Score Perfect Professional 1-4 file format

How far this format is understood, in numbers, and what is known, guessed and still open. The numbers are against the 116 songs of the corpus that SPP Lite 5.1 could open and save again in the SPP 5–7 format (the twins), and against PriMus' son2emil. Of the 119 songs of this format in the corpus, the 3 of SPP 1.0 cannot be opened by Lite; all 119 are read, and the 5 of SPP 1.2-K are another format.

Files up to SPP 4 start with f0 ee ed. The corpus holds 119 of them: 18 of the songs the converter was written for, the songs that come with the SPP 4.0 demo, those of the SPS and SPE 4.0 demos, and 3 of the SPP 1.0 disk. Their version strings are 11.0, 11.4, 12.0, 12.1, 13.0, 13.1, 14.21, 14.90, 14.92, 14.93 and 14.94: a 1, then the version of SPP that saved them (son2emil names 13.0 format version 3, and 14.94 4.94), so from SPP 1.4 to 4.94. The layout found so far is the same in all of them.

SPP 1 and 2 ran on the Atari platform only. SPP 3 and 4 were also available for Windows, and later for the Mac, presumably by emulating the Atari: the Windows version of SPP 4 looks like an Atari program, drawing its own menu bar. The format fits that origin: big endian, the byte order of the Atari's 68000 processor, and text in the Atari's character set. SPP 5 was a rewrite, for Windows only, with the SPP 5–7 format. SPP 4 reads all these files; the demo of SPP 4.0 opens them but cannot save. SPP Lite 5.1 opens them too, down to SPP 1.4, and saves them in the SPP 5–7 format.

The Standard (SPS) and Education (SPE) editions are assumed to use the same format as SPP: Scoretec offered version 4.0 in all three, and later said that songs could be exchanged between them. Their 4.0 demos cannot save, but come with songs (8 for SPS, 26 for SPE): all start with f0 ee ed, carry the same version strings (11.4 to 14.93), read with the same reader and convert without errors. 15 of them are byte for byte the demo songs of SPP 1.4 to 4.21; the other 19 (from SPP 4.90 to 4.93) are not in the SPP demos.

SPP 1.0 (version string 11.0) writes this format already. SPP 1.2-K, the short version of 1.2 ("Kurzanleitung"; the program says 1988 and 15.7.1989, a year before SPP 1.0 of 26.6.1990) that came on the utility disk of the magazine KEYS and was made for it, by the same author, Klaus Kleinbrahm, a music teacher (his own words in the disk's LIESMICH.TXT; it needs an Atari ST with 1 MB and runs in the high resolution only, as all of them do: "läuft nur in der hohen Auflösung"). The text says that songs made with 1.2-K can be loaded and edited by the main program, Score Perfect itself. Unlike SPP 1.0, it does not write this format: its songs start with f0 ee ec and are text, lines ending in 0d 0a as written by a BASIC program: the version "1.2", the title, a line of numbers, then per measure a line with the length of its notes and a line with the notes in quotes, as fe, number, type, position records. Not decoded; the converter says "SPP 1.2 format: not supported". The corpus has 5 of them, from the SPP 1.2 disk image.

It is the format of the time before PriMus forked off from SPP: PriMus' son2emil.exe (in PriMus Free) converts these files to its text format, emil, and refuses the SPP 5–7 format.

One file of the SPP 4.0 demo (Lieder - ALTA_TRI.SON) is no song: all its 7708 bytes are e5, so the data was never written properly or was lost in copying.

Every song was saved again by SPP Lite 5.1, in the SPP 5–7 format, driven by a script under Wine; the converter reads those, and all 116 convert to valid MusicXML, with as many measures as son2emil counts where it has been run. The 3 songs of SPP 1.0 have no pair: SPP Lite 5.1 cannot open them, and saves an empty song; the SPE and SPS songs have no son2emil output. So each song comes as a pair: the same music in both formats. research/oldformat.py lines the SPP 1–4 file up against the converter's reading of the SPP 5–7 one, field by field; research/compare_older.py compares the converter's reading of both. son2emil's output of the originals is a second reference. In the corpus, the resaved songs are in songs/old_resaved_by_spp51/ and son2emil's output in songs/old_converted_by_son2emil/, each in the same collection and version folders as the originals.

What SPP 5.1 makes of these songs, and what that taught about the SPP 5–7 format, is on the SPP 5-7 file format page.

Later findings came from two other sources. PriMus' son2emil.exe was decompiled with Ghidra: it reads the notes' optional records, the lists after the measure arrays, the layout and the hairpins, so its code says what a field is (the f6 record, the hairpin walk). The test songs research/test_songs/maarten/ were made by hand in SPP for the SPP 5–7 format, with one thing changed at a time, and are on the SPP 5–7 format's page.

• Integers are big endian: 0c 00 is 3072.
• Strings have a 2-byte length in front: 00 06 4f 62 6f 65 20 32 is "Oboe 2".
• Text is in the Atari ST character set, which the converter reads as code page 437 with two exceptions, 9e ß and a9 ©: 81 ü, 84 ä, 94 ö, 9e ß ("Baß", "Händel"). But the Windows SPP 4.93 stored a typed © as a9, its Windows byte, which is ⌐ in the Atari set.
• Durations are in divisions of 768 per quarter note, as in the SPP 5–7 format.
• Staff positions count diatonic steps, C4 is 35, plus 100 for a note head on the other side of the stem, as in the SPP 5–7 format.

From the header to the time signature table, the file is read block by block (in SPP 2 and later); the notes are found by their shape, as the sections before them are not all decoded. research/layout_older.py prints this map for a song.

All 119 songs are read this way.

Checked in all 119 songs:

Right after the header, per staff: its name, a string, then 15 bytes.

Right after the staff records: settings of a fixed length, 122 bytes from SPP 3.0 on, 121 in SPP 2 (the one song of SPP 1.4 has shorter ones, not worked out). The 6th short (bytes 10-11) is the tempo in beats per minute, equal to the tempo of the resaved songs in all 116; Bach's Toccata has 88. They start with 00 05 to 00 08 and hold the guitar tuning 34 39 3e 43 47 4c (E A D G B E, as in the SPP 5–7 format); not decoded further.

Then three lists of staff ranges: each its length in bytes, then pairs of staff numbers, first and last. The first list holds the braces, the second the brackets, as the SPP 5–7 format's part groups (in the other order); they agree with those of the resaved songs in all songs. The third list, other ranges, comes back unchanged as list 6 of the SPP 5–7 format. Beethoven 5: braces 19-20, brackets 1-8, 9-13 and 14-20. Then 2 bytes (00 01, 00 02 or 00 04), and then the time signature table.

A table of the time signatures the song uses: its length in bytes, then 12-byte entries:

After it three arrays, each the number of measures, then a byte per measure:

• the time signature: an index into the table, from 1 (MEL_10 changes from 3/8 to 2/4);
• the key, + 20, concert pitch;
• the bar line at the end of the measure: in the low 4 bits the SPP 5–7 format's bar line code minus 1, so 00 dotted, 01 normal, 02 double, 03 final, 04 repeat end, 05 repeat start of the next measure, 06 and 07 repeat end and start. Bits 10, 20, 40 and 80 are not the bar line: see below.

Against the resaved songs, the keys agree in 96 songs, the time signatures in 94 and the bar lines in 88; the rest differ by multi-measure rests and the bits below.

The bar line byte has more than the bar line, found by reading son2emil: bit 80 is "hide this staff" (son2emil writes HideStaff after the bar line, and ShowStaff where the next measure has the bit off), and bits 10, 20 and 40 of the second measure's byte are one song-wide number, the style of the barline through a staff group that son2emil writes in Set grouping: 0 continuous (1-2), 1 1:2, 2 m, 3 and others other characters (+ is 5, ' is 6). The other measures repeat the bits (Beethoven's Fifth: all 502 measures 3x, Locus iste: all 48 1x), which is why they are on every measure of a song or on none. The converter reads none of these.

Right after the per-measure arrays come lists, each with its length in bytes first. The first list has records of 6 bytes, ff, 00, a measure, a kind and 2 bytes, for example ff 00 0f 05 31 54; it is empty in most songs (00 00) and its meaning is unknown (kinds 1, 2, 4, 5, 7, 9, 11, 13 occur, and two measures of Beethoven's Fifth, second movement, where SPP Lite 5.1 numbers one lower have a kind 1 record). The second list holds the voltas: 00 7e bytes (7 per record), then per record ff ff and 5 bytes: 00, the measure (from 1), the number of the ending, 78 00. A volta of several measures has a record for each, so the first measure of a run of the same number is where SPP 5.1 puts it. After the voltas come the texts. All 115 voltas of the 97 songs are where SPP Lite 5.1 has them; before the first list was taken into account, five big songs had none and the records of the MEL songs seemed to have a first byte that was not 00.

The bar line codes 08 to 0f of the low bits are not known: the converter logs them. The bar line bits (see above) have nothing to do with voltas: none of the measures with a volta has any of them.

After the voltas: records separated by ff ff, first the texts, then the numbered sections (below). A text:

The converter puts every text above the staff; the title is the first of the texts in the title style (the lowest measure, then the lowest staff, then the first in the file: the same as SPP Lite 5.1 makes of it in 67 of the 69 songs that kept a title) and is not kept as a text, the others are texts, and a header title that is "Kein Titel" is none; there is no composer or subtitle field in this format, so Toccata's "J. S. Bach", in the same style, stays a text and not the title as the header's own title string has it, and a text outside the song, or one whose length does not fit, is skipped. The header's title is often "Kein Titel"; the text in the title's style is the real one ("Menuett II", "INVENTIO 1"). 565 of the 870 texts of the resaved songs match; SPP 5.1 itself drops texts and moves some in long songs (by 156 measures in Beethoven's Rondo), where son2emil agrees with the original.

After the texts the file goes on with records separated by ff ff, each starting with its number: 02 the clef changes, if there are any (see Clef changes); 03, 04 and 05, short and not decoded; 06 the MIDI mixer, with its labels in Atari text (LAUTSTÄRKE, HALL, CHORUS, PAN), not decoded; then in most songs 08, 09, 0a, 0b and 0d or 0e, not decoded; some songs have a section 00 before the texts. A section can hold ff ff itself, so they cannot be cut apart by splitting. In some songs section 09 holds a 2-byte number per measure, the measure numbers as SPP shows them, with a number repeated where a measure of a first ending repeats one; SPP Lite 5.1 does not keep them. The layout strings, the last thing before the notes, are not a section of this list (see Hairpins); so the notes block is still found by its shape.

After the texts: a record starting 02 (after ff ff), then per staff a length (6 per change), then per change ff ff and 4 bytes: the measure (2 bytes, from 1), the number of the note in it (from 1, rests count), the clef (the code of the SPP 5–7 format, 00 G, 01 F). In Bach's Toccata and Fugue: 02 | 00 00 | 00 0c | ff ff 00 11 01 01 ff ff 00 12 02 00 | 00 72 | ..., so no change in the first staff, then an F clef at the first note of measure 17 and a G clef at the second of measure 18 in the second. All 371 changes of 33 songs sit where SPP Lite 5.1 has them.

Per staff record (each voice, see staff records), one block per measure: a 2-byte length, then the notes; then an empty block, ending the staff (the last staff's can be missing). An empty measure is an empty block too, and the empty measures after the last one with notes are left out: the staff record says how many measures are stored. The block is found by its shape; as the search finds its first note, the first staff's leading empty blocks lie before that and are counted back. Found in all 97 songs. With the stored counts, note counts and pitches agree with the resaved songs in 99.6 % of the notes.

The optional record fa (crescendo) or fb (diminuendo), 13 bytes: the code, then pairs (high, low) with the value (high − 50) × 100 + low: the start offset a6, a y, the length a2, another y, and two more. It is the same record as SPP 5's tag d1 (there with 100 instead of 50). A crescendo is on its first note, a diminuendo on its last. The end is not in the file: PriMus' son2emil works it out from the places of the notes, and so does the converter (hairpins.py), found by reading son2emil with Ghidra:

• The target is the place of the note plus a2 × the zoom of the line / 200 + a6, in layout units. Forward (crescendo), from the next note on, the end is the note that is nearest to the target, nearest being the distance to the note's place plus 8, until a note is not nearer than the one before. At a barline the width of the measure is taken off the target. Backward (diminuendo) the same without the 8, and in both directions the first note has to be nearer to the target than the hairpin's own start is, that is, closer than |a2 × zoom / 200 + a6|, and the width of the measure before the one entered, counted from the one left, is added (an off-by-one in son2emil, followed here).
• The place of a note: its column (note byte 4) in the layout string of its measure, plus 2 and the first setting number (the 3rd short of the settings) × the zoom / 200. The width of a measure: the width in its layout string plus the 4th short of the settings × the zoom / 200.
• The layout strings, one per measure and one more, are the last thing before the notes: a 2-byte length, then 4 bytes, the width (2 bytes), 3 bytes, and records of 10 bytes (14 in later versions): the time, the x of the column, and 6 (or 10) more.
• The zoom belongs to a line, from the end of the file, after the notes: 9 shorts, the number of pages and a short, then per page the number of lines, and per line the zoom, the first and last measure, a short per staff and a string. A line with first measure 0 is no line. Measures that no line covers get the 7th short of the settings as zoom. When no note of the hairpin has a place in its layout string, the hairpin gets no other end and is left out.

Against son2emil's output for all songs: 271 hairpins agree, 5 are only in son2emil's output, 7 only ours. Of the songs with hairpins, only the SPP 1.4 one has no layout strings that can be found. A measure can have a short string (no columns); those of the MEL songs are, and without allowing them the chain does not add up.

After the fd records (or the tones) comes f6 and a mask byte, found by reading son2emil; records in this order. Lengths include the code byte:

Bit 01 of the stem byte says an f6 record follows (47000 notes). The 40 byte of the fd mask is read by son2emil and not used either.

The grace note: ec is there on all 369 grace notes SPP Lite 5.1 has and on no other note. The grace note keeps the duration of its type. After f6 may come e8, with the fd mask bit 01: a mask (bit 02 dotted slur, in son2emil's (:) and records e7, e6, e5 (hide stem) and e3; not read.

Not one record as in SPP 5: each measure of the run has its rest, ending in the record f6 02 f4 NN, NN the place in the run (1, 2, 3), as in fe 01 00 08 01 1a 21 64 01 32 f6 02 f4 02. A run is the measures marked 1, 2, ... in every staff; the converter makes them one measure with the count, as SPP 5 has it. In Mozart's bassoon duet 5, the only song where SPP Lite 5.1 made multi-measure rests, all four (3, 2, 3 and 3 measures) are found exactly where it has them.

The record f6 04 f3 CC (or f6 14 f3 CC, followed by an f1 record) and 5 more bytes (00 46 00 00 00 on the first note of a line, zeros after) is on every note under an octave line, CC the code of the SPP 5–7 format's e2 (only 01 occurs). A rest in between has none. In Beethoven's Für Elise (15 notes) and Das Lied vom Traum (152 notes) every note under a line is where SPP Lite 5.1 has it, with the same pitch in all but one.