MusicXML voice numbers should be unique within a multi-staff part (reuse of voice numbers confuses Dorico)

[gregchapman-dev]

music21 version

7.3

Problem summary

Dorico believes the notes are in the same voice, so they end up rendered as a cross-staff two-note chord. Musescore doesn't mind.

Here's an example MusicXML file (produced by music21) that re-uses voice numbers within a multi-staff part.

<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE score-partwise  PUBLIC "-//Recordare//DTD MusicXML 3.1 Partwise//EN" "http://www.musicxml.org/dtds/partwise.dtd">
<score-partwise version="3.1">
  <movement-title></movement-title>
  <identification>
    <creator type="composer"></creator>
    <encoding>
      <encoding-date>2022-06-18</encoding-date>
    </encoding>
  </identification>
  <defaults>
    <scaling>
      <millimeters>7</millimeters>
      <tenths>40</tenths>
    </scaling>
  </defaults>
  <part-list>
    <score-part id="P1">
      <part-name />
    </score-part>
  </part-list>
  <!--=========================== Part 1 ===========================-->
  <part id="P1">
    <!--========================= Measure 1 ==========================-->
    <measure number="1">
      <attributes>
        <divisions>1</divisions>
        <staves>2</staves>
        <clef number="1">
          <sign>G</sign>
          <line>2</line>
        </clef>
        <clef number="2">
          <sign>F</sign>
          <line>4</line>
        </clef>
      </attributes>
      <note>
        <pitch>
          <step>C</step>
          <octave>5</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>1</staff>
      </note>
      <note>
        <pitch>
          <step>D</step>
          <octave>5</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>1</staff>
      </note>
      <note>
        <pitch>
          <step>E</step>
          <octave>5</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>1</staff>
      </note>
      <note>
        <pitch>
          <step>F</step>
          <octave>5</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>1</staff>
      </note>
      <backup>
        <duration>4</duration>
      </backup>
      <note>
        <pitch>
          <step>C</step>
          <octave>3</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>2</staff>
      </note>
      <note>
        <pitch>
          <step>D</step>
          <octave>3</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>2</staff>
      </note>
      <note>
        <pitch>
          <step>E</step>
          <octave>3</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>2</staff>
      </note>
      <note>
        <pitch>
          <step>F</step>
          <octave>3</octave>
        </pitch>
        <duration>1</duration>
        <voice>1</voice>
        <type>quarter</type>
        <staff>2</staff>
      </note>
      <barline location="right">
        <bar-style>regular</bar-style>
      </barline>
    </measure>
  </part>
</score-partwise>

And here's the Dorico rendering of that file:

I'm working on a simple test case that can trigger generation of that MusicXML (the test case I have involves my Humdrum parser).

[gregchapman-dev]

I don't blame Dorico, they're trying to pay correct attention to which notes go in which voice, and I'm happy about that.

[gregchapman-dev]

I would be happy to work on this, once I get an opinion about whether this is a fix you would want. My proposal would be to do something simple in the MusicXML writer, like pre-allocate 4 voice numbers to each staff (start staff 2 with voice 5), and let the chips fall where they may.

[gregchapman-dev]

Looks like the voice numbering happens in moveMeasureContents (in partStaffExporter.py). It looks suspiciously like it's trying to make voice numbers unique, but perhaps it doesn't know enough, and I can plumb more info down to it (and update it's uniqueness algorithm a bit)? Definitely looking for advice.

[gregchapman-dev]

Ah. Looks like the following code is the culprit:

            if elem.tag == 'note':
                voice = elem.find('voice')
                if voice is not None:
                    if otherMeasureLackedVoice and voice.text:
                        # otherMeasure assigned voice 1; Bump voice number here
                        voice.text = str(int(voice.text) + 1)
                    else:
                        pass  # No need to alter existing voice numbers <-- HERE BE DRAGONS
                else:
                    voice = Element('voice')
                    voice.text = str(maxVoices + 1)

[gregchapman-dev]

Simple (well, sort of) test that I have added to partStaffExporter.py, which fails without my fix, and succeeds with my fix:

    def testJoinPartStaffsD2(self):
        '''
        Add measures and voices and check for unique voice numbers across the StaffGroup.
        '''
        from music21 import layout
        from music21 import note
        s = stream.Score()
        ps1 = stream.PartStaff()
        m1 = stream.Measure()
        ps1.insert(0, m1)
        v1 = stream.Voice()
        v2 = stream.Voice()
        m1.insert(0, v1)
        m1.insert(0, v2)
        v1.repeatAppend(note.Note('C4'), 4)
        v2.repeatAppend(note.Note('E4'), 4)
        ps1.makeNotation(inPlace=True)  # makeNotation to freeze notation

        ps2 = stream.PartStaff()
        m2 = stream.Measure()
        ps2.insert(0, m2)
        v3 = stream.Voice()
        v4 = stream.Voice()
        m2.insert(0, v3)
        m2.insert(0, v4)
        v3.repeatAppend(note.Note('C3'), 4)
        v4.repeatAppend(note.Note('G3'), 4)
        ps2.makeNotation(inPlace=True)  # makeNotation to freeze notation

        s.insert(0, ps2)
        s.insert(0, ps1)
        s.insert(0, layout.StaffGroup([ps1, ps2]))
        root = self.getET(s)
        measures = root.findall('.//measure')
        notes = root.findall('.//note')
        # from music21.musicxml.helpers import dump
        # dump(root)
        self.assertEqual(len(measures), 1)
        self.assertEqual(len(notes), 16)

        # check those voice and staff numbers
        for mxNote in notes:
            mxPitch = mxNote.find('pitch')
            if mxPitch.find('step').text == 'C' and mxPitch.find('octave').text == '4':
                self.assertEqual(mxNote.find('voice').text, '1')
                self.assertEqual(mxNote.find('staff').text, '1')
            elif mxPitch.find('step').text == 'E' and mxPitch.find('octave').text == '4':
                self.assertEqual(mxNote.find('voice').text, '2')
                self.assertEqual(mxNote.find('staff').text, '1')
            elif mxPitch.find('step').text == 'C' and mxPitch.find('octave').text == '3':
                self.assertEqual(mxNote.find('voice').text, '3') # FAIL: it is '1' instead of '3'
                self.assertEqual(mxNote.find('staff').text, '2')
            elif mxPitch.find('step').text == 'G' and mxPitch.find('octave').text == '3':
                self.assertEqual(mxNote.find('voice').text, '4') # FAIL: it is '2' instead of '4'
                self.assertEqual(mxNote.find('staff').text, '2')

[gregchapman-dev]

Fixed in PR #1336