Start and end notes under `Ottava` spanner exported incorrectly due to `toWrittenPitch()`

[NeroBlackstone]

music21 version

7.3.3

Problem summary

Pitch of some notes incorrect after .musicxml file parsed by music21.

Steps to reproduce

from music21 import *
s=converter.parse('1.musicxml')
s.show()

Expected vs. actual behavior

Original file:

After parse:

So C5 is incorrectly parsed as C4.

More information

musicxml file:
1.zip

OS: linux-lts 5.15.64-1

[mscuthbert]

Hi Nero -- what does the MusicXML file output from MuseScore look like in m1 (beginning) and m8 (end)? There are four possible problems:

1. MuseScore is not outputting MusicXML the way you intend the piece to be interpreted (either user error or MuseScore error; not music21).
2. Music21 is not reading the MusicXML properly (music21 problem -- we would fix); the Stream is incorrect.
3. Music21 is not outputting the MusicXML properly (music21 problem -- we would fix)
4. MuseScore is not reading properly formatted MusicXML properly (MuseScore error; not music21)

Can you isolate what is happening w/ the MusicXML specs and determine which of these 4 errors is happening here? Otherwise, if you need someone to do this for you you may need to look into paid consulting support.

[NeroBlackstone]

Thank you, mscuthbert.

I have read .musicxml file before and after music21 parser.
And I think it's possible music21 bug:

For your question 1, I think muse score generate correct .musicxml file.
For example in measure 1: (Original pitch in .mscz is B-5 F5 E-6 F6)
After musescore .musicxml export:

<measure number="1" width="207.21">
...
<pitch>
          <step>B</step>
          <alter>-1</alter>
          <octave>5</octave>
</pitch>
....
<pitch>
          <step>F</step>
          <octave>5</octave>
</pitch>
...
<pitch>
          <step>E</step>
          <alter>-1</alter>
          <octave>6</octave>
</pitch>
...
<pitch>
          <step>F</step>
          <octave>6</octave>
</pitch>
</measure>

It's good, right? So it's not muscscore bug.

For question 2:
Execute:

from music21 import *

s = converter.parse('1.musicxml')
piano_rh = s.parts[0]
print(piano_rh.measure(1).pitches)
s.show()

s.write(fp=r'out.musicxml')

Console output is:

[<music21.pitch.Pitch B-2>, 
<music21.pitch.Pitch D3>,
 <music21.pitch.Pitch F3>, 
<music21.pitch.Pitch B-5>,
 <music21.pitch.Pitch F5>, 
<music21.pitch.Pitch E-6>, 
<music21.pitch.Pitch F6>]

I don't know the first three note come from where.

For question 3:
out.musicxml is:

<measure number="1" width="207.21">
...
<pitch>
          <step>B</step>
          <alter>-1</alter>
          <octave>4</octave> <!-- WHAT'S WRONG ,should be 5-->
</pitch>
...
<pitch>
          <step>F</step>
          <octave>5</octave>
</pitch>
...
<pitch>
          <step>E</step>
          <alter>-1</alter>
          <octave>6</octave>
</pitch>
...
<pitch>
          <step>F</step>
          <octave>6</octave>
</pitch>
...
</measure>

So I think it's music21 bug.

My python version is 3.10

[jacobtylerwalls]

toWrittenPitch(), usually needed for musicxml display, calls Ottava.undoTransposition(), which isn't--since musicXML encodes pitches under ottavas at sounding pitch (or in music21's terms, as "non-transposing spanners").

So toWrittenPitch should probably take an argument undoOttavaTransposition=True so that the musicxml exporter can call it with the value False.

This is only affecting the first and last notes of the spanner because undoTransposition() ignores the intermediary elements of the spanner, see #249.

[NeroBlackstone]

Any other alternatives to music21 to recommend?
I'm sorry for post something unrelated to this bug, but I have an experiment to do.🥲
Thank you very much!

[jacobtylerwalls]

As a workaround, you can set the attribute atSoundingPitch to False on each Score and Part object (or PartStaff, in your case, a subclass of Part):

from music21 import *
myScore = converter.parse('../Downloads/1.musicxml')
myScore.atSoundingPitch = False
for p in myScore.parts:
    p.atSoundingPitch = False

myScore.show()

This will work here since you don't have transposing instruments such as clarinets or trumpets.

[mscuthbert]

Definitely the bug in not processing intermediate notes in a spanner is something that I want to fix. It should not be too hard to do if someone wants to take it on. The Q is what types of elements are included in the intermediary sections? Presumably not rests or clefs, etc. but definitely notes and chords but not unpitched and we can't determine what is likely to be in there based on first and last elements alone.

[mscuthbert]

Okay -- I think I got an answer to "What to include" -- (1) we add to spanner.Spanner (the base object) a set of what objects to include in autoAdd (.autoAddClasses). If defined, we add only those classes during autoadd. I think that Ottava and Slur can be defined to have {GeneralNote} as their set (since they will transpose .pitches and Rests have empty .pitches, they're okay to include. StaffGroups will have Part set in their autoAddClasses.

(2) If undefined and there are no elements then it's treated as Music21Object. If undefined and there is one element, use that class. If undefined and there are two elements, then eventually use the intersection of their MRO.

It still doesn't solve the problem of knowing when to end IF the two elements are in different hierarchies. It probably needs to do something like this https://leetcode.com/problems/lowest-common-ancestor-of-a-binary-tree/ (They're not binary trees, but given that there are only two objects, it works out to be the same problem.

[jacobtylerwalls]

After responding to @gregchapman-dev at gregchapman-dev/converter21#8 (comment), I now think we need ottavasToSounding=True to handle the option to convert either transposing or non-transposing spanners to Sounding pitch as musicxml requires, despite everything else being converted to written pitch. Musicxml documents expect sounding pitch w/r/t transposition and written pitch w/r/t ottavas.

[mscuthbert]

yes. agreed.

[gregchapman-dev]

I will work up a PR for this. (Please assign to me.)

It sounds like there are a few bits to this:

1. Add ottavasToSounding=True option to stream.toWrittenPitch.
2. Call stream.toWrittenPitch with that parameter in m21ToXml.py (and maybe in other writers, including my Humdrum writer).
3. Figure out a way to allow processing every element that is intended by the spanner, even if the spanner only includes start and end element. (I have some questions about the intended approach here... should we actually add the intermediate notes to the spanner at some point? Or have an iterator that iterates over all the elements that should be in the spanner?)
4. Figure out a way to successfully process a spanner that starts and ends in different sub-hierarchies of the score. This is perhaps separable into another PR, but as long as I'm in here, I think this PR would benefit from it.
5. Maybe factor out a separate routine that would help apps like musicdiff compare a transposing Ottava's notes with a non-transposing Ottava's notes (perhaps just a routine that returns all the sounding-pitch notes from an Ottava).

[gregchapman-dev]

Oh, and I am going to start this PR from my SpannerAnchor (PR #1479) branch, since MEI <octave> elements often end with a timestamp instead of a note/chord. I might even find some SpannerAnchor bugs.