Using XML for Musical Representation
Michael Good
Recordare
Guest Lecture for Music Representation class
(Music 254)
Stanford University
May 17, 2000
Copyright © 2000 Recordare. All rights reserved.
Outline
- Why another music representation language?
- Why use XML?
- XML versions of MIDI and NIFF
- MusicXML: An updating of MuseData and Humdrum using Internet standards
My Background
- BS and MS in computer science, MIT
- Developed Scot language for Music 11 (now Csound) for BS thesis
- After MS, worked in human-computer interaction for 18 years
- Started Recordare in January
- Active choral singer, former trumpeter
Why Another Music Representation Language?
- 20+ formats described in Beyond MIDI
- Each commercial music program has its own proprietary format
- The only commonly used interchange format is MIDI
- MIDI is good for performance but not sufficient for notation or analysis
HTML and MIDI as Catalysts for Growth
- HTML and MIDI both hit the "sweet spot" of simplicity and power
- Powerful enough for real-world information and music
- Simple enough for lots of people to learn
- The result: explosive growth in the Web and the electronic music industry
The Limits of HTML and MIDI
- People want to do more than the formats were designed for
- Neither format has extensibility designed in
- HTML is too unstructured for automated electronic commerce
- MIDI does not represent concepts needed for notation and analysis
The New Standard: XML
- XML: Extensible Markup Language
- Like Humdrum, XML provides a syntax for other markup languages
- Like HTML, XML balances human and computer readability
- Very wide support within industry, especially for electronic commerce
Why Are People So Excited About XML?
- Finally, a simple, standard, text-based format for complex structured data
- If you can represent it hierarchically, you can represent it in XML
- You can leverage commodity software and existing Web infrastructure
- Separates semantics, structure, and presentation
More Good Things About XML
- XSL (Extensible Style Language)
- Standard data transformation and formatting language
- In MusicXML, XSLT allows switches between part-based and time-based hierarchies
- XML documents are in Unicode
- "Terseness in XML markup is of minimal importance" - W3C
Designing an XML Representation for Music
- XML by itself is a meta-markup language
- Domain-specific languages are defined with a Document Type Definition (DTD)
- Previous attempts at musical DTDs have generally been too simple for real use
- With a standard DTD in place, musical scores can be exchanged on the Web and between different musical software programs
MusicXML Representation Goals
- Represent common western musical notation from 17th century onwards
- Represent popular and classical music
- Extensible for future support of early music and non-standard modern scores
MusicXML Application Goals
- Support notation, performance, and analysis applications
- Sufficient, not optimal, for these applications
- Support interchange with any musical program with a published data format
- 100% coverage of MuseData
Why MusicXML Instead of NIFF or SMDL?
- NIFF and SMDL are both oriented towards machine readability, not human readability
- NIFF is oriented towards notation; difficult to use for other applications
- SMDL trades off ease of use with common notation for full generality
- MusicXML leverages Internet standards
XML Versions of MIDI and NIFF
- Provide a text representation of these binary formats
- Used by MusicXML as an intermediate representation for file import and export
- Why build your own internal data structure when you can just use XML?
- Examples: Recordare logo (Verdi):
MIDI XML Example
<?xml version="1.0" standalone="yes" ?>
<MIDI>
<Format>1</Format>
<Tracks>2</Tracks>
<TicksPerBeat>24</TicksPerBeat>
<TimestampType>Delta</TimestampType>
<Track number="0">
<SetTempo>
<Delta>0</Delta>
<Value>750000</Value> <!-- MM = 80 -->
</SetTempo>
<KeySignature>
<Delta>0</Delta>
<Fifths>-1</Fifths> <!-- One flat -->
<Mode>0</Mode>
</KeySignature>
<EndOfTrack>
<Delta>192</Delta>
</EndOfTrack>
</Track>
<Track number="1">
<TrackName>
<Delta>0</Delta>
<Value>Mezzo-soprano</Value>
</TrackName>
<ProgramChange>
<Delta>0</Delta>
<Channel>1</Channel>
<Number>54</Number> <!-- GM Voices -->
</ProgramChange>
<NoteOn> <!-- Initial A -->
<Delta>0</Delta>
<Channel>1</Channel>
<Note>69</Note>
<Velocity>90</Velocity>
</NoteOn>
<NoteOff>
<Delta>60</Delta> <!-- Half tied to eighth -->
<Channel>1</Channel>
<Note>69</Note>
<Velocity>90</Velocity>
</NoteOff>
<NoteOn>
<Delta>0</Delta>
<Channel>1</Channel>
<Note>65</Note>
<Velocity>90</Velocity>
</NoteOn>
<!-- Rest of notes omitted... -->
<EndOfTrack>
<Delta>24</Delta> <!-- quarter rest -->
</EndOfTrack>
</Track>
</MIDI>
NIFF XML Example
<?xml version="1.0" standalone="yes" ?>
<NIFF>
<SetupSection>
<!-- Skip these next 3 for here... -->
<ChunkLengthTable />
<NiffInformation />
<FontDescriptions list="yes" />
<StringTable>
<String Offset="0">Mezzo-soprano</String>
<String Offset="14">M.</String>
</StringTable>
<Parts list="yes">
<PartDescription>
<PartID>0</PartID>
<PartName>0</PartName>
<PartAbbreviation>14</PartAbbreviation>
<Staves>1</Staves>
<MidiChannel>0</MidiChannel>
<MidiCable>-1</MidiCable>
<Transpose>0</Transpose>
</PartDescription>
</Parts>
</SetupSection>
<DataSection>
<Page list="yes">
<PageHeader /> <!-- Skip headers here -->
<System list="yes">
<SystemHeader />
<Staff list="yes">
<StaffHeader />
<TimeSlice type="2">
<StartTime Numerator="0"
Denominator="1" />
</TimeSlice>
<Clef Shape="1" StaffStep="2"
OctaveNumber="0" />
<KeySignature StandardCode="8" />
<Stem>
<LogicalPlacement Horizontal="0"
Vertical="1" Proximity="0" />
</Stem>
<Slur>
<NumberOfNodes>2</NumberOfNodes>
<ID>0</ID>
</Slur>
<Notehead Shape="3" StaffStep="3">
<Duration Numerator="1" Denominator="2" />
<PartID>0</PartID>
</Notehead>
<Tie>
<NumberOfNodes>2</NumberOfNodes>
<ID>0</ID>
</Tie>
<!-- Rest of notes omitted... -->
</Staff>
</System>
</Page>
</DataSection>
</NIFF>
MusicXML Example (May 2000 version)
<?xml version="1.0" standalone="no" ?>
<!DOCTYPE score-partwise SYSTEM "musicxml-p.dtd">
<score-partwise>
<movement id="M1">
<part-list>
<score-part id="M1-P1">
<part-name>Mezzo-soprano</part-name>
<part-abbreviation>M.</part-abbreviation>
</score-part>
</part-list>
<part id="M1-P1">
<measure number="1">
<attributes>
<key>
<fifths>-1</fifths>
</key>
<divisions>2</divisions>
<staves>1</staves>
<clef>
<sign>G</sign>
<line>2</line>
</clef>
</attributes>
<note>
<pitch>
<step>A</step>
<octave>4</octave>
</pitch>
<duration>4</duration>
<tie />
<type>half</type>
<stem>up</stem>
<notations>
<slur type="start" number="1"/>
<tied/>
</notations>
</note>
<!-- Notes omitted until last B-flat -->
<note>
<pitch>
<step>B</step>
<alter>-1</alter>
<octave>4</octave>
</pitch>
<duration>1</duration>
<type>eighth</type>
<stem>up</stem>
<beam number="1">end</beam>
</note>
</measure>
<measure number="2">
<!-- Notes omitted until last rest -->
<note>
<rest />
<duration>2</duration>
</note>
</measure>
</part>
</movement>
</score-partwise>
MusicXML Converter Status as of May 17, 2000
- Proof of concept that MusicXML supports performance, notation, and analysis formats
- Reads MuseData and NIFF files
- Writes Standard MIDI Files
- Future directions: Support Humdrum and commercial formats
How Much Bigger Are The MusicXML Files?
- XML compresses very well
- Zip compresses MusicXML about 30x
- Comparison with MuseData for first movement of Haydn's Symphony #104:
- Uncompressed (11 score files)
- MuseData 319K, MusicXML 2.19M (7x)
- Compressed (.zip)
- MuseData 35K, MusicXML 69K (2x)
XML Document Type Definitions (DTDs)
- XML is a meta-markup language
- A DTD defines the specific markup language or tag set
- Documents without DTDs can be well-formed XML, but not validated XML
- Standardization thus becomes a DTD design issue
MusicXML DTD Sample: note
<!-- Internal entities to simplify note definitions -->
<!ENTITY % full-note "(chord?, (pitch | rest))">
<!ENTITY % voice-track "(footnote?, level?, track?)">
<!-- Definition of the note element -->
<!ELEMENT note ((((cue | grace), %full-note;) |
(%full-note;, duration, tie?)),
%voice-track;, type?, dot*,
accidental?, time-modification?,
stem?, staff?, beam*, notations*,
text*)>
MusicXML DTD Sample: pitch and slur
<!ELEMENT pitch (step, alter?, octave)> <!ELEMENT step (#PCDATA)> <!ELEMENT alter (#PCDATA)> <!ELEMENT octave (#PCDATA)> <!-- Slur is an empty element with attributes --> <!ENTITY % beam-level "(1 | 2 | 3 | 4 | 5 | 6)"> <!ENTITY % start-stop "(start | stop)"> <!ELEMENT slur EMPTY> <!ATTLIST slur type %start-stop; #REQUIRED> <!ATTLIST slur number %beam-level; "1">
Working With XML
- Two main XML programming models
- Accessible from many programming languages (Java, C++, Visual Basic)
- Tools available from multiple vendors
- My examples use DOM and Visual Basic
Analysis Examples
- Two examples adapted from Humdrum problem list:
- Do Bach's pieces really have 90% of the notes in one of two durations?
- Is pitch correlated with duration?
- Show duration distribution on bar chart
- Show correlation analysis on scatterplot
Durations in Bach BWV 6, 2nd Movement
Correlations in Mozart Quartet #7, 1st Movement
Distribution Analysis Code
Set oNodes = oRoot.selectNodes("//note | //divisions")
Do
Set oNote = oNodes.NextNode
If oNote Is Nothing Then Exit Do
If oNote.tagName = "divisions" Then
nMultiplier = nDivisions \ CLng(oNote.Text)
Else
Set oTmp = oNote.selectSingleNode("duration")
If Not oTmp Is Nothing Then ' Exclude grace and cue notes
nDuration = CLng(oTmp.Text) * nMultiplier
nCounts(nDuration) = nCounts(nDuration) + 1
End If
End If
Loop
Correlation Analysis Code
' Code structure is same as for distribution analysis.
' The Else clause for handling notes changes:
Else
Set oDuration = oNote.SelectSingleNode("duration")
Set oPitch = oNote.SelectSingleNode("pitch")
If (Not oDuration Is Nothing) And _
(Not oPitch Is Nothing) Then ' No rests either
nDuration = CLng(oDuration.Text) * nMultiplier
nPitch = MidiNote(oPitch) ' Separate function
oWorksheet.Cells(i, 1) = nDuration
oWorksheet.Cells(i, 2) = nPitch
End If
End If
Future Directions for MusicXML
- Test, debug, and iterate with more music, formats, and applications
- More general release once internally debugged
Summary
- XML is the Internet standard for representing complex structured data
- Representing music in XML is a natural way to publish musical scores, not just recordings, on the Internet
- Standard music DTD must be well-designed from musical, human, and computer perspectives
Acknowledgments
- Eleanor Selfridge-Field for inviting me to make this presentation to her class
- For their encouragement and prior work in musical score representation:
- Eleanor Selfridge-Field and Walter Hewlett, CCARH, Stanford University
- Barry Vercoe, MIT Media Lab
- David Huron, Ohio State University School of Music
- Gerd Castan for the idea of creating an XML version of NIFF
Home - Music - Software - MusicXML - Events - Search - Store - About Us
Copyright © 2000 Recordare. All rights reserved.