Piano playback: das große F klingt immer staccato

• Jun 25, 2021 - 06:44

Guten Morgen!

Beim Verwenden des Grand Pianos (MuseScore General SF, auch HQ) erklingt das "große F" im Bass grundsätzlich staccato (auch mit Pedal).
Kann mir bitte jemand helfen? Dieses "Schluckauf-F" macht mir sämtliche Wave-exports für Youtube etc. kaputt...
Besten Dank vorab und viele Grüße!



"When using the grand piano (MuseScore General SF, also HQ), the "big F" in the bass is basically staccato (also with pedal). ‎
‎ Can someone please help me? This "hiccup-F" makes me all wave-exports for Youtube etc. broken... ‎
‎ Thank you in advance and many greetings!‎"

Please attach the score, or an excerpt, demonstrating the issue. Point out the exact location of the "big F" -- i.e., staff and measure. Also mention the exact soundfont we should load to listen to the attachment.


In reply to by Jm6stringer

If I understand it correct, with "big F" is meant the Helmholtz pitch notation (https://en.wikipedia.org/wiki/Helmholtz_pitch_notation) which corresponds to F2 in scientific pitch notation (https://en.wikipedia.org/wiki/Scientific_pitch_notation).

@polyphonicus: see attached example. I can't hear a staccato there. So as mentioned, best way is to attache the corresponding score and describe, in which measures it doesn't sound as expected.

Attachment Size
F.mscz 5.93 KB

Dear Jm6stringer, dear kuwitt,

thank you very much for paying attention to my post and for your answers!!

I guess that I chose the wrong expression when I wrote "staccato". It is not a real staccato but - at least for my ears - quite a different sound. Actually it affects F, E, Eb and somewhat D and Db. These notes appear to be "strangled". What I mean is that these 5 notes do not have the full sonority, do have a shortened climax an immediately fade away. The notes before and after (C2 and Gb2) do have the full sound.
Sorry for my english - I hope that I have expressed myself in an understandable way...



In reply to by Jojo-Schmitz

Hallo Herr Jojo-Schmitz,

tatsächlich benutze ich noch MuseScore 3.1, allerdings mit dem HQ-soundfont. Könnte es tatsächlich an der Version 3.1 liegen?
Beste Grüße und vielen Dank!


Hello Jojo-Schmitz,

indeed I still use MS 3.1, but with the HQ-soundfont. Could version 3.1 be the reason for my problem?
Thanks & regards!


In reply to by Jojo-Schmitz

I thought that maybe at one of the company parties in Las Vegas while you were smoking a good Cohiba on the roof in the pool with the CEO, you could share my ideas with him. ;)
All jokes aside, as a music teacher I understand and know what you mean. Nevertheless, thank you very much for your time and best regards!

In reply to by Polyphonicus

Ich hör’s auf jeden Fall, in 3.2.3 mit dem HQ-Soundfont, aber nur da, wo die Pedalmarkierungen mit ins Spiel kommen. Beim A₂ ein bißchen, beim G₂ dann nicht, und beim F₂ stark. (Die Helmholtz-Notation hab’ ich nie gelernt in Musik, ich kann auch nur SPN.)

Ich hab’ jetzt erstmal ein Meeting, kann da aber nachher mal ein bißchen nachforschen (in den Soundfont reingucken, mit verschiedenen Versionen ausprobieren, usw). Dadurch, daß es nur bei den Pedalmarkierungen zum Vorschein kommt, liegt aber der Verdacht nahe, daß es sich um einen Fehler in der Software handelt, der u.U. in einer neueren Version bereits korrigiert wurde. Das kann ich aber testen.

Update: Ich hör’s jetzt auch in der nicht-Pedal-Version, aber nicht in TimGM6mb zum Beispiel.

It’s audible to me (3.2.3), but only in the line with the pedal markings. I’ve got a meeting in a bit, but I’ll investigate later. Suspecting a, possibly fixed in a later version, bug in the synthesiser.

Update: I’m also hearing it without the pedal markings, but it seems soundfont-dependent. Investigating.

In reply to by mirabilos

OK, this looks like a soundfont issue indeed.

It’s interesting how well you figured out which samples belong to which notes. There are two samples in play, each in “Piano MF” and “Piano FF” variants, each variant in (L) and (R) stereo channels. The sample “Piano MF/FF D1 (L/R)” is used for the notes C♯₂/D♭₂ and D₂ (which you coloured orange for semi-bad), the sample “Piano MF/FF E1 (L/R)” is used for the notes D♯₂–F₂ (which you coloured red for fully bad); all other notes use different samples. (The octaves used in denoting pitches in this particular soundfont are off by one by historical accident and fixing it would not be worth the effort.)

Inspecting the preset “000:000 Grand Piano” in Polyphone shows that the sound is comprised of the instruments “Piano MF-low” and “Piano FF-low”. When I play the F₂ (or other affected notes) in those instruments (using Polyphone’s keyboard), they sound much softer and better than when I play them in the preset, which has a harsh, metallic sound (which is probably the sound that happens when the key hits the piano frame hard, which is why it’s worse in the FF version).

In MuseScore, playing your example score at mp instead of mf seems to soften it a bit, too.

So I guess this has to do something with the parameters on the preset. I’ll mail the soundfont designer and see whether he has a way to make this sound less bad.

In the meantime, we can investigate other soundfonts. MuseScore_General (MuseScore 2.2 and up) uses the “Splendid Grand piano from AKAI S5000” in both the LQ and HQ versions. Its antecessor (MuseScore 2.0/2.1) Fluid (R3) Mono sounds even worse for the D₂ but a little better for the F₂. The MuseScore 1.x soundfont TimGM6mb has a surprisingly nice F₂ but its D₂ and lower degrade as well, it might be an option. Or something else entirely.

Depending on how much emphasis you place on music playback (as opposed to music notation), be aware that soundfonts are constructed from samples of actual instruments.

Example: for piano:
(There you will see names like Yamaha, Baldwin, Kawai)
So, each different soundfont may be more (or less) consistent across its pitch range from low to high.
You may discover one with a 'große F' that you prefer.


I exported the file with MuseScore 3.05 (I keep some older versions for reference) and checked both waveform and spectrogram (listening with MS 3.6.2 I don't feel noticeable difference). In the waveform you can see there is a decay approximately equal to that of the other notes. But in the spectrogram you can see that the note mainly consists of the base harmonic and some harmonic of higher degrees, what doesn't happen at this same level for the other notes. F and Fb are using the same audio sample, this is why they sound the same.

What is happening? This region of the piano has 2 identical strings per note. When they are vibrating in the same phase, there is a constructive interference, which we perceive as an increased volume. When they vibrate in opposite phase, there is a destructive interference, which sounds like the volume decreases for some harmonics. This second case is happening, but actually the volume decrease is very little. How do you "solve" that on an actual piano? With fine tuning on one of the twin strings. But this is a virtual piano with recorded samples, so there is nothing to do except for:
1. Using another sample or readjusting the sample regions in the SFZ/SF2 file (you would need to edit the virtual instrument file)
2. Using another virtual piano instead
But there is actually nothing wrong with that. You'll never find perfectly tuned piano in the real life so there will always be some "effects" like this, it sounds natural. Even professional virtual instruments have those little tuning offsets to sound more natural. But if this is really annoying in this case, better to download another piano SFZ, there are quite a few available, like already reported.

In reply to by Ludwig van Benteuer

I am the designer of the MuseScore_General SoundFont. Ludwig is right. The issue lies in the characteristics of the sampled instrument in combination with how it was recorded. The original sample set that I used (AKAI "Splendid Grand") features samples for more of the keys, but I removed them to save RAM. I will investigate if there are D#2 or F2 samples to maybe use instead. Unfortunately, this entire sample set feels like it lacks sustain to me, but I felt it was the best overall sampled piano we could use that was compatible with our SoundFont's MIT license.

I have created an alternate version of the piano that uses a previously unused F2 sample rather than the E2 sample for the notes D#2 through F2. You can test it out by downloading this file, then extracting and importing it into MuseScore. If your archiver of choice cannot extract a 7-zip file, you can download 7-zip. The first preset in the file is the original version of the piano, "Grand Piano", and the second preset has the swapped sample, "Grand Piano-alt" for comparison.

I'm not sure this swap is an improvement (at least to my ears), as it seems to transition a bit more jarringly to the sample below, but if we get a consensus that the F2 sample is preferred, I can make this change in the main SoundFont.

In reply to by s.chriscollins

Again, not an audiophile or anything, but the -alt version sounds better to me. Not perfect, but better. More “even” tones (in themselves, not in the descending scale, but I don’t perceive the slightly different transition as too problematic).

Let’s see whether there are other opinions.

Good morning everyone!

Many thanks to everyone for their efforts, all the great answers and the very interesting information. I did not expect that there would be such a lively participation on the topic and am very pleasantly surprised by the support and the community - thanks again to everyone!
And I have to say that I am very relieved that other people also notice the sound effect. I've already started to doubt myself a little. :)

In fact, I forgot to mention that the strength of the effect is different at different volume levels.

The image of the spectrogram and the explanation of the vibrations of the two strings regarding their impact on volume were both VERY interesting. So far I only knew that the reason for the effect of "choked" notes on the piano was too deep grooves in the felts of the hammers. Over the years, the strings leave deeper and deeper grooves in the felt of the hammers. At some point these are so deep that they almost completely surround the string when they come into contact and thus prevent it from vibrating. The result is strongly shortened and stifled tones, very similar to the present case.

Actually I really like the "Grand Piano sound" from the "MuseScore General SF HQ" and it would be fantastic if you could "correct" the notes in question.
I downloaded the "MuseScore GM piano test 322502" soundfont and tried it out in the "Example of shortened tones bass" as well as in various piano pieces. I think that the notes Db2 to F2 now sound a little longer (which is of course an improvement), but now differ slightly in timbre from the rest of the notes. I am unable to say whether these tones should be permanently replaced in the "GrandPiano sf". Would it theoretically be possible to subsequently edit the samples of the problem tones with an audio program (e.g. audacity) and insert them back into the sound font? Unfortunately, I have absolutely no idea about such things, so this idea may be completely nonsense.

I'm sorry to be so pedantic with these 5 bass notes, but I may be able to clarify this in my (soon) following comment.

On the meaning of soundfonts and the playback of scores in general:

In my experience, the sound quality of a score is very important. You can write a masterpiece, but if it is played back by the music notation program without any sound enhancement, it will get significantly less attention from the audience. Most listeners (e.g. on YouTube) are impatient and if a piece of music does not sound good it is immediately canceled. Very few people perceive the quality of a composition regardless of the sound quality of its reproduction.

As for the notation options, I think musescore is fantastic and, thanks to the excellent work of the creators of this program, I have no need for another program. With Musescore, I can easily notate even the most complicated piano notes (e.g. 4- or even 5-part-fugues). I therefore think that in the future the focus should be placed on the constant optimization of the playback quality (e.g. through sound fonts that sound as good as possible and are as "error-free" as possible). I know this is very easy to write and claim ...

Unfortunately, as a composer you usually have neither the means nor the time to practice your own compositions and record them as professionally as possible. Therefore, one is dependent on the sound result of the notation software and, as described above, this contributes significantly to the success of the composition. But I am convinced that you can also achieve a professional sounding interpretation of a piece of music (at least piano music) by using the playback of a notation program.

I think it is just as important to have the simplest and most time-saving input functions possible for basic interpretation parameters such as ritardandi/accelerandi or the implementation of the accent rules of a time signature (this absolute basic rule of music does not exist in musescore at all so far).

In this context, I had posted a feature request regarding a very basic but often neglected element of musical interpretation: the tonal differentiation of the different beats within a time signature (https://musescore.org/en/node/315610). Unfortunately it has been a little bit ignored so far. In my experience, this parameter makes a decisive contribution to the sound quality of an interpretation.

To illustrate what I mean, I have attached a file with a short example in two versions: the famous second Prelude in c-minor of J.S.Bachs WTC I. The first notation without any sound processing or interpretation, the other with (unfortunately still extremely laborious and time-consuming) sound processing and interpretation. The problem is that entering all the sound parameters (especially the velocity for each individual note) takes much longer than entering the notation of the musical text.

I would be very grateful if you could listen to these two sound samples and then give your opinion.

It is my hope that the Grand Piano-sf can be "corrected" in the bass section and that accelerandi / ritardandi and the accentuation rules can be set in future Musescore versions.

With the best of thanks to everyone and greetings,


In reply to by Polyphonicus

Please keep the scope of one issue in one report. You had already posted the “interpretation / more human-like playback” issue in a different report, so let’s keep this for an actual problem in the soundfont.

As for the soundfont improvements… it’s hard to get instruments recorded in a quality good enough to make soundfonts from. The “Splendid Grand piano” was released in 1999 to use with a hardware synthesiser and has been adapted for soundfont use, but nobody can easily re-record any keys. Direct audio manipulation takes it further away from a real recording and into “plastic sound” (im Sinne von künstlich) territorium, which other people might dislike. I guess we’ll have to live with it until there’s a chance to record a good-quality piano again (on my personal wishlist, vocals and possibly harpsichord are more on top though) or can get a hold of one under a suitable licence (i.e. CC-Zero or MIT/BSD/ISC/…).

For now, do you think Chris should switch to the -alt samples?

The spectral analysis thing is also great. I admit I barely understand anything about it. If any direct sample manipulation happens, does that thing help? How did you (Ludwig) produce it?

MuseScore by default applies some effects as Zita1. For doing analysēs like that, it’d be better to temporarily turn those effects off (View → Synthesiser, 「Master Effects」 tab, change Effect A to NoEffect).

In reply to by mirabilos

"Please keep the scope of one issue in one report. You had already posted the “interpretation / more human-like playback” issue in a different report, so let’s keep this for an actual problem in the soundfont."

  • Ok, sorry, of course I will.

"Direct audio manipulation takes it further away from a real recording and into “plastic sound” (im Sinne von künstlich) territorium,..."

  • I see...plastic sound is not good.

"..., which other people might dislike."

  • I am one of them. :)

"For now, do you think Chris should switch to the -alt samples?"

  • I think that this effort is not necessary. As you have already written, we will probably have to live with the sound font as it is.

In reply to by Polyphonicus

The last paragraph makes no sense… we have to live with the soundfont as is, but Chris can still make small changes; selecting which individual samples of the recorded piano to use is one such change that’s easily done (and he has, in fact, already done all the work for the test soundfont from the link above).

What we cannot easily do is make bigger changes, except to swap complete instruments out with different recordings, or add new recordings as new presets.

In reply to by mirabilos

Another thing I have been thinking about: the Splendid Grand samples are not phase aligned, meaning that when played back on certain devices--especially those that produce mono sound (or stereo sound without much separation)--some frequencies between the left and right channels may cancel eachother out, leading to a perception of greater sound decay or other filtered effects. This could be why the sample in question sounds better when played back on some devices vs. others.

When I created the MuseScore version of this piano, I attempted to manually adjust the phase alignment, but I have since learned that my methodology was only helping the alignment at the very beginning of the sample. I have better ways of doing this now, and I plan to do a re-alignment of these samples in the coming weeks. Perhaps we should wait until the re-alignment is complete before making a decision on this?

In reply to by mirabilos

I don't think direct sample manipulation in an audio editing software would work because manipulation a complex sound like the one from a piano is very tricky and difficult and would more probably lead to an artificial sound (im Sinne von künstlich). Except for doing things that are well described by relatively simple mathematical models, such as adding reverb, equalizing etc.

One really nice thing you can actually do with equalization, for example, is simulating mutes and dampers. I created a couple of virtual instruments for brass with mutes just by using EQ filters and the results sometimes are surprisingly good (as long as the original samples are good as well). I will share the ones I have done at some point, when I get them well adjusted. :)

The spectral analysis I did with Audacity. It natively has an option to show the spectrum instead of the waveform when you go to the dropdown menu from a track.

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