Whatever method is used, the aim is assumed to be an encoding which uses an internal music representational language (MRL). In real music acquisition systems various overlaps occur, such as when an organ keyboard is used in conjunction with a graphics tablet [Wittlich 1978a]. This overlap suggests an alternative categorization which involves dividing up the information encoded in a page of printed music into three basic types and associating each with an input device. Thus, pitch may be entered using an organ keyboard, rhythm could be determined by selecting from an on-screen menu of durations using a mouse, and miscellaneous symbols, like slurs, dynamics or text, could be entered using a purpose-built keyboard.
2.2.1 Automatic pattern recognition of printed music
Past and present work on pattern recognition of printed music
-7-
is described in detail in chapters 3, 4 and 5. Chapter 3 also provides a discussion of work in the general field of pattern recognition of binary images.
2.2.2 Operator input
There are two musician-only methods which attempt to capture all possible information represented in a score. One uses a directly-connected musical instrument (usually an organ keyboard) while the other involves analysing a soundtrack recording of a performance of the score.
2.2.2.1 Directly-connected instrument
It is impractical to encode all the information represented in a score by sole use of an organ or electronic piano-type keyboard (hereafter organ keyboard), and it would also appear impossible to convert into 100% accurate notation all the information present in an organ keyboard performance [Knowlton 1971, 1972]. A major problem discussed by Raskin [1980a, 1980b] and Clements [1980a] is rhythm-fitting, since no performance contains, throughout, durations which are fully consistent with the notation. Voice assignment has been found to be an intractable problem [Maxwell 1983, 1984] [Byrd 1984]. For example, if two voices out of three in a three-voiced composition are sounding in unison, it is impossible for the system to know whether two or three voices are actually sounding.
Enharmonic notation is another problem of some magnitude. In
-8-
tonal music, guidelines might be formulated to assist the machine in determining which particular notation is meant, but for atonal music the problem appears insurmountable [Byrd 1984] [Gomberg 1975, 1977].
Thus, provided the user circumvents the limitations of using an organ keyboard as an input device, it can still form part of a practical system.
2.2.2.2 Soundtrack analysis
Early work in this field by Moorer [1975, 1977] limited input to no more than two voices with further restrictions on the harmonic relationship between simultaneously sounding notes. He stated that 'In general, the system works tolerably well on the restricted class of musical sound'. More recently, Piszczalski and Galler [Piszczalski 1977, 1979, 1981] have, according to Moorer [Byrd 1984, p.78], 'refined monophonic music transcription to a fine art'.
Further work by Chafe [1982], Foster [1982] and others, used artificial intelligence techniques to automatically determine clef, key and time signatures, and enharmonic notation. However, Piszczalski and Galler [Piszczalski 1979, p.203] have asserted that 'the state of the art is nowhere near approaching the successful automatic transcription of polyphonic music recorded on a single track.'
Hence, it can be concluded that recognizing music from a
-9-
soundtrack combines all the problems associated with using a directly-connected instrument for input with those involving spectral analysis and psychoacoustics.
2.2.2.3 The compromise solution to the input problem
From the previous discussion it will be realized that, except in very restricted circumstances, methods which appear ideal do have flaws which prohibit their use as sole input devices. Hence, whether a musician or non-musician is entering the music data, the problem becomes one of optimizing the man-machine interface, with any musical ability on the part of the operator serving to speed this interaction. The main types of peripheral device available to any user for entering music data are:- QWERTY keyboard, digitizer (graphics tablet), and pointer (light-pen/mouse/joystick/tracker-ball).
As has been previously stated, the three basic types of information present on a page of printed music (pitch, rhythm and other symbols) tend to associate themselves with particular peripherals, depending to a great extent on the type of music being encoded.
The QWERTY keyboard is always available, which is its main advantage. With suitably defined function keys and/or an overlay together with carefully designed software, the ordinary keyboard can be quite successful, although it is still physically limited by the number of keys. Some systems still rely on the QWERTY keyboard for direct input of alphanumeric music-representational
-10-
data, although more use is being made of menu-driven systems with on-screen icons [Hewlett 1988].
The graphics tablet has been separated here from other 'pointing' systems because, where pitch is concerned, it has a unique advantage, i.e., the actual printed score (assuming one exists) can be placed on the tablet (a page at a time) and the noteheads picked out with the pen or stylus, a method successfully used by Wittlich et al, [1977, 1978a]. Hence, this is the only 'pointing' system which uses the score directly. This speeds up entry considerably because it eliminates the repeated movement of the eyes from score to screen. Alternatively, the tablet can be divided up (using an appropriate overlay) into regions, each representing a particular symbol, pitch, etc., so that all aspects of the music may be entered in this way.
The different types of pointer - light-pen, mouse, joystick, tracker-ball - work in a similar way, in that they select symbols from a menu, but vary in their ergonomic convenience. Use of these devices facilitates screen feedback, hence the adoption of abbreviations (e.g., for repeated pattern entry) or short-cutting techniques, such as having various permutations of chord patterns or arpeggio figures available in the menu.
2.2.2.4 Practical assessment of input methods
The following list gives a general view of which peripheral
-11-
best suits each category of music information:-
1. Pitch
Digitizing tablet (tracing the score with a stylus) i) Organ keyboard (depending on the complexity of the music and the musical ability of the operator)
ii) Pointers
2. Rhythm
i) Pointers (+ menu of patterns on screen)
ii) Digitizing tablet (+ menu of patterns)
iii) Organ keyboard (+ metronome click-track)
3) Text, slurs, etc.
i) Pointers or QWERTY keyboard
ii) Digitizing tablet
The order chosen for 2 i/ii and 3 i/ii above is determined by the greater inconvenience involved in moving between score, screen and tablet as opposed to score and screen.
In a test undertaken by the author, the graphics tablet method of entering letter names of pitches has yielded results of approximately 90% accuracy. The opening three lines of the flute part for J. S. Bach's Sonata in C were scanned three times each, with individual lines averaging 48 notes taking about 30 seconds. The results were obtained using a graphics tablet having a resolution of 0.001", i.e., 1,000 points per inch, or approximately 60 points between stavelines.
-12-
Taking into account the improvements that could be made to the system used, the results were quite promising. Modifications might include using a cursor with a lens and cross-hair instead of the pen utilized in the test system, plus some form of guidance as to the proximity of the pointer to one of the horizontal pitch lines (i.e., stavelines and centre-lines of spaces) in order to permit re-entry of wayward values. It has to be accepted that in general the stave will not be horizontal and any calculations involving vertical measurements must take this deviation into consideration. Although the bowing of stavelines (mentioned as a problem by Wittlich et al. [1978a]) was not encountered, it must be planned for when indicating the position of the stave (i.e., not only indicating end-points). For a non-musician, a 100-notes-per-minute, 90% accurate system for entering pitch values (letter names only) might form a valuable part of a practical set-up. Wittlich et al. [1978a] cited user-friendliness and speed as the most favourable features of this sort of system.
A deciding factor that emerges here is the type of music involved. For example, the above test was carried out on a sample of Bach, which consisted of more-or-less continuous semiquavers, making entry of the rhythm values particularly easy for a 'mouse and icons' or similar system, with good repeat facilities. In certain cases, however, the pitches also form patterns and this may be an example where a mouse could be used to achieve results faster, picking from a limited set of previously defined patterns, rather than entering all pitches via the graphics tablet or organ keyboard. On the other hand, types of twentieth century music which dispense with the key signature and employ numerous
-13-
accidentals could not beneficially be entered using a system where one method is used for entering the letter names of pitches (i.e., organ keyboard or tracing the score on a graphics tablet) and another is used for specifying the desired accidentals. Thus, the choice of method can be seen to be related to both the type of information being encoded (pitch, rhythm or other symbols) and the style of the music.
During the design of Music Representational Languages, much effort has been expended to support abbreviations, for instance to avoid repeated encoding of constant rhythm values or to provide a means of tagging certain patterns (of rhythms or pitches) so that these can easily be recalled. However, this feature must surely be most easily implemented by the software controlling a 'pointer and icons' system rather than the syntax of the M.R.L. (Music Representational Language). The only possible exception to this might be where the QWERTY keyboard is used to enter raw data, where minimizing the number of keystrokes involved in encoding is paramount. In this case, a cursor-key-controlled 'pointer and icons' system may still be faster and will certainly be easier to learn and use [Carter 1984]. As the abbreviated M.R.L. data would almost certainly have to be expanded to the 'long-hand' version at some stage (as in user-DARMS to canonical-DARMS discussed in section 2.3.1.1), program simplification would be achieved by having the user-interface software output standard M.R.L. data. Abbreviations in M.R.L.'s are covered in Section 2.3 - REPRESENTATION. Optimization of the man-machine interface is discussed, for example, by Card, Moran and Newell [Card 1983] in the context of a formal research project, and by Freeman [1986] in
-14-
a personal assessment of one particular application (word-processing).
At this point, mention should be made of Armando Dal Molin and his purpose-built music-entry terminals. Dal Molin's significant contribution to computerized music-setting and printing is covered in numerous publications [Dal Molin 1973, 1975, 1976, 1978] [Sargeant 1971]. His Musicomp (PCS 500) terminal [Dal Molin 1978] utilizes a combination of pitch keyboard (i.e. four stacked octaves of white keys) and music-orientated QWERTY keyboard built side-by-side into the console. This permits use of the left hand for Pitch and right hand for Character (symbol) and Spacing, hence the acronym PCS. Good on-screen feedback and editing facilities are significant features of the system, which produces near engraving-quality output on a phototypesetter.
2.3 REPRESENTATION
In addition to an input method, a means of internally representing the score is required, as all or part of the information present in the score must be stored electronically. A score can be stored as a graphic image, i.e., as an 'electronic photograph', so that the holding computer has no knowledge of the meaning of the symbols in the image, but this method is of limited practical use. Alternatively, the information contained in the score can be stored using a Music Representational Language (M.R.L.), enabling reconstruction of the score using appropriate software, albeit with inevitable slight differences in layout.
-15-
A Music Representational Language is a symbol system, normally using ASCII characters (but sometimes binary or decimal numbers) to represent musical pitches, rhythms or other miscellaneous notational symbols. For example, the letters A to C might represent the pitches of the same names, and the numbers 2, 4 and 8, the durations of a half, quarter and eighth note (minim, crotchet and quaver). Hence, using this simple and restricted M.R.L., the following extract would be written G8 G8 A4 B4 C8 D8 G2.
Obviously, for a useful M.R.L., more symbols would be required to represent octaves, clefs, barlines, rests, dotted notes, slurs, etc. Another factor emerges here, that is, the amount of decision-making which should be built into the score-reconstruction' software, instead of being included in the M.R.L.. For example, in the music extract above, the pairs of quavers are beamed, but this is not indicated in the M.R.L. data; a score reconstruction might appear with separate quavers. Thus, either another symbol could be included, indicating beamed notes, or an automatic beaming facility could be included in the score- reconstruction software. The latter might cause problems, however; for instance it could not be determined automatically how, or even if, a number of notes should be beamed.
-16-
Once the score has been converted to M.R.L. data, it can be analysed by software, edited, performed electronically or reconstructed for printing. The hardcopy can be in the form of proofs or finished artwork, consisting of score, separate parts, or piano reduction. Thus, assuming that an M.R.L. data version of a score is desirable, the researcher has the choice of either using an existing M.R.L. or creating his own.
Unfortunately, in the past, music researchers have chosen, for several reasons, to invent their own M.R.L.'s, and continue to do so (see [Gourlay 1986]). In most cases, the M.R.L. was tailored to the particular requirements of an individual research project, and as the nature and scope of projects differed, so did the requirements of an M.R.L.. This variety created the problem which still exists today, i.e., the lack of a standard encoding scheme for representing music scores. The recently established Musical Instrument Digital Interface (M.I.D.I.) standard provides the equivalent for musical sound. The pressing need for a standard has frequently been stressed by authors writing on the subject of computers and music (e.g., [Lincoln 1970b] [Morehen 1979]), but researchers who have encoded large quantities of data using their own particular M.R.L. are understandably averse to re-encoding the information. This has meant that a barrier exists between workers in the field which obstructs the transportation of data and software for analysis, printing etc. Recently, a task group (the Music Information Processing Standards Committee) formed under the auspices of the American National Standards Institute has been asked to propose a standard for the interchange of musical
-17-
information. It is intended to use only printable characters as an extension of SGML (Standardized General Markup Language), the code which covers production aspects of text publication. The principal document produced by the committee includes an outline for the standard [MIPS 1988]. Also, a working group of the Music Library Association is currently examining the possibility of using encodings of music in library cataloguing [Hewlett 1986b].
Irrespective of whether the researcher is choosing an existing one or creating his own, various attributes of an M.R.L. have to be assessed; in particular the scope (i.e., completeness), structure, (hierarchical or sequential), and efficiency (i.e., compactness). Also to be considered are the use of mnemonics, number of passes, availability of software for syntax/semantic checking or translating, existence of a shorthand version, and compatibility with existing facilities.
2.3.1 Existing M.R.L.'s
Much of the literature on computers applied to music relates to the subject of particular M.R.L.'s [Heckmann 1967] [Brook 1970a] [Lincoln 1970a]. However, as few as six have been used in significant research, or have been specified thoroughly enough for practical use. A comprehensive survey of M.R.L.'s has been produced by Boody [1975], together with a more detailed analysis of six of these which met his list of criteria deemed necessary - but not sufficient - for a useful language. From consideration of these evaluations, three M.R.L.'s emerge as the most suitable for general use, with a definite ranking of DARMS, MUSTRAN, ALMA.
-18-
2.3.1.1 DARMS: Digital Alternate Representation of Musical Scores
A large proportion of the literature on computers applied to music mentions DARMS at some point. Originating in the 1960's, the DARMS Project has involved work by Stefan Bauer-Mengelberg (the M.R.L. itself), Raymond Erickson (Syntax-checker), Anthony Wolff and Bruce McLean (user-DARMS to Canonical-DARMS translater, [McLean 1980]), David Gomberg (score-layout programs) and Melvin Ferentz (production of printed score). There are three main papers [Bauer-Mengeiberg 1970] [Erickson 1975, 1983] which discuss DARMS in detail as well as three publications [Gomberg 1975, 1977] [Wolff 1977] covering aspects of representation and music-setting which form part of the DARMS project, with its aim of producing publication-quality music from DARMS data (See also [Gomberg 1972] [Wolff 1972] [Feldman 1973] [Bauer-Mengelberg 1972, 1974a, 1974b] [Erickson 1977]). A user manual has also been produced [Erickson 1976].
DARMS was specifically designed [Erickson 1983 p.176] 'to capture accurately all the information provided by the composer, but not those details of layout within the province of the engraver or autographer.' DARMS has become the most widely used M.R.L. because it can be used to encode most music published since c.1600 which uses Common Musical Notation (C.M.N.).
The encoder can make use of extensive abbreviations which are permitted in user-DARMS but these are expanded by the Canonizer software into their complete version, known as
-19-
'Canonical-DARMS'. Hence, there may be several versions of a score in user-DARMS, but there will be only one Canonical-DARMS representation. Briefly, the representation uses numbers for pitches (21 to 29 cover lines and spaces on the stave), letters
for rhythm-values (Q, E, S, etc Quarter, Eighth and Sixteenth
notes respectively, i.e. crotchet, quaver and semiquaver) and
other ASCII characters for miscellaneous symbols, e.g.:(# ,-, *) =
respectively. Chords and polyphonic music are catered for, the latter by a method called Linear Decomposition Mode, i.e., encoding one voice per pass over the score.
2.3.1.2 MUSTRAN: Music Translater
MUSTRAN [Wenker 1969, 1970, 1972a] was developed by Jerome Wenker, originally for ethnomusicological purposes, but the revised version, MUSTRAN II [Wenker 1972b, 1974, 1977] includes more 'art-music' notation, representing C.M.N. almost as comprehensively as DARMS.
MUSTRAN (II) uses more mnemonic symbols than DARMS and has better software support (translator and utilities). The Indiana University Computer Music System [Wittlich 1978b], which incorporates Donald Byrd's SMUT (System for Music Transcription) software [Byrd 1984] uses MUSTRAN as its Representational Language. It was chosen because of its mnemonic code, relative completeness and existing translator and syntax analyser. Both the original MUSTRAN and MUSTRAN II have been described in detail by Wenker [see Wenker 1969 to 1977] while details of the numerous MUSTRAN utility programs now available at Indiana University can
-20-
be found in [Hewlett 1985].
2.3.1.3 ALMA: Alphanumeric Language for Music Analysis
ALMA [Gould 1970], which can cope with most C.M.N. (including polyphony), evolved from the 'Plaine and Easie Code' [Brook 1964, 1965, 1970b] invented by Barry S. Brook and Murray Gould, which while sometimes used (see [Rösing 1985]), is limited to monophonic music representation. ALMA supports abbreviations, optional multi-pass encoding, user-determined order of encoding and user-defined representational symbols. ALMA also supports a novel feature which helps with encoding repetitions, i.e., cyclic duration definition. This means that a recurring rhythmic pattern can be encoded once only and it will automatically be cyclically applied to following groups of the same number of notes.
A common problem with both MUSTRAN and ALMA is the use of letters to represent pitches, for although this seems appropriate at first (being suitably mnemonic), it means that a separate octave indicator has to be used. For example, in ALMA, the apostrophe (') is used once for each octave above middle C and commas (,) correspondingly indicate octaves below middle C. Thus, a passage of music which alternates across the boundary between two octaves will, when encoded, consist of numerous commas or apostrophes compared to the number of actual pitch symbols. This inefficiency is overcome in ALMA by an abbreviation facility which allows the encoder to indicate only a change of octave relative to the present one (+ or - a fourth relative to the present note). Obviously, this problem does not occur in DARMS, with its numeric
-21-
pitch-representation system.
2.3.2 M.R.L. structure
The structure of the M.R.L. is for the most part determined by the method of encoding polyphonic music. Various ways of referencing the notes in multi-part music have been adopted or proposed (e.g., [Clements 1980a, 1980b] [Maxwell 1983, 1984]).
One method is to encode voices separately with a tag of some sort to indicate which is which, but this can be wasteful if a composition consists almost entirely of one voice with just a few chords, since the subsidiary voices will have to be encoded throughout. Alternatively, encoding can be orientated around vertical 'time slots', so that all notes occurring simultaneously may be referenced together. Different voices then 'switch in and out' as and when they are required. Both Clements and Maxwell have favoured the latter approach, defining everything in the time domain, because the system thereby avoids any constrictions which would arise through the use of a musical hierarchy. Clements [1980a] also describes a 'neutral' M.R.L. implemented at the University of Western Ontario, which contains enough information in its data to facilitate ready conversion to either a 'sound output file' (for performance) or DARMS format (for printing).
2.3.3 The Editor
The minimum features of an editor have been formalized by Clements [1980a], as have the varieties of operational scope
-22-
needed [Buxton 1980, 1981]. Basically, the editor requires a data pointer, together with the following functions, the operation of which should be self-evident: delete, insert, change, transfer, search, substitute and transpose. The five categories of scope defined are: simple, block of time, local attributes, contextual attributes and named structural entities. In order, these refer to single note or entire score, notes within a certain time interval, notes encompassed within limits of pitch and duration, a note or notes within a certain musical context, and notes of a defined motif or theme.Various arguments have been put forward supporting different structures for M.R.L.'s; when consideration is given to the incorporation of an editor into the system, however, opinions [Maxwell 1984] [Clements 1980a] come down in favour of a sequential, time-ordered structure. This facilitates access to the data by the editor because information which is close together in the notated score will also be close in the Music Representational Language form.
It should be stressed that when entering information, the user can be buffered from the raw M.R.L. data by. the use of certain types of user-interface, as described in sections 2.2.2.3 and 2.2.2.4. Hence, different assessment criteria must be applied to M.R.L.'s when data is input via a QWERTY keyboard, compared to, say, a mouse and icons system. When using the former, as has been stated previously, minimizing the number of keystrokes is of paramount importance, so abbreviations (shorthand) and mnemonics (for ease of use) would be prime factors for consideration.
-23-
If another means of input is being employed, however, e.g., organ keyboard or graphics tablet, then comprehensiveness and efficiency are perhaps the most relevant features to be considered.
The issue of translation seems to be gaining in importance at present, as researchers accept the fact that there is no standard method of representing printed music, and opt for encoding data in as crude a form as possible so that it can then be translated into whichever M.R.L. is later required. If only pitches and rhythms are required initially (perhaps so that some form of analysis can be undertaken), and then later printed scores are required and only DARMS-compatible printing software is available, then it should be reasonably easy to convert the encoded data into the DARMS format and at the same time add previously omitted information. However, despite solving the problem of researchers' tenacious defence of their own M.R.L.'s (i.e., this approach allows them to retain their own data format), the result is a proliferation of translation software.
2.3.4 Problems concerning score representation in the context of printing
One example of this has already been mentioned, i.e., whether to indicate beaming in the M.R.L. or to try to formalize the rules for beaming of notes and to incorporate this knowledge in the score-reconstruction software. To take another example, the slur
-24-
shown below
is 'attached' to the two notes when included in the M.R.L. data, but its exact position and curvature will probably be determined by the music-setting software. Some M.R.L.'s do, however, make provision for stem direction (for example) to be encoded (e.g., U and D for up and down stem respectively) if desired.
A detailed discussion of some of the problems involved in representing all the complexities of music notation can be found in Wolff [1977]. For example, sometimes too many notes are present in a bar, either because the notes are an appoggiatura written in small notation or because a triplet sign has been omitted. These features have to be indicated in some way to prevent syntax-checking software producing an error. Another problem relates to horizontal positioning, where notes to be sounded simultaneously are of a similar pitch and must be separated out horizontally; this separation can also lead to 'too many notes in a bar' as well as variations in layout.
In coping with the above cases there is a trade-off between the interpretative ability of the encoder, the knowledge-based rules programmed into the score-reconstruction software and the graphical information included in the M.R.L.. Donald Byrd [Byrd 1984] asserts that fully-automatic reconstruction of high-quality music notation is not possible without artificially intelligent
-25-
software for this purpose. As this is not available at present, a compromise has to be reached; hence the trade-off mentioned above.
Most of the problems concerned with score-representation seem to be caused by the transfer from two dimensions (pitch and time) to the single-dimensional string of ASCII characters. These difficulties are compounded by the fact that there is not always a one-to-one correspondence between music-notational symbol and ASCII character(s). This is due to the graphical variations which can occur in some musical symbols (e.g., slurs).
2.4 RECONSTRUCTION
It is the actual hardware used which has the most influence on reconstruction. Once the master copy of the printed page has been produced, it can be mass produced by traditional means, such as offset lithography. The software of a music printing system can be designed to be device-independent, at least within certain categories of printer. It is easier, however, to convert from a vector-image (or random-scan) construction scheme to a raster-scan format than vice versa. Vector-drawing image construction means building up the image by drawing line segments (vectors); raster-scan implies plotting a large number of points while making horizontal sweeps over the imaging area. In the former case, small segments may not be drawn accurately and producing dots can prove to be a problem, whereas with the latter it is the number of points per unit area (i.e., the resolution) which is crucial to producing a good image (which appears not to be made up of
-26-
individual dots).Numerous types of printer are available for producing hardcopy computer output. Music printing systems have tended in the past to use three main varieties, the dot-matrix printer, plotter and phototypesetter. The primary features which determine these choices are price, resolution and speed. The printers' best attributes are shown by category in the following table. The phototypesetter, for example, provides the best resolution and speed at the worst (= highest) price.
| price | resolution | speed | |
|---|---|---|---|
| dot-matrix | 1 | 3 | 2 |
| plotter | 2 | 2 | 3 |
| phototypesetter | 3 | 1 | 1 |
The dot-matrix printer is widely available and very popular because of its price and versatility. It can produce scores at a resolution of up to approximately 360 dots per inch (d.p.i.) which is adequate, although beams which slope only slightly will appear as a 'staircase' and slurs will not be smooth. These disadvantages can be minimized by producing oversized originals and then using photoreduction at the production stage. More often, though, the intention is to use the dot-matrix printer for producing draft copies for proof-reading or one-off prints for use in an educational situation or by a composer. Often the dots which constitute the image are large compared to the 'grid' on which they can be positioned, so that adjacent dots overlap and, for
-27-
example, a line two dots wide would be less than twice as wide as a line of single dot thickness.
In the past, the plotter has been by far the most widely used output device for computer printing of music within the context of research projects (see section 2.5). Its two main forms are the X-Y flat-bed pen plotter and the drum plotter. In the former, the pen travels over both axes, whereas with the latter the pen moves over one axis only and the paper moves over the other. Although the plotter is relatively slow, it can resolve up to a thousandth of an inch, but then suffers from a problem similar to the dot-matrix printer in that the ultimate resolution is decided by the pen being used, and not by the 'grid' upon which it is positioned. Again, photoreduction can be used to improve the perceived resolution and assist in eliminating 'staircase' effects.
The phototypesetter also exists in two forms: mechanical and digital [Seybold 1983], but the latter is rapidly taking over from the former. Both types produce output on film but the mechanical typesetter uses templates and an optical system to produce very high quality images with an accuracy in the region of +/- one thousandth of an inch. The use of templates dictates that the character-set is built into the hardware. This restricts the number of symbols to less than that needed to cover all orientations of beams and slurs, and is less flexible than the digital approach. In a digital typesetter, the number of characters/symbols available is limited solely by the on-line storage capacity. So, although the digital typesetter is the more
-28-
expensive variety (tens of thousands of pounds), it has enormous flexibility and is able to produce images of very high quality (approx. 2500 d.p.i.) making it the ideal music printing device.
More recently, other designs of printer have been developed and refined and hence have become available for music printing [Weber 1986a, 1986b]. Specifically, these are varieties of non-impact machines, such as thermal transfer, electro-erosion, electrostatic and xerographic printers.
Thermal transfer printers are capable of approximately 300 d.p.i., an example being the IBM Quietwriter 7 typewriter/printer, but the short life-span of the ribbons used leads to high running costs.
Electro-erosion printers (such as the IBM 4250) produce up to 600 d.p.i. (low typeset quality) by using a high-density printhead to burn away the metallic surface covering a black backing paper, which is then used to produce final output on film.
Electrostatic printers produce a resolution of up to 400d.p.i. Their print quality and contrast, however, are not as good as the xerographic printers described below. A disadvantage of electrostatic printers is their requirement of special paper.
Of all xerographic (or electrographic) printers, the laser printer is, at present, the most popular form. It is being used widely for music printing (see section 2.5) where it is seen to benefit from some of the advantages of the phototypesetter (high
-29-
speed and resolution) but with none of the disadvantages of the plotter and dot-matrix printer. Other varieties of xerographic printer use a different light source such as light-emitting diodes, or a liquid crystal shutter to control the light. Working on principles similar to those of a photocopier, the modulated light source directed onto a rotating drum produces an image on the charged drum surface. This is then used to transfer toner onto the paper, where it is heat-fused. A diode or liquid crystal shutter as the light source or controller, respectively, gives the printer greater reliability because it cuts down on moving parts. Speeds for laser printers vary between six and 200+ A4-size pages per minute, prices are from £1,000 upwards, and maximum resolution is about 600 d.p.i.. Duplex printing (i.e. on both sides of the page) and A3 paper-handling - both important features for music printing - are now becoming available in cheaper machines.
Normally, assuming a laser printer (or similar) has the facility to address all points over its imaging area, the information actually transmitted over the interface between host computer and printer is a coded description of the original image. This produces significant data compression compared to transmitting a complete bit-map representation. A raster image processor (R.I.P.), normally contained in the printer, converts the transmitted code into a raster image and controls the marking engine itself. It is the R.I.P.'s own command language which determines the text/graphic printing capabilities of any particular printer. There is a similarity between the laser printer and phototypesetter in that, in both cases, the host computer has merely to transmit information regarding which symbol
-30-
(character) is to be printed, where on the page, and perhaps in what orientation. The printer can then construct the bit-map form of the image and thus offload a large amount of work from the main computer. Currently, the effective standard for this type of coded data transfer between computers and printers is Adobe Systems' 'Postscript', although other page description languages (PDL's) are in use, for example Xerox's Interpress. In practice, this means that the same data can be sent to an appropriate laser printer or phototypesetter, and the only difference in the output image will be the change in resolution. Also, 'Display Postscript' is now becoming available, providing even more device-independence, by enabling a Postscript encoded image to be displayed on-screen.
2.5 APPENDIX
The following table can be viewed as an enlarged version of the survey given in Donald Byrd's thesis 'Music Notation by Computer' [Byrd 1984]. References have been added pertaining to systems which Byrd mentions. Information on systems which were recent at the time of his writing, has been expanded upon, and details of some new systems have been added. Also, the information is presented in a clear tabular form. Ideally, the same features of each system would have been examined to facilitate direct comparison; the systems vary so much, however, that this is not possible, especially considering the varying amount and age of information which is available. Examples of output from significant systems are to be found in [Clements 1980a] and [Hewlett 1985, 1986b, 1987, 1988]. Where the information provided
-31-
is based on [Byrd 1984], the subjective descriptions of output (e.g., 'high-quality' scores) have been retained and where the current author's opinion is being expressed, a similar scale has been adopted, i.e., low or poor, adequate, reasonable, high, very high or very good, excellent or engraving-quality.
As the price of technology falls, significant developments are occurring at a rapid pace. For example, the hardware used in the Mockingbird system developed at Xerox's Palo Alto Research Center, deemed 'prohibitively expensive' in 1984, may in the not-so-distant future come within reach of most music publishers' budgets, if not those of music copyists. More music printing software is becoming available for popular microcomputers, which is having a significant impact, especially where the Postscript page description language is used, with its aforementioned flexibility. The possibility of proofing output on a laser printer and then using a bureau service to produce final typeset material from the same data is proving attractive in music production (as well as general graphics applications) because, while giving access to top-quality output, it removes the need to invest heavily in typesetting equipment.