Electronic Music Editing
The preparation of encoded musical materials for processing is an essential part of computer-assisted musicology. Without electronic data to process, none of the manipulations described in this book would be possible. The encoding and editing of musical data has been an activity that has occupied various scholars since the earliest forays into computer-assisted musicology (e.g., Bronson, 1959).
In this chapter we identify and describe some of the tasks and issues involved in preparing electronic musical documents in the Humdrum format. In general, the following discussion pertains to the production of electronic documents representing musical score information. However, the basic procedures are applicable to any kind of data — from sound recordings to historical choreographies of ballets.
The Process of Electronic Editing
Electronic editing entails translating one representation into another representation. In many cases, an electronic edition is prepared from existing source documents — such as printed scores or manuscripts. In other cases, an electronic edition may be prepared from another electronic form — such as MIDI. When translating between two formats, it is important to be familiar with both representations. Descriptions of a number of popular electronic music formats may be found in Eleanor Selfridge-Field’s encyclopedic Handbook of Musical Codes (1997).
The principal processes involved in electronic editing include identifying possible sources and alternative versions, selecting and encoding the material, proofing the material (using both automated and manual methods), adding reference and editorial information, writing appropriate research notes, generating possible analytic information, and resolving issues related to copyright, distribution, and data integrity.
Establishing the Goal
Encoding and editing music is a time-consuming and labor-intensive process. Before starting to work on a electronic edition, it is appropriate to ask: what is the goal? What purpose will be served by the edition? Is the objective to provide a performance copy? A historically accurate document? A document suitable for analysis? If so, what sort of analysis?
In general, it is wise to avoid trying to cater to all needs at once. There is nothing wrong with creating a narrowly-defined document that serves a modest goal. For example, if our intention is to pursue a study of rhythm, there is little need to encode data that is extraneous to this goal. Accordingly we might omit stem-direction, beaming, textual underlay, and even pitch information.
Note that it is always possible to incorporate additional information at a later date. Additional information can be inserted into a Humdrum encoding by using the assemble and cleave commands. In general, it is important that the encoding of electronic documents not consume all of a researcher’s efforts and resources. The most common problem that beset early projects in computational musicology was that researchers rarely got past the stage of inputting data. Early researchers typically ran out of time, money or enthusiasm before they could turn to using the materials they had input.
Documenting Encoded Data
More important than generating a complete and accurate database is generating complete and accurate documentation so that future users of the information understand the limitations of the materials. Whether or not the electronic edition is “complete,” the materials are almost useless without proper documentation about what information is present, what information has been omitted, what sources were used, and what interpretations have been made. As long as the electronic materials are well documented, users can make effective use of whatever you create.
An ethnomusicologist may have inaccurately transcribed pitches, but even approximate pitch information can be used for a study of pitch contour. Of course, it is better to have clean data than messay data, but even messy data can be highly useful if the nature of the mess is well understood by researchers. In fact, most historical information is a mess.
Concretely, Humdrum provides several reference records that allow electronic editors to identify what information has been encoded and what information has been omitted or interpreted.
In light of the editorial goal, you can procede to select source materials for encoding. In many cases, this will involve selecting printed musical scores. Begin by spending time with the materials. What special problems or challenges are raised? What sort of Humdrum representation would be suitable for the proposed edition? Are there special notational symbols that suggest a new representation should be designed? Can special representational problems be accommodated within an existing Humdrum representation? How would users be made aware of extensions or modifications to the representation?
In the case of musical scores, there is rarely a single source or Urtext. In most cases, multiple sources are available. How do you want to approach the problem of variant sources? One approach is to select a single source and remain true to it. A second approach is to identify the principal variant sources and encode each independently. A third approach is to encode several variants within a single document and provide Humdrum “versions” that allow users to select particular “readings” from a single file. A fourth approach is to create your own “critical” edition — based on a close re-examination of the original materials.
An important consideration will be the copyright status of the materials you examine. Some materials are old enough that their copyrights have expired. Encoding older materials raises two difficulties. First, older sources are typically not the best quality editions available. Second, access to printed versions of older sources may be difficult for other scholars using your data. Some publishers (such as Dover Publications) specialize in reprinting musical scores whose copyrights have lapsed. Although these sources may not be the best available, they are often convenient because other scholars can easily purchase the modern Dover reprints that correspond to your electronic edition.
The best materials are usually under copyright. If you encode this material without permission, you will never be able to distribute the encodings — even on a non-profit basis. Most modern publishers profess an interest in electronic publishing, but (at the time of this writing) few publishers have any pertinent expertise, and fewer yet are willing to encourage or allow the preparation of electronic documents from their publication catalogue by third parties. It is likely to take decades before the financial and legal issues are satisfactorily addressed.
In some cases, you will be translating from one electronic format to another — such as from MIDI to kern. Once again, be conscious of any copyright issues that are raised. Few experiences are more discouraging than discovering that your work cannot be distributed because you failed to consider seriously the copyright issues involved.
Selecting a Sample from Some Repertory
Quite often, there are insufficient resources to encode an entire corpus. Many repertories are simply too large to consider creating an exhaustive electronic edition. Nevertheless, you might choose to encode a subset or selection of works from the given corpus.
A common expectation is that studying a sample subset of works will inform us about the corpus as a whole. That is, it is commonly expected that studying a number of works by, say, Offenbach, the scholar may be able to identify general characteristics of Offenbach’s writing. In order for this assumption to be valid, statisticians tell us that the sample of works must be a representative sample — free of all sorts of possible selection biases.
From a research perspective, the best sample is a random sample. Suppose, for example, that there are 2,000 items in a particular repertory, but you are able to encode just 100 items. You might be tempted to select the first 100 items, or to select a subset of items that share some common feature that makes the collection seem “coherent.”
However, picking and choosing what to encode will prevent researchers from being able to draw general conclusions about the repertory as a whole. For example, selecting the first 100 items will typically bias the sample to early works in a composer’s career. Selecting 100 random items provides a very good statistical sample of a population of 2,000 items.
Unfortunately, since random sampling has little precedence in traditional humanities research, many scholars are resistent to the idea. The value of random sampling has been established beyond a doubt by statisticians. This is not the place to rehearse the detailed arguments. Simply take my word for it: if you can’t encode a complete corpus, the very best solution is to select a random sample.
In making such a random sample, it is essential to resist the temptation to select a “random” sample “by eye.” Establish a truly random procedure (such as flipping coins or using a random number table) and methodically follow the procedure.
Incidentally, it is common to run out of resources before completely encoding the selected materials. As a result, you may end up encoding only half or two-thirds of the projected materials. If you began encoding the materials in (say) chronological order, then the resulting database will be biased toward the early works of the repertory. In order to avoid introducing an unwanted bias, it is also prudent to encode the selected materials in a random order.
Once you have established your materials and have decided on the type of encoding, you can go ahead and begin encoding the documents in random order. Use whatever resources are available to you. These might include scanning software, MIDI performance capture, or the Humdrum encode command. Begin by encoding a sample section or sections. Spend some time determining ways to increase your productivity.
As you encode the selected materials, editorial problems or questions will inevitably arise. As you gain experience, you may realize that earlier encoding practices were not the best. You may want to return to these problems and encode them in a different manner. Be sure to keep notes — either pencil marks on a page, or local comments in a file — so that you can easily revisit these problem sites later. Again, it is valuable to encode works in random order in order to avoid possible confounds arising from editorial experience. That is, you don’t want a scholar’s conclusions about differences between early works and late works to be merely an artifact of the electronic editor’s increasing experience.
Typically, it is more efficient to encode individual parts and then assemble all parts into a single full score.
In the case of the kern representation, all parts are represented at concert pitch. It is typically easier to encode the parts as written and then transpose the result using the Humdrum trans command. For example, material for B-flat trumpet or B-flat clarinet can be transposed using the following command:
trans -d -1 -c -2
In the case of clarinet in A, a suitable transposition would be:
trans -d -2 -c -3
The trans command adds a transposition
interpretation to the output in order to identify that the material
has been shifted. In the kern representation,
transposed instruments must be explicitly identified using a special
“transposing-instrument interpretation” (see Humdrum Reference
Manual — Section 3 for details). A suitable interpretation
can be created by adding the upper-case letter
I prior to the
in the appropriate tandem interpretation. In the case of a horn in
F for example, the transpostion interpretation would be modified
Humdrum provides standardized instrumentation indicators. Three
different types of indication are appropriate: (1) the instrument
name as indicated in the source, (2) standardized instrument name,
and (3) instrument class. Standardized instrument names can be found
in Appendix II. For example, the standard
indicator for “harpsichord” is
Standardized instrument class designators include
keyboard instruments and
*ICidio for percussion instruments, etc.,
and instrument grouping designators — such as
ripieno instruments and
*IGacmp for accompaniment instruments.
These instrument class designators can also be found in Appendix
In addition, the original instrument name (as found in the score) should also be encoded as a Humdrum local comment.
Humdrum tools prefer to have explicit information indicating the beginning of the first measure. If a file does not begin with an anacrusis (“pickup”) then it is appropriate to encode an “implicit” first barline. For a hypothetical file containing five spines, we would need to insert the following line just before the first note(s) in the work:
=1 =1 =1 =1 =1
Recall that the common system for representing barlines makes a distinction between the logical function of a barline and it’s visual or orthographic appearance. For example, the common system for barlines distinguishes between double barlines whose function is to indicate the end of a work or movement, and double barlines that simply delineate sections within the course of a work or movement. It is possible for a barline at the end of the work to be “functionally” a double barline, yet appear visually as a single barline.
Functional double barlines are encoded with a double equals sign (==) whether or not they are visually rendered as double barlines.
Functional single barlines are encoded with a single equals sign (=) whether or not they are visually rendered as single barlines.
The specific visual appearance may be encoded following the equals
sign(s). The vertical line
| represents a ‘thin’ line and the
! represents a ‘thick’ line. A typical final
double bar would be encoded as
== with the style being implicitly
Most mid-movement double bars are encoded with two thin lines and so would be encoded:
A common encoding error is to render mid-movement double barlines as functional rather than orthographic double-bars.
The kern representation makes a distinction between whole-tone and semitone trills and mordents. Typically, each ornament must be examined manually and the correct code selected.
In some cases, the size of the trill or mordent will be ambiguous
and so some sort of editorial decision will be necessary. One
possibility is to add the kern
x signifier immediately following
t. This indicates that the trill size is an “editorial
The kern representation treats appoggiaturas in a special way. In general, kern is oriented to representing things in a manner closer to how they sound. Consequently, appoggiaturas are encoded as they would be logically performed. For example, a quarter-note preceded by an appoggiatura (small note) would be performed as two eighth-notes. Similarly, a dotted quarter-note preceded by an appoggiatura would be performed as a quarter-note followed by an eighth-note.
All appoggiaturas must be re-encoded in a way that reflects their
likely performance. At the same time, the two notes forming the
appoggiatura must be marked in the kern representation: the initial
note of the appoggiatura is marked by the upper-case letter
and the final (second) note of the appoggiatura is marked by a
It is helpful to break-up large works/movements into smaller sections
that can be labelled. In a binary work, for example, it may be
useful to label the
B sections. In a sonata-allegro
work, it may be useful to label the introduction, exposition,
development, recapitulation, etc. Some works include explicitly
notated labels. These labels may be traditional, e.g. “Coda,” or
they may reflect programatic descriptions, such as the section
entitled Il canto degl’uccelli [The song of the birds] in
Vivaldi’s The Four Seasons.
Where appropriate, suitable section labels should be created and encoded using the Humdrum Section Label designator. Remember that section labels can include the space character:
If you include section labels, you must also include a Humdrum “Expansion List” to indicate how the sections are connected. The Humdrum thru command causes a through-composed version of a file to be generated according to the expansion list. For example, an expansion list for a simple binary work may be encoded as:
Remember that expansion lists ought to be encoded prior to the first section label.
Whenever a work/movement includes repeats or Da Capos, section labels and expansion lists must be encoded. In some cases, there is more than one way of interpreting how to realize the repeats. The most “conventional” realization should be encoded with the unnamed expansion list. This will specify the default expansion using the Humdrum thru command. Suppose for example, that you are encoding a typical minuet and trio. The conventional performance practice involves repeating all sections of both the minuet and trio, but then avoiding the repeats in the minuet following the Da Capo. A suitable expansion list might be:
An alternative expansion list might be encoded as follows (notice the expansion-list-label ossia):
Editorialisms in the **kern Representation
Humdrum provides several ways of encoding editorialisms. These include editorial footnotes, local comments, global comments, interpretation data, sic and ossia designations, version labels, sectional labels, and expansion lists.
The kern representation provides several
special-purpose signifiers to help make explicit various classes
of editorial amendments, interpretations, or commentaries. Five
types of editorial signifiers are available: (1) sic (information
is encoded literally, but is questionable) signified by the
character; (2) invisible symbol (Unprinted note, rest or barline,
but logically implied) signified by the
y character; (3) editorial
interpretation, (a “modest” editorial act of interpretation —
such as the interpretation of accidentals in musica ficta) signified
x character; (4) editorial intervention (a “significant”
editorial intervention) signified by the
X character; (5) footnote
(accompanying local or global comment provides a text commentary
pertaining to specified data token) signified by
One of the most onerous impositions of the kern representation is the requirement that the music be interpreted into a coherent spine organization. Why not avoid interpreting the voicings?
The answer to this question is that editorial interventions are often essential clarifications that make a document useable. Without voicing information, users would be unable to calculate melodic intervals, for example. Without melodic intervals, it may be impossible to search for themes, motives, and other patterns. Editorial interpretations are not simply unwarranted obfuscations. This does not mean that interpretations are “correct” and so it may be necessary to provide several alternative or plausible interpretations of an artifact.
One of the advantages of computers is that it is possible for documents to undergo continuous revision. In research, it is common for documents to be reinterpreted, annotated, or recast in light of newly found documents.
x signifies an “editorial interpretation” —
that the immediately preceding signifier is interpreted. The kern
xx also signifies an editorial interpretation where the
immediately preceding data token is interpreted. The kern
signifies an “editorial intervention” — that the immediately
preceding signifier is an editorial addition. The kern
also signifies an editorial intervention where the immediately
preceding data token is an editorial addition. The kern
designates a invisible symbol — such as an unprinted note or
rest that is logically implied. The kern
Y signifies an
editorial sic marking — that the information is encoded
literally, but is questionable. The kern
? signifies an
editorial footnote where the immediately preceding signifier has
an accompanying editorial footnote (located in a comment record).
?? signifies an editorial footnote where the immediately
preceding data token has an accompanying editorial footnote (located
in a comment record).
Adding Reference Information
Reference information must be added to each file. This information provides “library-type” information about the composer, date of composition, place of composition, copyright notice, etc.
As many reference records should be added as possible since these are immensely useful to Humdrum users. Essential reference records include the following:
||title (in original language)|
||movement number (if appropriate)|
||opus number (if appropriate)|
||date of composition|
||place of composition|
||publisher of electronic edition|
||date & owner of electronic copyright|
||date electronic edition released|
||country of copyright|
||encoder of document|
||electronic edition version|
||file number, e.g. 1 or 4 (1/4)|
||checksum validation number (see below)|
Where appropriate, the following reference records should also be included:
||title (English translation)|
||title of larger (or parent) work|
||country of composition|
||date first published|
||place first published|
||scholarly catalogue name & number|
||form of work|
||genre of work|
||style of period|
In general, place essential reference records at the beginning of a document. These will include the composer, title of the work, etc. Less important reference records should be placed at the end of the file. Minimizing the number of reference records at the beginning of a file makes it more convenient for users looking at the contents of a Humdrum file.
Refer to the Humdrum Reference Manual for further information about the types and format for different reference records.
Once you have encoded your document, you should create a error-checking strategy. The Humdrum humdrum command can be used to identify whether the final encoded output conforms to the Humdrum syntax:
Use the Humdrum proof k command to identify any syntactical errors in any encoded kern data:
proof -k full.krn
One of the best ways to ensure that musical data makes sense is to listen to it. The Humdrum midi and perform commands can be used to listen to your data. The c option for midi causes the Humdrum data to be displayed while the MIDI data is performed. This can further help in locating errors.
midi -c full.krn | perform
The perform command allows you to pause
(press the space bar), to move to a particular measure (type a
measure number followed by enter), to increase (type
<) or decrease
>) the tempo, and to return to the beginning of the
score (type enter). There are many other functions within the perform command; refer to the Humdrum Reference
Manual — section 4 for further details.
Data Integrity Using the VTS Checksum Record
When using electronic documents, it is often useful to modify the document for some purpose. After a while, the user will become confused about the status of a document. Is this the original distribution file? Did I make some modification to this file that I’ve forgotten about? Has someone tampered with this data?
Humdrum provides a means for ensuring that a particular file is what it purports to be. The veritas command provides a formal means for verifying that a given Humdrum file is identical to the original distribution file and has not been modified in some way.
The veritas command works by looking for a VTS reference record in the file. It then calculates a “checksum” for the file (excluding the VTS record itself) and compares this value with the encoded VTS value. If these values differ, a warning is issued that the file has been modified in some way.
Once you are certain that an encoded Humdrum file is completely finished, you should calculate a “checksum” value to be encoded in a Humdrum “VTS” reference record.
In order to calculate the checksum value for a given file, use the following command:
Open the original file and move to the bottom of the document. Then
read in the calculated checksum value. Finally, insert the
reference record designator.
You can check that everything is fine by invoking the veritas command:
The command will complain only if the VTS checksum value does not correspond to the computed checksum for the file. Finally, be sure to include the checksum value in an index or README file for the distribution. This provides a public venue for users to determine whether the VTS record itself has not been modified.
Preparing a Distribution
Finally, you may want to prepare the material you have encoded for
public distribution. Rename the score files and collect them into
a coherent repertory. If your data is encoded in the kern format, be sure to use the
extension. Place all resulting Humdrum files in a single directory.
README file similar to others in Humdrum data distributions.
The file should contain a repertory title, a brief paragraph
describing the historical background for the works, a paragraph
describing the personnel involved in the production, a copyright
and license notice, and a table of contents. Avoid tabs in this
file, and ensure that no line is greater than 80-characters in
It is wise to also add a
LICENSE file that reiterates whatever
licensing agreement is entailed for the distributed data.
Electronic editions of music might be cited in printed or other documents by including the following information. The “author” (e.g. !!!COM:), the “title” — either original title (!!!OTL:) or translated title (!!!XEN:). The editor (!!!EED:), published (!!!YEP:), date of publication and copyright owner (!!!YED:), and electronic version (EEV:). In addition, a full citation ought to include the validation checksum (!!!VTS:). The validation number will allow others to verify that a particular electronic document is precisely the one cited. A sample citation to an electronic document might be:
In this chapter we have reviewed the principal issues involved in preparing electronic music documents in Humdrum.