation software should never be subjective.

>> So simply use a different sound set, and let the millions of others enjoy the amazing results of the hard work of the Muse Sounds team

This is no solution to the fundamental flaws with MuseSounds, and I'm quite sure you realise it. Stop being such a PR cheerleader and contribute something useful.

In reply to by KSaul

Playback is always subjective. No two players will play the same passage the exact same way.

Anyhow, if you are encountering some sort of fundamental flaw, please be sure to report them separately, with sample score and steps to reproduce the problem.

In reply to by Marc Sabatella

By "sf" I mean the dynamic marking "sforzato". Take a quiet piece of music in musescore 4 , mark it piano overall and then add "sf" to a note, preferably at least a minim, and listen. I guarantee you that you can not possibly call the result realistic. The extra problem is that I can not even turn off "play" in the properties dialogue like I can for staccato dots and marcato signs. And I need to keep the "sf" markings in the score; at the end it is supposed to be for people who play the music (and I don't want to mess with the composer's intentions).

I was perfectly happy with the options in MS3 in this regard; I don't mind doing it by hand: I then get to decide on the nuance rather than the machine. I suppose there were sound technical reasons why that solution could not be kept for version 4. In MS3 I could listen and make the sound perfect "in my head" as they say. This is no longer easy with the disruptions I get from the immobility (and sometimes just incorrectitude) of the markings.

In reply to by Marc Sabatella

As I said do the experiment. Or let me explain: If I enter a sf in a piano surrounding it blares out ffff. If it is a long note it blares out the entire length of it. It is truly shocking. In fact it is shocking even in a forte environment. BTW "fp" on the other hand is correctly stressing only the beginning of the note but it is too feeble.

It is not a matter of a specific score. It is ugly every time everywhere.

In reply to by azumbrunn

It is most definitely not every time everywhere - I just tried it and it worked as expected. So pleas,e if you want someone to investigate, please attach a specific score. Then if someone can confirm the issue unique to that score, you can open an issue n GitHub so the developers can take it from there.

Seems this is somewhat related to the above: something very strange is going on with MuseSounds Violas written Ab4. See a few measures attached demonstrating the problem. Depending on duration of the Ab and the note that follows it, there sounds a ghost Gb4 on top of it - not very pleasant when trying to move to a chord that has a G natural. Would appreciate validation of problem encountered. Thanks in advance for your help

OS: Windows 10 Version 2009 or later, Arch.: x86_64, MuseScore version (64-bit): 4.1.1-232071203, revision: e4d1ddf

Attachment Size
debug.mscz 17.35 KB

Do you still have an unanswered question? Please log in first to post your question.