Human Performance

• Jan 7, 2017 - 19:15

I downloaded a free trial version of Finale last night in order to test it out and see if it actually lives up to its name. In it, there's a setting called "Human Performance", in which you can make your score sound more natural.

Does Musescore have something similar? If not, will it?


There are several improvements being worked on for 3.0 such as mid note dynamic changes and slurs sounding like slurs. These will go a long way toward making it sound like a human performance. I would dare say that Finale probably sounds better than MuseScore, but if you listen closely I'm sure you will find defects in playback.

In reply to by mike320

The main reason playback does not sound human is rubato. If you could imitate that and had only one dynamic level the music would still sound more like from a--lazy but human--musician.

The problem is that rubato and its exact handling depends on the genre of music you want to play. Not just is e.g. classical music different from jazz but even Mozart and Brahms are played differently. So automation of this is really tricky.

In reply to by azumbrunn

Very good, very insightful, Thanks for verbalizing this.

I have been studying my own (piano/keyboard) playing of my writing for the last while and have been gaining more insight into the numerous subtle time and rhythm adjustments I constantly make, exploring different shades in the same piece(s). Impossible to catch the subtlety in a form other than performance.

Notation is just too clumsy a way of trying to realize and analogue experience in a digital form, no matter how finely drawn.

If you want it to sound like humans there's just no getting around having to use humans (which I think is great!).


In reply to by azumbrunn

Even in pieces that are meant to be played in strict time, there is usually some natural imprecision (both deliberate and otherwise) that characterizes a human performance sound. Also some variation in volume above and beyond what the dynamics specify - bring out the melody slight, putting a bit more emphasis on the strong beats or on syncopations, etc.

Right now we do provide means for tweaking these things, via the Pianoroll Editor & Inspector. I suspect a plugin - or perhaps eventually a built-in tool - could do a decent job of automating this without any major new underlying facilities being needed.

In reply to by Marc Sabatella

Obviously tools for manipulating rubato effects would allow to imitate more closely a human performance. But I would not call this "automated". If done well it is a lot of work which has to be done by trial and error.
I am not sure this is right but here goes: Our bodies have all sorts of rhythms going, breathing, walking, heart beat, nerve cell firings, sleep wake cycle etc. And I believe that the deep connection we develop to music is at least in part due to those rhythms in our physiological routine. Such that teaching rubato to a computer is probably one of the more difficult AI problems around.
This connection I think is also the reason for the constant emphasis by many teachers on breathing, even for musicians who don't play a wind or brass instrument and who don't sing.

In reply to by azumbrunn

I would like to restore this interesting thread about human playback. Your post ( azumbrunn • Jan 8, 2017 - 02:38) is the one that MuseScore developers should consider while implementing such a feature (assuming it will be done...).

In my opinion automation here does not make sense at all. Why don't make this totally MANUAL for the human score writer? You can think at adding a sort of "overlay" score that is only designed for tuning of tempo and dynamics, even note by note, or just at selected points. This additional score should not interfere with the real score but can be overlapped (eg. in a different color) with the option to view/unview it or disable it, of course. In this score view, you should be able to do things like: add pauses where main score does not explicitly states; slow down tempo at the end of a trill; begin a trill slower and progressively speedup it as usual in baroque performances, and so on...

I'm not a musician nor a developer, but I'd like to post this as a proposal for a feature, what do you think?

underquark directed me to this thread. Using an audio editor, I did an cursory examination of a digital wave pattern and a human performance. I was surprised at how perfect the digital patterns were compared to the live human performance. Looking at the human performance, I roughly estimated that a three or four minute performance contains about one million irregularities in pitch, duration, volume and timbre. There are a multitude of spikes and lags and dips in every millisecond of sound.

And that's the charm of a human performance - the struggle against those irregularities. You would think that the listener would jump for joy to hear near-perfect wave patterns. but not so - we seem to derive our enjoyment from hearing the energy of that struggle as it's happening.

If only I could introduce all of those irregularities back into the digital wave forms. But that would defeat the convenience of digital. It would take forever to deform the wave pattern a million times. Would it not?

MS is for music notation and not playback in audio form.

[EDIT: "Uh, everything's under control, situation normal..." "--Who is this?" ...Anyway, I can't reproduce the situation with any of my soundfonts, so it must have been a transient issue of some kind with my computer.]

I thought that MuseScore 2 was already doing this, I have noted that when I repeat four of the same note, every fourth note is slightly muted.

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