Here are four files -- the short article about the program published in "Current Directions in Computer Music", the longer article, STAN-M-19, a final portion left out of the latter since it was specific to the operating system, and a translation of the SAIL code to C.
=========================================== AUTOMATIC COUNTERPOINT by Bill Schottstaedt
This file contains both the "Current Directions" paper on automatic counterpoint, and the original SAIL program, STAN-M-19, slightly rewritten to reduce duplicate verbiage.
For several hundred years composers have praised species counterpoint as one of the best ways to learn to write music. In this discipline the student learns to shape and combine simple melodies through a carefully organized sequence of baby steps. One well known exposition is "Gradus ad Parnassum" by J J Fux (1725).
A program was written following as closely as possible the exposition of species counterpoint given by Fux. Fux's guidelines and rules were used to solve counterpoint exercises automatically. The rules were then extended and modified until the results of the program were an acceptable match with Fux's examples. All five species in Fux's five modes were implemented, accomodating up to six voices. The program itself with a detailed commentary is available in Schottstaedt (1984).
WHY WRITE A COUNTERPOINT SOLVER?
There are several reasons to undertake this task. From a programmer's point of view, it is fun and not particularly difficult to write such a program. The world of species counterpoint is simple enough that the criteria for judging solutions are reasonably clear, and the results look deceptively like music. From a music theorist's point of view, the program can imitate an unusually inept but earnest student whose unexpected mistakes and misunderstandings can make it clear where the rules are inadequate. Finally, from a composer's point of view, species counterpoint seems like a plausible place to start in the development of a smarter working environment for composers of computer music.
Although Fux's definitions and rules are exceptionally clear and carefully presented, they are
incomplete. For example, in his list of the ways voices can move, Fux ignores the case in which
neither voice moves. Also, in the edition I used, Fux does not explicitly define the difference
between a skip and a step. Omissions such as these are easy to handle. It is more difficult to make
explicit the proper way to handle tradeoffs between rules and guidelines. For example, Fux
mentions that it is desirable to maintain "variety" in the melodies. This variety seems to entail a mix
of melodic intervals coupled with the avoidance of repetitions of any one pitch. The overall melodic
contour should be smooth, but not without some independent character. A program cannot satisfy
these constraints unless it can recognize good and bad situations, and make choices among them.
The first requires an explicit statement of what "smooth" means; the second requires some explicit weighting and searching method to decide when an acceptable solution has been found. In all but the simplest cases, each solution runs up against at least a few of the guidelines, so the program must be able to choose the "best" from many imperfect solutions.
The basic rules of species counterpoint describe how melodies are formed and combined. Fux repeatedly stresses that most of the rules are guidelines, not absolutes. In our implementation we define the relative importance of the rules by assigning each rule a penalty. The higher the penalty, the worse it is to break the associated rule. The relations among these penalties largely determine what solutions the program finds.
Prohibitions (infinite penalty):
- Parallel fifths are not allowed.
- Parallel unisons are not allowed.
- In two part counterpoint the final chord must be a unison or an octave; it can include the third or fifth when there are more than two parts.
- The next to last chord must have a leading tone.
- Dissonances must be handled correctly (this covers a host of rules).
- Voices must stay within the mode of the cantus firmus.
- Some melodic intervals are not allowed.
- A melody must stay within the range of a twelfth.
- Cadences must be correct (this covers a host of rules).
- Direct motion to an octave or fifth over three chords where the intervening interval is not at least a fourth is not allowed.
- A dissonance as a passing tone must fill in a third.
- A nota combiata must be resolved correctly.
- A ligature must be resolved correctly.
- There are certain situations where a ligature is not allowed.
- Augmented intervals are not allowed.
- Doubled leading tones are not allowed.
- The leading tone must be resolved correctly.
Very bad infractions (penalty is 200):
- Avoid direct motion to a perfect fifth.
- Avoid direct motion to an octave.
- Voices should not move outside certain ranges.
- Avoid a unison on the down beat.
Bad infractions (penalty is 100):
- Avoid unisons in two part counterpoint.
- Try not to repeat a note on the upbeat.
- A melody should stay within the range of an octave.
- Eighth notes should not skip.
- There are various rules concerning the rhythmic values of tied notes.
- A six-five chord must be prepared correctly.
- A six-five chord must be resolved correctly.
- Avoid direct motion to a tritone.
Other rules (penalty is given in square brackets):
- Avoid a sixth followed by motion in the same direction .
- Strive for triads .
- In fourth species, avoid passages without ligatures .
- Avoid direct motion to a perfect consonance in the inner voices .
- Avoid tritones near the cadence in lydian mode .
- Avoid a leap to the cadence .
- Avoid moving from a tenth to an octave .
- Avoid a skip to an octave .
- Avoid a sixth preceded by motion in the same direction .
- Avoid a tritone outlined by a melody .
- Avoid skips in all voices at once .
- Avoid a fifth or octave followed by motion in the same direction .
- Avoid a repeated pattern of four notes .
- Avoid notes on the outer edges of the ranges .
- Avoid leaps of an octave .
- Avoid doubling the third .
- Avoid doubling the sixth .
- Try not to skip from a unison .
- Avoid a repeated pattern of three notes .
- Avoid a fifth or octave preceded by motion in the same direction .
- Avoid any skip followed by motion in the same direction .
- Avoid two skips that do not fill in a triad .
- Do not repeat a note on the down beat .
- Avoid a repeated note on the fourth beat .
- Avoid a skip followed by a skip in the opposite direction .
- Avoid direct motion to a tritone in the inner voices .
- Avoid doubling the fifth .
- Avoid doubling the bass twice .
- Avoid a repeated pattern of two notes .
- Avoid perfect consonances .
- Avoid leaps of a sixth .
- Avoid a tritone between voices .
- Avoid a skip followed by a skip .
- Avoid direct motion .
- Avoid compound intervals .
- Avoid any skip preceded by motion in the same direction .
- Avoid upper neighbor notes .
- Avoid lower neighbor notes .
- Strive for variety in a melody (this covers several rules which try to avoid a melody made up entirely of skips, constant repetition of one pitch, constant use of the same interval and so on) .
- Avoid skips to a down beat .
- Strive for contrary motion .
- Try to keep the upper voices from wandering far apart .
- Avoid repeated voice crossing [penalty dependent on the number of crossings]
These penalty values arise partly from Fux's comments about which rules are more important than
others (these get a higher penalty), and partly from experience running the program. Obviously
some of the penalties do not apply to every species of counterpoint. The penalties are positive (that
is, there are no rewards), because we want to abandon a line as soon as its penalty gets too high.
This implies that the penalty function should not be able to decrease. Even the smaller penalties have a noticeable effect on the outcome of the counterpoint search. The solver is consistently "bent" in the direction determined by the penalties.
SEARCHING FOR SOLUTIONS
There are many ways to search for acceptable solutions to a counterpoint problem. The main constraint is compute time. If we make an exhaustive search of every possible branch of a short (10 note) first species problem, we have 16^10 possible solutions (there are 16 ways to move from the current note to the next note). Even if we could check each branch in a nanosecond, an exhaustive search in this extremely simple case would take 1000 seconds (about 20 minutes). Because we hope to handle problems far more complex than this simple one and hope to do it reasonably quickly, we must find a smarter search method.
After we made several false starts, Bernard Mont-Reynaud suggested that we use a best-first search.
We compute the penalty associated with each melodic interval from the current pitch, then continue recursively using the best result first. If forced to abandon a branch, we back up and try the next best interval until a complete solution is found. Unfortunately, the first such solution may not be very good. By accepting the smallest local penalty we risk falling into a bad overall pathway. If the program is told to search every branch, we once again get bogged down in long computations. To get around this, as soon as we get a solution from one branch, we drop back to the beginning of the exercise, and try a different beginning interval. By trying a new starting interval, we maximize the chance of finding a truly different solution. As before, we abandon any branch when its accumulated penalty is above that of the best complete solution found so far.
Even with these optimizations, fifth species remains a problem. The number of choices increases in this species because it generally has more notes in the counterpoint melody, and because we have several possible rhythmic values to assign each note. Several optimizations can be added to reduce this problem: we search only for those solutions that are much better than the current one, and once a solution has been found we don't spend too much time on any other interval. Compute times for multi-voice fifth species remain high despite all this machinery.
In early versions of the program we solved one voice at a time and built up the entire ensemble by layering. This is satisfactory for three voices because the first voice is usually a good melody, and the second voice still has enough degrees of freedom to find an acceptable solution. As more voices are added the later layers become less and less acceptable. To get a solution in which all voices are reasonably good, the entire ensemble must be calculated together.
See page 39 and 40 for a completely spelled out example.
Currently the program has no provision for starting a melody with a rest, nor does it reward invertible counterpoint and imitation. It tends to let voices get entangled in each other, and makes inadequate judgements about overall melodic shapes. Certain implied vertical clashes, tritones in particular, are not noticed. Given these caveats the program does well in a reasonably short time.
We give below a few examples of the program's output. It should be kept in mind that species
counterpoint exercises are not expected to be great music. The five species are 1) note against note,
2) note against two notes, 3) note against four notes, 4) note against two notes with ligatures, and 5)
note against a "free" counterpoint, that is a melody in different rhythmic values using any of the
possibilities of the preceding species. The numbers running beneath the counterpoint melodies show
the penalty associated with that note. The cantus firmus is the voice without any running penalties.
Each cantus firmus is taken from Fux.
J. J. Fux (1725). The Study of Counterpoint from Johann Joseph Fux's Gradus ad Parnassum, translated and edited by Alfred Mann. W W Norton and Co, New York, 1971.
W. G. Schottstaedt (1984). Automatic Species Counterpoint. Stan-M-19, CCRMA, May 1984.
========================================== Automatic Species Counterpoint (Stan-M-19)
Species counterpoint as presented by J. J. Fux in "Gradus Ad Parnassum" appears to be a ready made case for a rule based "expert system". In programs of this sort, knowledge is encoded as a list of IF..THEN statements. Obvious examples from Fux are:
IF Melodic-Leap > Octave THEN Try-Something-Else.
IF Interval-With-Bass = Fourth AND Species = First-Species THEN Try-Something-Else.
These attributes can easily be defined in such a manner that a computer program can use them to find acceptable solutions to species counterpoint problems. In this paper we present a program that can write counterpoint of this sort. Rather than get bogged down in circumlocutions, we provide the actual code, and the reader can if he so desires re-implement the entire program. The language used is SAIL, an Algol-like language fully described elsewhere. The original implementation was done in Pla, an offshoot of SAIL also described elsewhere. We translated everything to SAIL to speed up execution. This file is itself an executable program which solves species counterpoint problems as described in the text.
The octave is divided into 12 semitones. We can therefore give a unique integer to every possible pitch. In our examples, this value is the distance in semitones from the given pitch to low C (16 Hz).
Rhythms can be represented as a certain number of eighthnotes.
Notes within a voice do not overlap. Each note's onset time can therefore be represented as a certain number of eighthnotes from time 0. Of course a prior assumption is that there are things called voices that can be assigned a unique succession of notes. Each note has a duration, an onset time, and a pitch. If more than one voice exists, there is also a notion that a "vertical" interval exists between these voices equivalent to the melodic interval necessary to jump from one voice to the other.
The cantus firmus and the starting note of each counterpoint voice are specifed by hand. We also currently assume that the cantus firmus is entirely in whole notes, but it would not be difficult to remove this restriction.
We must first define our terms. These include the interval names, consonance and dissonance classifications, and so on. To define the intervals:
BEGIN "foox" REQUIRE 20000 SYSTEM_PDL; DEFINE !="COMMENT"; DEFINE Unison= 0, MinorSecond= 1, MajorSecond= 2, MinorThird= 3, MajorThird= 4, Fourth= 5, Tritone= 6, Fifth= 7, MinorSixth= 8, MajorSixth= 9, MinorSeventh= 10, MajorSeventh= 11, Octave= 12;
Each interval is defined by the number of semitones within it. These intervals are impervious to octave interpolations and transpositions.
We classify each interval as either a perfect consonance, imperfect consonance, or dissonance:
BOOLEAN ARRAY PerfectConsonance[0:12]; SIMPLE PROCEDURE PerfInit; BEGIN PerfectConsonance[Unison]:=TRUE; PerfectConsonance[Fifth]:=TRUE; PerfectConsonance[Octave]:=TRUE; END; REQUIRE PerfInit INITIALIZATION; BOOLEAN ARRAY ImperfectConsonance[0:12]; SIMPLE PROCEDURE ImperfInit; BEGIN ImperfectConsonance[MinorThird]:=TRUE; ImperfectConsonance[MajorThird]:=TRUE; ImperfectConsonance[MinorSixth]:=TRUE; ImperfectConsonance[MajorSixth]:=TRUE; END; REQUIRE ImperfInit INITIALIZATION; BOOLEAN ARRAY Dissonance[0:12]; SIMPLE PROCEDURE DisInit; BEGIN Dissonance[MinorSecond]:=TRUE; Dissonance[MajorSecond]:=TRUE; Dissonance[MinorSeventh]:=TRUE; Dissonance[MajorSeventh]:=TRUE; Dissonance[Fourth]:=TRUE; Dissonance[Tritone]:=TRUE; END; REQUIRE DisInit INITIALIZATION;
To ascertain whether a given interval is a dissonance, we merely use that interval as an index into the Dissonance array. The other arrays are used similarly. Under certain circumstances, the fourth is considered an imperfect consonance, but we will leave that complication until later. (In SAIL all Boolean variables are initialized to False , so we need only set the ones that should be True . These assignments take place at initialization time before the main program is executed).
We must also define the basic scales used in species counterpoint. The representation chosen here
takes advantage of the fact that the scales are only an octave in extent, so octave transpositions can
be removed before a decision is made as to whether a given pitch is in a mode. However, we do
have to take into account transpositions by some interval other than an octave. In the definition
given here, we assume that the tranposition (if any) has also been removed Ð from this code's viewpoint every mode starts on pitch 0 and has only 12 pitches to choose from. Any pitch that is in the mode has a value of 1 in the corresponding mode array. A value of 0 means that the given pitch is illegal (or at least a chromatic alteration) in that mode.
DEFINE Aeolian=1,Dorian=2,Phrygian=3,Lydian=4,Mixolydian=5,Ionian=6,Locrian=7; PRESET_WITH 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1; INTEGER ARRAY _Ionian[0:11]; PRESET_WITH 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0; INTEGER ARRAY _Dorian[0:11]; PRESET_WITH 1, 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0; INTEGER ARRAY _Phrygian[0:11]; PRESET_WITH 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1; INTEGER ARRAY _Lydian[0:11]; PRESET_WITH 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 1, 0; INTEGER ARRAY _Mixolydian[0:11]; PRESET_WITH 1, 0, 1, 1, 0, 1, 0, 1, 0, 0, 1, 0; INTEGER ARRAY _Aeolian[0:11]; PRESET_WITH 1, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0; INTEGER ARRAY _Locrian[0:11]; BOOLEAN PROCEDURE InMode(INTEGER Pitch,Mode); BEGIN INTEGER Pit; Pit:=Pitch MOD 12; ! remove any lingering octave information; CASE Mode OF BEGIN [Ionian] RETURN(_Ionian[Pit]); [Aeolian] RETURN(_Aeolian[Pit]); [Dorian] RETURN(_Dorian[Pit]); [Phrygian] RETURN(_Phrygian[Pit]); [Lydian] RETURN(_Lydian[Pit]); [Mixolydian] RETURN(_Mixolydian[Pit]); [Locrian] RETURN(_Locrian[Pit]) END; END;
Although defined here, the Ionian and Locrian modes have not actually been used during testing and may contain unnoticed bugs.
Certain intervals are not allowed melodically because they are considered hard to sing:
BOOLEAN ARRAY BadMelodyInterval[0:12]; SIMPLE PROCEDURE BadInit; BEGIN BadMelodyInterval[Tritone]:=TRUE; BadMelodyInterval[MajorSixth]:=TRUE; BadMelodyInterval[MinorSeventh]:=TRUE; BadMelodyInterval[MajorSeventh]:=TRUE; END; REQUIRE BadInit INITIALIZATION;
Other "bad" intervals such as the augmented second can be avoided by forbidding chromatic
alterations within the mode. The latter are necessary unfortunately at cadences, so more elaborate
checks will be made later. We must also avoid a leap greater than an octave and a leap of a minor
sixth down. The leap of a minor sixth up is acceptable.
BOOLEAN PROCEDURE BadMelody(INTEGER Intv); RETURN( ABS(Intv)>Octave OR ! more than an octave either way; BadMelodyInterval[ABS(Intv)] OR ! tritone or major sixth or seventh; Intv=-MinorSixth); ! downward minor sixth;
We will often need to distinguish a "skip" (or leap) from a "step". These are not explicitly defined in my version of Fux, but it is clear that a skip is any interval greater than a major second, and that a step is an interval that is not a skip and also not a unison.
BOOLEAN PROCEDURE ASkip(INTEGER Interval); RETURN(ABS(Interval)>MajorSecond); BOOLEAN PROCEDURE AStep(INTEGER Interval); RETURN((ABS(Interval)=MinorSecond) OR (ABS(Interval)=MajorSecond));
The intervals are further divided up into the groups second, third, sixth, seventh, and so on. We therefore define the obvious procedures:
BOOLEAN PROCEDURE AThird(INTEGER Interval); RETURN((Interval=minorThird) OR (Interval=MajorThird)); BOOLEAN PROCEDURE ASeventh(INTEGER Interval); RETURN((Interval=minorSeventh) OR (Interval=MajorSeventh)); BOOLEAN PROCEDURE AnOctave(INTEGER interval); RETURN((Interval <> Unison) AND ((ABS(Interval) MOD 12)=0)); BOOLEAN PROCEDURE ATenth(INTEGER Interval); RETURN((ABS(Interval)>14) AND AThird((ABS(Interval) MOD 12)));
The last procedure (ATenth) is used to implement only one rather obscure rule.
Now we define the four kinds of "motion": direct, oblique, contrary, and no motion:
DEFINE DirectMotion=1, ContraryMotion=2, ObliqueMotion=3, NoMotion=4; INTEGER PROCEDURE MotionType(INTEGER Pitch1,Pitch2,Pitch3,Pitch4); COMMENT Pitch1 and Pitch2 are from Voice 1, Pitch3 and Pitch4 from voice 2. That is the passage in question is: Voice1: Pitch1 Pitch2 Voice2: Pitch3 Pitch4 This procedure decides which kind of motion these pitches indicate. Either voice may go up (Pitch2>Pitch1), down (Pitch2<Pitch1), or hold its pitch (Pitch1=Pitch2). ; IF (Pitch1=Pitch2 OR Pitch3=Pitch4) ! at least one voice does not move; THEN IF (Pitch1=Pitch2 AND Pitch3=Pitch4) ! neither voice moves; THEN RETURN(NoMotion) ELSE RETURN(ObliqueMotion) ELSE IF ((Pitch2-Pitch1)*(Pitch4-Pitch3)>0) ! either both >0 or both <0; THEN RETURN(DirectMotion) ELSE RETURN(ContraryMotion); direct contrary oblique no motion motion motion motion
Two other procedures will be useful later when we apply rules involving direct motion to a perfect consonance and consecutive leaps:
BOOLEAN PROCEDURE DirectMotionToPerfectConsonance( INTEGER Pitch1,Pitch2,Pitch3,Pitch4); RETURN(PerfectConsonance[ABS(Pitch4-Pitch2) MOD 12] AND MotionType(Pitch1,Pitch2,Pitch3,Pitch4)=DirectMotion); BOOLEAN PROCEDURE ConsecutiveSkipsInSameDirection( INTEGER Pitch1,Pitch2,Pitch3); RETURN( (((Pitch1>Pitch2) AND (Pitch2>Pitch3)) OR ((Pitch1<Pitch2) AND (Pitch2<Pitch3))) AND ASkip(Pitch2-Pitch1) AND ! first interval is a skip; Askip(Pitch3-Pitch2)); ! so is second;
The extreme high and low pitches in a melody define its overall range. Fux emphasizes on many occasions that this range should be constrained to fit human vocal ranges.
DEFINE HighestSemitone=72; ! this is a high C (1025 Hz); DEFINE LowestSemitone=24; ! this is a low C (65 Hz); ! these are liberal ranges; BOOLEAN PROCEDURE OutOfRange(INTEGER Pitch); RETURN((Pitch>HighestSemitone) OR (Pitch<LowestSemitone)); BOOLEAN PROCEDURE ExtremeRange(INTEGER Pitch); RETURN(Pitch>(HighestSemitone-3) OR Pitch<(LowestSemitone+3));
These procedures give a rather liberal range check. We also need code to find the overall melody range so that it normally does not exceed an octave and a fifth. This code, however, depends on the representation we choose for voice data.
For the purposes of this description we will use a simple representation which uses parallel arrays for onset time, duration, and pitch. First we have integers for the transposition (BasePitch), the current mode (Mode), and the length of the cantus firmus in eighth notes (TotalTime).
DEFINE MostNotes=128; ! most notes any one voice can have; DEFINE MostVoices=8; INTEGER BasePitch,Mode,TotalTime;
We arbitrarily limit the longest melody to 128 notes and the number of voices to 8 (once again, these are purely for array allocation purposes and are not built into any of the counterpoint writing procedures). We need a place to hold the pitch, duration, and onset time data for each voice, including the cantus firmus.
INTEGER ARRAY CtrPt,Onset,Dur[1:MostNotes,0:MostVoices]; INTEGER ARRAY TotalNotes[0:MostVoices];
The cantus firmus is voice 0 in this scheme, but is not otherwise distinguished from the other voices.
Each note of voice V keeps its note information for note number N in these three parallel arrays.
OnSet[V,N] is the begin time in eighth notes of the note, Dur[V,N] is its duration (in eighth notes), and CtrPt[V,N] is the pitch of the note (in semitones above the low c). Next we need a variety of procedures to access the data in these arrays.
INTEGER PROCEDURE Us(INTEGER n,v); RETURN(Ctrpt[n,v]); ! returns pitch of note N in voice V; BOOLEAN PROCEDURE LastNote(INTEGER n,v); RETURN(n=TotalNotes[v]); ! returns true if N is the last note of V; BOOLEAN PROCEDURE FirstNote(INTEGER n,v); RETURN(n=1); ! returns true if N is the first note; BOOLEAN PROCEDURE NextToLastNote(INTEGER n,v); RETURN(n=(TotalNotes[v]-1)); ! returns true if N is the next to last note; ! these are used mainly for cadential formulas; PROCEDURE SetUs(INTEGER n,p,v); Ctrpt[n,v]:=p; ! sets the pitch of note N in voice V to P;
We also need a procedure that returns the total range of the voice V at the note CN (counting from the beginning of the melody), given the current pitch CP .
INTEGER PROCEDURE TotalRange(INTEGER Cn,Cp,v); BEGIN ! return total range of melody so far (including CP); INTEGER Minp,Maxp,i; MinP:=Cp; Maxp:=Cp; FOR i:=1 STEP 1 UNTIL Cn-1 DO BEGIN MinP:=MinP MIN Us(i,v); MaxP:=MaxP MAX Us(i,v); END; RETURN(MaxP-MinP); END;
We present most of the data analysis procedures in this form. In words, we have a proposed pitch (CP) in a voice (V) at the point (CN) in that voice, and need a decision about the acceptability of that pitch.
Since we assume for simplicity's sake that the cantus firmus is always in whole notes, it is easy to write a procedure which returns the cantus firmus note at any given eighth note beat N .
INTEGER PROCEDURE Cantus(INTEGER n,v); RETURN(CtrPt[((Onset[n,v] DIV 8)+1),0]);
A similar procedure is needed to return the current pitch in some other voice (we could use this procedure for the cantus firmus also because it is just voice 0, but as long as we are assuming the cantus moves in whole notes we can handle it separately to save some time):
INTEGER PROCEDURE VIndex(INTEGER Time,VNum); BEGIN INTEGER i; FOR i:=1 STEP 1 UNTIL TotalNotes[VNum] DO IF Onset[i,vNum] <= Time AND Onset[i,vNum]+Dur[i,vNum]>Time THEN DONE; RETURN(i); END; INTEGER PROCEDURE Other(INTEGER Cn,v,v1); RETURN(Ctrpt[VIndex(Onset[Cn,v],v1),v1]); INTEGER PROCEDURE Bass(INTEGER Cn,v); BEGIN INTEGER j,LowestPitch; LowestPitch:=Cantus(Cn,v); FOR j:=1 STEP 1 UNTIL v-1 DO LowestPitch:=LowestPitch MIN Other(Cn,v,j); RETURN(LowestPitch); END;
This is not an optimal search procedure, but normally there are not too many notes in a melody. We can therefore search laboriously from the beginning for the note in voice V1 that sounds at the same time as the note CN in voice V.
We must also define various rhythmic values:
DEFINE WholeNote= 8, HalfNote= 4, DottedHalfNote= 6, QuarterNote= 2, EighthNote= 1;
A measure contains 8 eighth notes (triple time is not supported).
INTEGER PROCEDURE Beat8(INTEGER n); RETURN(n MOD 8); ! 0=first beat,7=last;
A downbeat occurs on the first (0-th) beat of the measure.
BOOLEAN PROCEDURE DownBeat(INTEGER n,v); RETURN(Beat8(Onset[n,v])=0);
An upbeat can be considered to be any beat that is not a downbeat. There are several cases where beat 4 (the second half note beat) is a downbeat, but these will be handled separately.
BOOLEAN PROCEDURE UpBeat(INTEGER n,v); RETURN(NOT DownBeat(n,v));
Fux mentions that it is desirable to maintain "variety" in the melodies. This variety seems to entail a mix of melodic intervals coupled with an avoidance of too many repetitions of any one pitch. To check the latter condition we define a procedure which looks for repeated pitches within a melody:
INTEGER PROCEDURE PitchRepeats(INTEGER Cn,Cp,V); BEGIN INTEGER i,k; i:=0; FOR k:=1 STEP 1 UNTIL Cn-1 DO IF Us(k,v)=Cp THEN i:=i+1; RETURN(i); END;
We also need to encourage the melody to contain a nice mixture of intervals (in third and fifth species). We keep track of how many times a given interval type has been used so far, then check to see if others have been used nearly as many times. The interval type ignores distinctions of major and minor, but does not ignore direction (a rising minor second is considered the same interval type as a rising major second, but not the same as a descending minor second).
DEFINE One=0, Two=2, Three=3, Four=4, Five=5, Six=6, Eight=8; INTEGER PROCEDURE Size(INTEGER MelInt); BEGIN INTEGER ActInt,IntTyp; ActInt:=ABS(MelInt); CASE ActInt OF BEGIN [unison] IntTyp:=One; [minorSecond][majorSecond] IntTyp:=Two; [minorThird][majorThird] IntTyp:=Three; [fourth] IntTyp:=Four; [fifth] IntTYp:=Five; [minorsixth] IntTyp:=Six; [octave] IntTyp:=Eight; ELSE PRINT("illegal melodic interval: ",MelInt) END; RETURN(IF MelInt>0 THEN IntTyp ELSE -IntTyp); END; BOOLEAN PROCEDURE TooMuchOfInterval(INTEGER Cn,Cp,v); BEGIN INTEGER ARRAY Ints[-8:8]; INTEGER i,k,MinL; ARRCLR(Ints); FOR i:=2 STEP 1 UNTIL Cn-1 DO BEGIN k:=Size(CtrPt[i,v]-CtrPt[i-1,v]); Ints[k]:=Ints[k]+1; END; k:=Size(Cp-Ctrpt[Cn-1,v]); MinL:=-8; FOR i:=-7 STEP 1 UNTIL 8 DO IF i <> k AND Ints[i]>Ints[MinL] THEN MinL:=i; RETURN(Ints[k]>(Ints[MinL]+6)); END;
Next we must define when a dissonance is legal in each species, and also when chromatic (altered) notes occur. The rules governing the cadential formulas are pretty clear, but the rules governing ficta seem to consist mostly of handwaves and exceptions. We first define dissonance handling in each species:
BOOLEAN PROCEDURE ADissonance(INTEGER Interval,Cn,Cp,v,Species); IF Species=1 OR Dur[Cn,v]=wholenote THEN RETURN(Dissonance[Interval])
In first species no dissonances are allowed, so we simply return true in every case in which Interval is a dissonance. Similarly, we do not allow wholenotes to form a dissonance (in fifth species).
ELSE IF Species=2 THEN IF (DownBeat(Cn,v) OR (NOT (Astep(Cp-Us(Cn-1,v))))) THEN RETURN(Dissonance[Interval]) ELSE RETURN(FALSE) passing tone
In second species, a dissonance is allowed only as a passing tone on an upbeat. Since we don't yet know what the next note will be, we must put off the rest of the dissonance check until later.
ELSE IF Species=3 THEN BEGIN INTEGER MelInt; IF Beat8(OnSet[Cn,v])=0 OR FirstNote(Cn,v) OR LastNote(Cn,v) THEN RETURN(Dissonance[Interval]); MelInt:=Cp-Us(Cn-1,v); IF (NOT AStep(MelInt)) THEN RETURN(Dissonance[Interval]); COMMENT 0 cannot be dissonant (downbeat) 1 can be if passing either way, but must be approached by step. 2 can be if passing 2 to 4 (both latter cons) 3 can be if passing but must be approached and left by step ; RETURN(FALSE); END cambiata
In third species, the downbeat cannot be a dissonance, but any of the other beats can if they are passing tones or cambiatas. As with second species, the dissonance resolution is checked later.
ELSE IF Species=4 THEN BEGIN INTEGER MelInt; IF UpBeat(Cn,v) OR FirstNote(Cn,v) OR LastNote(Cn,v) THEN RETURN(Dissonance[Interval]); MelInt:=Cp-Us(Cn-1,v); IF MelInt <> 0 THEN RETURN(Dissonance[Interval]); RETURN(FALSE); ! i.e. unison to downbeat is ok, but needs check later; END suspension
In fourth species, a dissonance is legal as a suspension. The suspension is handled here as a note repeated from the upbeat to the downbeat (consider it tied across the bar), so the melodic interval is obviously a unison.
ELSE IF Species=5 THEN BEGIN IF Beat8(Onset[Cn,v])=0 ! first note; THEN ! so can be tied from previous to diss now; IF Cp=Us(Cn-1,v) ! is a unison (assume tie); THEN RETURN(FALSE) ! so dissonance is ok here (check resolution later); ELSE RETURN(Dissonance[Interval]) ELSE ! not first beat; IF NOT AStep(Cp-Us(Cn-1,v)) ! cannot jump to dissonance; THEN RETURN(Dissonance[Interval]); RETURN(FALSE); END;
Finally, in fifth species we have passing note dissonances, cambiatas, and suspensions. Therefore, if
the downbeat (beat 0) of the measure is dissonant, the note must have been tied across the bar (the
melodic interval must be a unison). On any other beat the dissonance must be approached by step.
We will check later for its proper resolution.
Lastly, in multi-part counterpoint we often need to check for various kinds of pitch doublings:
BOOLEAN PROCEDURE Doubled(INTEGER Pitch,Cn,v); BEGIN INTEGER vNum; FOR Vnum:=0 STEP 1 UNTIL v-1 DO IF (Other(Cn,v,VNum) MOD 12)=Pitch THEN RETURN(TRUE); RETURN(FALSE); END;
The basic rules of species counterpoint describe how melodies are formed and combined. In most
cases any given rule can be broken if other factors necessitate it. Fux repeatedly presents the rules as
guidelines, not absolutes. In our implementation we define the relative importance of the rules by
assigning each rule a penalty. The higher the penalty, the worse it is to break the associated rule.
The relations between these penalties determine what kind of solutions the program will find.
Our penalty types and the associated penalty values are:
DEFINE Infinity="2^18"; DEFINE Bad!=100; DEFINE RealBad!=200; DEFINE UnisonPenalty =bad!, DirectToFifthPenalty =Realbad!, DirectToOctavePenalty =RealBad!, ParallelFifthPenalty =infinity, ParallelUnisonPenalty =infinity, EndOnPerfectPenalty =infinity, NoLeadingTonePenalty =infinity, DissonancePenalty =infinity, OutOfRangePenalty =realBad!, OutOfModePenalty =infinity, TwoSkipsPenalty =1, DirectMotionPenalty =1, PerfectConsonancePenalty =2, CompoundPenalty =1, TenthToOctavePenalty =8, SkipTo8vePenalty =8, SkipFromUnisonPenalty =4, SkipPrecededBySameDirectionPenalty =1, FifthPrecededBySameDirectionPenalty =3, SixthPrecededBySameDirectionPenalty =8, SkipFollowedBySameDirectionPenalty =3, FifthFollowedBySameDirectionPenalty =8, SixthFollowedBySameDirectionPenalty =34, TwoSkipsNotInTriadPenalty =3, BadMelodyPenalty =infinity, ExtremeRangePenalty =5, LydianCadentialTritonePenalty =13, UpperNeighborPenalty =1, LowerNeighborPenalty =1, OverTwelfthPenalty =infinity, OverOctavePenalty =bad!, SixthLeapPenalty =2, OctaveLeapPenalty =5, BadCadencePenalty =infinity, DirectPerfectOnDownbeatPenalty =infinity, RepetitionOnUpbeatPenalty =bad!, DissonanceNotFillingThirdPenalty =infinity, UnisonDownbeatPenalty =3, TwoRepeatedNotesPenalty =2, ThreeRepeatedNotesPenalty =4, FourRepeatedNotesPenalty =7, LeapAtCadencePenalty =13, NotaCambiataPenalty =infinity, NotBestCadencePenalty =8, UnisonOnBeat4Penalty =3, NotaLigaturePenalty =21, UnresolvedLigaturePenalty =infinity, NoTimeForaLigaturePenalty =infinity, EighthJumpPenalty =bad!, HalfUntiedPenalty =13, UnisonUpbeatPenalty =21, MelodicBoredomPenalty =1, SkipToDownBeatPenalty =1, ThreeSkipsPenalty =3, DownBeatUnisonPenalty =bad!, VerticalTritonePenalty =2, MelodicTritonePenalty =8, AscendingSixthPenalty =1, RepeatedPitchPenalty =1, NotContraryToOthersPenalty =1, NotTriadPenalty =34, InnerVoicesInDirectToPerfectPenalty =21, InnerVoicesInDirectToTritonePenalty =13, SixFiveChordPenalty =infinity, UnpreparedSixFivePenalty =bad!, UnresolvedSixFivePenalty =bad!, AugmentedIntervalPenalty =infinity, ThirdDoubledPenalty =5, DoubledLeadingTonePenalty =infinity, DoubledSixthPenalty =5, DoubledFifthPenalty =3, TripledBassPenalty =3, UpperVoicesTooFarApartPenalty =1, UnresolvedLeadingTonePenalty =infinity, AllVoicesSkipPenalty =8, DirectToTritonePenalty =bad!, CrossBelowBassPenalty =infinity;
These penalty values have come partly from Fux's comments (the more important rules get a higher penalty), and partly from experience running the program. Obviously not all the penalties apply to every species of counterpoint. The penalties are all positive (that is, there are no rewards), because we want to be able to abandon a line as soon as its penalty gets too high. This implies that the penalty function should never descend. Even the smaller penalties have a profound effect on the outcome of the counterpoint search. The solver is consistently "bent" in the direction determined by the penalties causing the music to reflect the slightest changes in that direction.
We need a procedure which examines the current note of the current voice and returns a measure of its goodness (the lower the penalty, the better the note).
FORWARD INTEGER PROCEDURE SpecialSpeciesCheck( INTEGER Cn,Cp,v,Other0,Other1,Other2,NumParts, Species,MelInt,Interval,ActInt,LastIntClass,Pitch,LastMelInt); FORWARD INTEGER PROCEDURE OtherVoiceCheck( INTEGER Cn,Cp,v,v1,Species);
We define the special attributes of each species later. These are checked by the SpecialSpeciesCheck procedure (declared here for convenience, but defined later to make the presentation more clear).
INTEGER PROCEDURE Check(INTEGER Cn,Cp,v,NumParts,Species); BEGIN ! return penalty associated with pitch CP in voice V at note CN; INTEGER Val ! accumulated penalty for CP; i,k, ! temp counters; Interval, ! current pitch to other voice current pitch; IntClass, ! octaves and direction removed from Interval; Pitch, ! CP pitch class; LastIntClass, ! previous such interval; MelInt, ! melodic interval in current voice; LastMelInt, ! last melodic interval in voice; Other0, ! current other voice pitch; Other1, ! previous other voice pitch; Other2, ! pitch 2 back in other voice; Cross; ! voice cross counter; BOOLEAN SameDir; ! did last 2 melodic intervals go same direction; IF v=1 THEN BEGIN Other0:=Cantus(Cn,v); ! current pitch in cantus; Other1:=Cantus(Cn-1,v); IF Cn>2 THEN Other2:=Cantus(Cn-2,v); END ELSE BEGIN Other0:=Bass(Cn,v); ! check current against bass line; Other1:=Bass(Cn-1,v); IF Cn>2 THEN Other2:=Bass(Cn-2,v); END; Val:=0 ! accumulated penalty; Interval:=Cp-Other0; ! vertical interval between current pitch and cantus; IntClass:=ABS(Interval) MOD 12; ! remove direction and octave info; MelInt:=(Cp-Us(Cn-1,v)); Pitch:=CP MOD 12;
The rules are applied here in an order which hopes to reduce redundant checks of various kinds. At the start we apply rules that apply to every pitch in the melodies.
! melody must stay in range; IF OutOfRange(Cp+BasePitch) THEN Val:=Val+OutOfRangePenalty; ! extremes of range are also bad (to be avoided); IF ExtremeRange(Cp+BasePitch) THEN Val:=Val+ExtremeRangePenalty; ! Chromatically altered notes are accepted only at the ! cadence. Other alterations (such as ficta) will be handled later); IF NOT NextToLastNote(Cn,v) THEN BEGIN IF (Species <> 2) THEN IF NOT InMode(Pitch,Mode) THEN Val:=Val+OutOfModePenalty ELSE ELSE IF (Cn <> TotalNotes[v]-2) OR (Mode <> Aeolian) OR (Cp <= Other0) OR (Intclass <> Fifth) THEN IF NOT InMode(Pitch,Mode) THEN Val:=Val+OutOfModePenalty; END ELSE BEGIN BOOLEAN WeHaveARealLeadingTone; WeHaveARealLeadingTone:=(Pitch=11) OR ((Pitch=10) AND (Mode=Phrygian)); Aeolian cadence in second species
The leading tone cannot be doubled, nor should the raised leading tone be preceded by its unraised form. The next to last chord of the counterpoint must have a leading tone somewhere.
IF WeHaveARealLeadingTone THEN IF Doubled(Pitch,Cn,v) ! leading tone doubled; THEN Val:=Val+DoubledLeadingTonePenalty ELSE ELSE IF (Pitch=10) ! unraised leading tone, but need raised form; THEN Val:=Val+BadCadencePenalty ELSE ! are we last Ð does anyone have one; IF NOT InMode(Pitch,Mode) ! maybe not the leading tone at all; THEN Val:=Val+OutOfModePenalty ELSE IF v=Numparts ! last voice Ð check that someone has it; THEN IF NOT Doubled(11,Cn,v) AND NOT Doubled(10,Cn,v) THEN Val:=Val+NoLeadingTonePenalty; END; IF Val >= infinity THEN RETURN(val); IF Cn>2 THEN BEGIN LastMelInt:=(Us(Cn-1,v)-Us(Cn-2,v)); SameDir:=((MelInt*LastMelInt) >= 0); END; IF Cn>1 THEN LastIntClass:=(ABS(Us(Cn-1,v)-Other1) MOD 12); ! dissonance must be handled correctly. The procedure ! ADISSONANCE returns true if a CP forms a dissonance that is ! not correctly prepared. Other rules later check that it is ! resolved correctly; IF ADissonance(IntClass,Cn,Cp,v,Species) THEN Val:=Val+DissonancePenalty; IF Val >= infinity THEN RETURN(val); Val:=Val+SpecialSpeciesCheck(Cn,Cp,v,Other0,Other1,Other2,NumParts, Species,MelInt,Interval,intClass, LastIntClass,Pitch,LastMelInt); IF v>1 THEN Val:=Val+OtherVoiceCheck(Cn,Cp,v,NumParts,species);
This procedure call looks at the current state from the point of view of the current species. There are a number of rules unique to each species.
IF FirstNote(Cn,v) THEN RETURN(Val); ! no further rules apply to first note; IF Val >= infinity THEN RETURN(val);
The "fundamental rule" defines which intervallic relationships are acceptable between two melodies.
Fux presents this as four rules, but as Martini remarks, these four are actually reducible to one Ð voices cannot move by direct motion to a perfect consonance. In multi-voice counterpoint, this rule can be broken, especially at cadences
! direct motion to perfect consonances considered harmful; IF NOT LastNote(Cn,v) OR NumParts=1 THEN IF DirectMotionToPerfectConsonance(Us(Cn-1,v),Cp,Other1,Other0) THEN ! we do have direct motion to perfect; IF IntClass=unison ! reached an octave or unison; THEN Val:=Val+DirectToOctavePenalty ELSE Val:=Val+DirectToFifthPenalty;
The expression "NumParts <> 1" checks that we are indeed in a multi-voice situation. In two part counterpoint we do not allow direct motion to a perfect consonance at all.
! check for more blatant examples of the same error; IF IntClass=fifth AND LastIntClass=Fifth THEN Val:=Val+ParallelFifthPenalty; IF IntClass=unison AND LastIntClass=unison THEN Val:=Val+ParallelUnisonPenalty; ! voices parallel in octave or fifths; ! which Fux seems to consider to be worse; ! than just direct motion to these; IF Val >= infinity THEN RETURN(val); ! certain melodic intervals are disallowed; IF BadMelody(MelInt) THEN Val:=Val+BadMelodyPenalty; IF Val >= infinity THEN RETURN(val);
This rule (the meodic interval check) is actually built into the search procedures given below so its inclusion here is mostly for completeness.
! must end on unison or octave in two parts, fifth and major third ! allowed in 3 and 4 part writing; IF (LastNote(Cn,v)) AND ! it is the last note; (IntClass <> unison) ! must be octave or unison here (2 parts); THEN IF NumParts=1 OR Interval<0 ! two part writing or bass; THEN Val:=Val+EndOnPerfectPenalty ELSE ! NumParts>1 so we must have 3 or more part writing; IF Intclass <> fifth AND Intclass <> MajorThird THEN Val:=Val+EndOnPerfectPenalty;
If we end on a major third, this should of course be a major third above the cantus, not below it (similarly for the fifth).
! penalize direct motion any kind (contrary motion is better); IF MotionType(Us(Cn-1,v),Cp,Other1,Other0)=DirectMotion THEN BEGIN Val:=Val+DirectMotionPenalty; IF IntClass=Tritone THEN Val:=Val+DirectToFifthPenalty; END; ! penalize compound intervals (close position is favored); IF ABS(Interval)>Octave ! avoid compound intervals wherever possible; THEN Val:=Val+CompoundPenalty;
Now we present a raft of rules involving skips. For example, it is considered bad to have too many consecutive skips in the same directions.
! penalize consecutive skips in the same direction; IF Cn>2 AND ConsecutiveSkipsInSameDirection(Us(Cn-2,v),Us(Cn-1,v),Cp) THEN BEGIN INTEGER totalJump; Val:=Val+TwoSkipsPenalty; totalJump:=ABS(Cp-Us(Cn-2,v)); ! do not let these skips traverse more than an octave, nor a seventh; IF ((totalJump>MajorSixth) AND (totalJump < Octave)) THEN Val:=Val+TwoSkipsNotInTriadPenalty; END; ! penalize a skip to an octave; IF ((IntClass=unison) AND (ASkip(MelInt) OR ASkip(Other0-Other1))) THEN Val:=Val+SkipTo8vePenalty; ! do not skip from a unison (not a very important rule); IF (Other1=Us(Cn-1,v)) AND ASkip(MelInt) THEN Val:=Val+SkipFromUnisonPenalty; ! penalize skips followed or preceded by motion in same direction; IF Cn>2 AND ASkip(MelInt) AND SameDir THEN ! skip preceded by same direction; BEGIN ! especially penalize fifths, sixths, and octaves of this sort; IF ABS(MelInt)<Fifth ! skip of third or fourth; THEN Val:=Val+SkipPrecededBySameDirectionPenalty ELSE IF ABS(MelInt)=Fifth OR ABS(MelInt)=Octave THEN Val:=Val+FifthPrecededBySameDirectionPenalty ELSE Val:=Val+SixthPrecededBySameDirectionPenalty; END; IF Cn>2 AND ASkip(LastMelInt) AND SameDir THEN BEGIN IF ABS(LastMelInt)<Fifth ! skip of third or fourth; THEN Val:=Val+SkipFollowedBySameDirectionPenalty ELSE IF ABS(LastMelInt)=Fifth OR ABS(LastMelInt)=Octave THEN Val:=Val+FifthFollowedBySameDirectionPenalty ELSE Val:=Val+SixthFollowedBySameDirectionPenalty; END; ! too many skips in a row Ð favor a mix of steps and skips; IF Cn>4 AND ASkip(MelInt) AND ASkip(LastMelInt) AND ASkip(Us(Cn-2,v)-Us(Cn-3,v))
! avoid tritones melodically; IF Cn>4 AND (ABS(Cp-Us(Cn-2,v))=Tritone OR ABS(Cp-Us(Cn-3,v))=Tritone OR ABS(Cp-Us(Cn-4,v))=Tritone) THEN Val:=Val+MelodicTritonePenalty;
As can be seen from the solutions given at the end of the paper, tritone handling is far from perfect.
Perhaps the MelodicTritone penalty should be higher, and more elaborate checks should be added for tritones over larger melodic distances.
! do not allow movement from a tenth to an octave by contrary motion; IF Species <> 5 AND NumParts=1 THEN IF ATenth(Other1-Us(Cn-1,v)) AND AnOctave(Interval) THEN Val:=Val+TenthToOctavePenalty;
Fux admits this rule (disallowing motion from a tenth to an octave) is without obvious musical justification, but states that the rule merely follows the practice of the great composers.
! more range checks Ð did we go over an octave recently; IF Cn>2 AND ABS(Cp-Us(Cn-2,v))>Octave THEN Val:=Val+OverOctavePenalty; ! same for a twelfth; IF Cn>3 AND TotalRange(Cn,Cp,v)>(Octave+Fifth) THEN Val:=Val+OverTwelfthPenalty; IF Val >= infinity THEN RETURN(val); ! slightly penalize repeated notes; IF (Cn>3) AND (Cp=Us(Cn-2,v)) AND (Us(Cn-1,v)=Us(Cn-3,v)) THEN Val:=Val+TwoRepeatedNotesPenalty; IF (Cn>5) AND (Cp=Us(Cn-3,v)) AND (Us(Cn-1,v)=Us(Cn-4,v)) AND (Us(Cn-2,v)=Us(Cn-5,v)) THEN Val:=Val+ThreeRepeatedNotesPenalty; IF (Cn>6) AND (Cp=Us(Cn-4,v)) AND (Us(Cn-1,v)=Us(Cn-5,v)) AND (Us(Cn-2,v)=Us(Cn-6,v)) THEN Val:=Val+ThreeRepeatedNotesPenalty-1; IF (Cn>7) AND (Cp=Us(Cn-4,v)) AND (Us(Cn-1,v)=Us(Cn-5,v)) AND (Us(Cn-2,v)=Us(Cn-6,v)) AND (Us(Cn-3,v)=Us(Cn-7,v)) THEN Val:=Val+FourRepeatedNotesPenalty; IF (Cn>8) AND (Cp=Us(Cn-5,v)) AND (Us(Cn-1,v)=Us(Cn-6,v)) AND (Us(Cn-2,v)=Us(Cn-7,v)) AND (Us(Cn-3,v)=Us(Cn-8,v)) THEN Val:=Val+FourRepeatedNotesPenalty;
These rules do not penalize sequences, nor do they catch widely separated pattern repetitions. We may add such checks to a later version of the program.
IF LastNote(Cn,v) THEN BEGIN INTEGER LastPitch; LastPitch:=(Us(Cn-1,v) MOD 12); IF ((LastPitch=11) OR ((LastPitch=10) AND (Mode=Phrygian))) AND (Pitch <> 0) THEN Val:=Val+UnresolvedLeadingTonePenalty; END;
Without this rule forcing the leading tone to go to the tonic, in multi-voice situations the program sometimes skips from the leading tone to some other note than the tonic to avoid direct motion to an octave.
IF Val >= infinity THEN RETURN(val); ! an imperfect consonance is better than a perfect consonance; IF PerfectConsonance[IntClass] THEN Val:=Val+PerfectConsonancePenalty; ! no unisons allowed within counterpoint unless more than 2 parts; IF NumParts=1 AND (Interval=unison) THEN Val:=Val+UnisonPenalty; IF Val >= infinity THEN RETURN(val); ! seek variety by avoiding pitch repetitions; Val:=Val+PitchRepeats(Cn,Cp,v)/2;
Without a few rules governing melodic practices, the counterpoint writer is as happy to bounce around by octaves as by steps. This can lead to some rather funny looking melodies.
! penalize octave leaps a little; IF AnOctave(MelInt) THEN Val:=Val+OctaveLeapPenalty; ! similarly for minor sixth leaps; IF MelInt=MinorSixth THEN Val:=Val+SixthLeapPenalty; ! penalize upper neighbor notes slightly (also lower neighbors); IF Cn>2 AND (MelInt<0) AND ! last interval was downward; AStep(MelInt) AND ! downward step in fact; (Cp=Us(Cn-2,v)) ! upper neighbor (this note repeated 2 back); THEN Val:=Val+UpperNeighborPenalty; IF Cn>2 AND (MelInt>0) AND ! last interval was upward; AStep(MelInt) AND ! upward step in fact; (Cp=Us(Cn-2,v) ! lower neighbor (this note repeated 2 back); THEN Val:=Val+LowerNeighborPenalty; ! do not allow normal leading tone to precede raised leading tone; ! also check here for augmented fifths and diminished fourths; IF (NOT InMode(Pitch,Mode)) AND ! must be raised leading tone; ((MelInt=minorsecond) OR ! and preceding was a normal leading tone; (MelInt=MinorSixth) OR ! preceded by augmented fifth; (MelInt=-MajorThird)) ! preceded by diminshed fourth; THEN Val:=Val+OutOfModePenalty; ! slightly frown upon leap back in the opposite direction; IF Cn>2 AND ASkip(MelInt) AND Askip(LastMelInt) AND NOT SameDir THEN BEGIN Val:=Val+(0 MAX ((ABS(MelInt)+ABS(LastMelInt))-8)); IF Cn>3 AND ASkip(Us(Cn-2,v)-Us(Cn-3,v)) THEN Val:=Val+ThreeSkipsPenalty; END;
This rule only penalizes a leap of more than a major third followed immediately by a leap in the opposite direction. Fux has an example that leaps up an octave, then immediately down an octave, so this is not a terrible flaw. More "common sense" checks of this sort are given below.
! try to approach cadential passages by step; IF NumParts=1 AND Cn >= TotalNotes[v]-4 AND (ABS(MelInt)>4) THEN Val:=Val+LeapAtCadencePenalty; ! check for entangled voices; Cross:=0; IF NumParts=1 THEN FOR k:=4 STEP 1 UNTIL Cn DO IF (Us(k,v)-Cantus(k,v))*(Us(k-1,v)-Cantus(k-1,v))<0 THEN Cross:=Cross+1; IF Cross>0 THEN Val:=Val+(0 MAX ((Cross-2)*3));
(Cross-2)*3 does not penalize the melody until there are three crossings.
! don't repeat note on upbeat; IF (Upbeat(Cn,v) AND (MelInt=Unison)) THEN Val:=Val+RepetitionOnUpbeatPenalty; ! avoid tritones near Lydian cadence; IF (Mode=Lydian) AND (Cn>TotalNotes[v]-4) AND (Pitch=6) THEN Val:=Val+LydianCadentialTritonePenalty; ! various miscellaneous checks. More elaborate dissonance resolution and ! cadential formula checks will be given under "Species definition"; IF Species <> 1 AND DownBeat(Cn,v) THEN BEGIN IF Species<4 THEN BEGIN IF MelInt=unison AND ! don't repeat upbeat note on downbeat; NOT LastNote(Cn,v) THEN Val:=Val+UnisonDownbeatPenalty; ! check for dissonance that doesn't fill a third as a passing tone; IF Dissonance[LastIntClass] AND ((NOT AStep(MelInt)) OR (NOT SameDir)) THEN Val:=Val+DissonanceNotFillingThirdPenalty; END; ! check for Direct 8ve or 5 where the intervening interval is less than a fourth; IF DirectMotionToPerfectConsonance(Us(Cn-2,v),Cp,Other2,Other0) AND (ABS(LastMelInt)<Fourth) THEN Val:=Val+DirectPerfectOnDownbeatPenalty; END; ! check for tritone with cantus or bass; IF IntClass=Tritone THEN Val:=Val+VerticalTritonePenalty; ! check for melodic interval variety; IF Cn>10 AND TooMuchOfInterval(Cn,Cp,v) THEN Val:=Val+MelodicBoredomPenalty; RETURN(Val); END;
We have not yet provided rules guiding the program toward good overall melodic shapes. It may be useful in this regard to treat a melody as a waveform and look for its low frequency components. We then presumably want the various melodies to have different such components or at least different initial phases.
There are differences between the species that can most easily be handled by providing a special set of rules for each species. This procedure is called in Check (given above). The two major special areas are cadential formulas and dissonance handling.
INTEGER PROCEDURE SpecialSpeciesCheck( INTEGER Cn,Cp,v,Other0,Other1,Other2,NumParts, Species,MelInt,Interval,ActInt,LastIntClass,Pitch,LastMelInt); BEGIN "SSC" INTEGER Val; IF Species=1 THEN RETURN(0); ! no special rules for 1st species; Val:=0; ! accumulated penalty;
In first species all vertical intervals must be consonances (this is handled by the procedure
ADissonance given above), the first and last intervals must be perfect consonances, imperfect consonance are better than perfect, and the next to last interval must be a major sixth if the cantus firmus is below or a minor third if the cantus firmus is above. All these rules are handled by Check, so we have no further rules to apply in first species.
In second species the interval at the downbeat must be consonant, but the upbeat can be dissonant if it fills in a third (a "passing tone"). The next to last measure has a fifth to a major sixth if the counterpoint is above, and a fifth to a minor third if it is below. The phrygian cadence is also special. Direct motion to a perfect consonance between successive down beats is accepted if the intervening interval is larger than a major third. ADissonance handles the passing tone check, so we need only define the cadences here.
IF Species=2 THEN BEGIN "2" IF NextToLastNote(Cn,v) AND (Pitch=11 OR Pitch=10) THEN ! we have the leading tone; IF ((Mode <> Phrygian) OR Interval >= 0) THEN ! phrygian below is special case; BEGIN IF LastIntClass <> Fifth THEN Val:=Val+BadCadencePenalty; END ELSE IF LastIntClass <> MinorSixth THEN Val:=Val+BadCadencePenalty; END "2" ELSE
Fourth species introduces suspensions and accented dissonances. The dissonance must be resolved by a step downward. Since there are more similarities between third and fifth species and second and fourth, we can save some code by combining third and fifth below.
BEGIN "3 4 5" ! 3rd, 4th, and 5th need more info; INTEGER k; IF Species=4 THEN BEGIN "4" IF DownBeat(Cn,v) AND MelInt <> Unison ! strongly encourage ligatures; THEN Val:=Val+NotaLigaturePenalty;
Fux says we should try to use a ligature wherever possible, so we penalize anything else.
IF Upbeat(Cn,v) AND Dissonance[LastIntClass] THEN BEGIN IF (MelInt <> -minorSecond AND MelInt <> -majorSecond) THEN Val:=Val+UnresolvedLigaturePenalty;
If the last interval was a downbeat and a dissonance, the dissonance must be resolved on this beat by a step downward.
IF ActInt=Unison AND (Interval<0 OR (ABS(Us(Cn-2,v)-Other2) MOD 12)=Unison) THEN Val:=Val+NoTimeForaLigaturePenalty;
The vertical interval cannot resolve to a unison or octave if the preceding downbeat was also a unison or octave (the ligature does not make direct motion to a unison or octave acceptable, but does enable one to move from one fifth to another). It is also bad to resolve to an octave if the cantus firmus is above the melody.
IF ActInt=Fifth OR actInt=Tritone THEN Val:=Val+NoTimeForALigaturePenalty; END;
And, finally, the resolution must be to a consonance (not a tritone for example).
END "4" ELSE
In third species in addition to the passing tone dissonance (which can occur on any beat except the
first), we must also accept cambiatas. The result is that any dissonance must be approached by step.
If the dissonance occurs on the second beat and is approached from above and is left by a third down, then it must be left by two steps up. In the latter case every interval must be consonant except the second (and the fourth if it is a passing tone). In addition, this particular dissonance (the cambiata) can only be a seventh if the cantus firmus is below, or a fourth if it is above (in two parts).
BEGIN "3 5" INTEGER Above,Cross; Above:=(Interval >= 0); ! added check to stop optimizer from changing 4th beat passing tones into ! repeated notes+skip; IF (Beat8(Onset[Cn,v])=6 OR Beat8(Onset[Cn,v])=7) AND Cp=Us(Cn-1,v) THEN Val:=Val+UnisonOnBeat4Penalty; ! skip to down beat seems not so great; IF Beat8(Onset[Cn,v])=0 THEN BEGIN IF ASkip(MelInt) THEN Val:=Val+SkipToDownbeatPenalty; IF Cn>2 AND (ActInt=unison OR ActInt=Fifth) THEN ! look for parallel 8ve or 5 on downbeat; BEGIN INTEGER i; IF Species=5 THEN FOR i:=Cn-1 STEP -1 UNTIL 1 DO BEGIN IF Beat8(OnSet[i,v])=0 THEN DONE; END ELSE i:=Cn-4; IF (ABS(Us(i,v)-Bass(i,v)) MOD 12)=ActInt THEN Val:=Val+DownbeatUnisonPenalty; END; END; ! check for cambiata not resolved correctly (on 4th beat); IF Beat8(Onset[Cn,v])=6 AND AThird(ABS(LastMelInt)) AND Dissonance[ABS(Us(Cn-2,v)-Other2) MOD 12] AND ( (MelInt<0) OR (ABS(MelInt) <> MajorSecond AND ABS(MelInt) <> MinorSecond)) THEN Val:=Val+NotaCambiataPenalty; IF Val >= infinity THEN RETURN(Val); IF Species=3 AND Cn>1 AND Dissonance[LastIntClass] THEN ! cambiata or passing tones?; BEGIN CASE Beat8(Onset[Cn,v]) OF BEGIN  IF (NOT AStep(MelInt)) OR ! (4 was diss or 3 was diss); (NOT AStep(LastMelInt)) OR ((MelInt*LastMelInt)<0) THEN Val:=Val+DissonancePenalty;  Val:=Val+DissonancePenalty; ! can't happen (1 can't be diss);  IF (NOT AStep(LastMelInt)) OR ! cambiata or passing tone are alike in this; (ABS(MelInt)>MajorThird) OR ! step if passing, third if cambiata; (MelInt=0) OR ! neither being a unison; ((LastMelInt*MelInt)<0) ! both continue same direction; THEN Val:=Val+DissonancePenalty ! so lose; ELSE ! now check more cases; IF (NOT AStep(MelInt)) ! passing tone ok (so ELSE clause ok); THEN BEGIN IF Above THEN BEGIN IF NOT ASeventh(LastIntClass) THEN Val:=Val+DissonancePenalty; END ELSE IF LastIntClass <> Fourth THEN Val:=Val+DissonancePenalty; END END; END; IF Species=5 THEN BEGIN "5" INTEGER LastDisInt; IF Cn>1 AND Beat8(Onset[Cn,v])=0 AND (Cp <> Us(Cn-1,v)) AND (Dur[Cn,v] <= Dur[Cn-1,v]) THEN Val:=Val+(NotaLigaturePenalty/3); IF Cn>3 AND Dur[Cn,v]=halfnote AND Beat8(Onset[Cn,v])=4 AND Dur[Cn-1,v]=quarternote AND Dur[Cn-2,v]=quarternote THEN Val:=Val+HalfUntiedPenalty; IF Dur[Cn,v]=EighthNote AND DownBeat(Cn,v) AND Dissonance[ActInt] THEN Val:=Val+DissonancePenalty; IF Val >= infinity THEN RETURN(Val); IF Cn>1 THEN LastDisInt:=ABS(Us(Cn-1,v)-Other1) MOD 12; IF Cn>1 AND Dissonance[LastDisInt] THEN ! cambiata or passing tones?; BEGIN "Diss" CASE Beat8(Onset[Cn-1,v]) OF BEGIN  IF LastDisInt=Fourth AND ! dissonance was a fourth; MelInt=Unison AND ! tied to current; (Other0-Other1)=unison AND ! and bass is also motionless; Beat8(Onset[Cn,v])=0 ! and tied across bar; THEN ! "consonant fourth"!; ELSE ! cambiata or passing tone or suspension; IF (NOT AStep(MelInt)) OR ! (4 was diss or 3 was diss); (NOT AStep(LastMelInt)) OR ((MelInt*LastMelInt)<0) OR (Dur[Cn-1,v]=eighthnote) OR (Dur[Cn-1,v]=quarterNote AND Dur[Cn-2,v]=halfNote) THEN Val:=Val+DissonancePenalty;  IF (NOT AStep(MelInt)) OR ! (4 was diss or 3 was diss); (NOT AStep(LastMelInt)) OR ((MelInt*LastMelInt)<0) THEN Val:=Val+DissonancePenalty;  BEGIN IF (Dur[Cn-2,v]=eighthnote) OR (Dur[Cn-2,v]<Dur[Cn-1,v]) THEN Val:=Val+NoTimeForALigaturePenalty; IF (MelInt <> -minorSecond AND MelInt <> -majorSecond) THEN Val:=Val+UnresolvedLigaturePenalty; IF ActInt=Fourth OR ActInt=Tritone THEN Val:=Val+NoTimeForALigaturePenalty; IF ActInt=Fifth AND Interval<0 THEN Val:=Val+NoTimeForALigaturePenalty; IF ActInt=0 AND (ABS(Us(Cn-2,v)-Other2) MOD 12)=0 THEN Val:=Val+NoTimeForaLigaturePenalty; IF LastMelInt <> unison THEN Val:=Val+DissonancePenalty; END;  IF (NOT AStep(LastMelInt)) OR ! cambiata or passing tone are alike in this; (ABS(MelInt)>MajorThird) OR ! step if passing, third if cambiata; (MelInt=0) OR ! neither being a unison; (Dur[Cn-1,v]=eighthnote) OR ((LastMelInt*MelInt)<0) ! both continue same direction; THEN Val:=Val+DissonancePenalty ! so lose; ELSE ! now check more cases; IF (NOT AStep(MelInt)) ! passing tone ok (so ELSE clause ok); THEN BEGIN IF Above THEN BEGIN IF NOT ASeventh(LastIntClass) THEN Val:=Val+DissonancePenalty; END ELSE IF LastIntClass <> Fourth THEN Val:=Val+DissonancePenalty; END END; END "Diss"; IF Cn>1 AND Dur[Cn-1,v]=eighthnote AND (NOT AStep(MelInt)) THEN Val:=Val+EighthJumpPenalty; IF Cn>1 AND Dur[Cn-1,v]=halfnote AND Beat8(Onset[Cn,v])=4 AND MelInt=Unison THEN Val:=Val+UnisonUpbeatPenalty; END "5"; END "3 5"; END "3 4 5"; RETURN(Val); END "SSC";
It is perhaps interesting that we do not need to add any special encouragement for cambiatas Ð they occur as a side effect of the rules (once the dissonance handling is defined as acceptable). The passing tone (second note of the cambiata) is more likely to jump to an imperfect consonance (the leap down by a third making a sixth with the cantus) than to a perfect consonance (the latter is slightly penalized), so we get cambiatas simply as a result of the preference for imperfect consonances.
Each added voice must obey all the normal rules with regard to the bass voice, but need not be quite so particular in relation to other voices. We assume here that the bass voice is either the cantus firmus or the first of the voices calculated by the Search mechanism. The extra rules for multi-voice counterpoint deal mainly with the makeup of chords.
INTEGER ARRAY IntervalsWithBass[0:7]; ! 0 = octave, 2 = step, 3 = third, 4 = fourth, 5 = fifth, 6 = sixth, 7 = seventh;
We need to look at the chord currently formed by the counterpoint and decide whether the pitch doublings are acceptable. In general, the leading tone should not be doubled, the fifth is rarely doubled, the third and sixth (figuring from the bass) can be doubled, but rarely tripled, and the octave can occur as often as necessary. A full chord (octave, third, and fifth for example), is better than a partial one (octave, octave, and third for example). In any case, the chord must contain at least one imperfect consonance. A root position triad is better than a first inversion triad (octave, third, and fifth is better than octave, third and sixth). Dissonances between upper voices are acceptable as long as the voices are individually consonant with the bass, but the 6-5 chord is a special case. In this chord the upper voices contain both the fifth and sixth above the bass, and according to some sources, the fifth should be treated as a form of suspension. To simplify matters we allow the 6-5 chord only in fifth species.
PROCEDURE AddInterval(INTEGER n); BEGIN ! collect current notes in chord; INTEGER ActInt; ActInt:=(CASE (n MOD 12) OF (0,2,2,3,3,4,4,5,6,6,7,7)); IntervalsWithBass[ActInt]:=IntervalsWithBass[ActInt]+1; END;
After each voice has checked that its proposed pitch agrees with the basic rules, it must also check for agreement with the other voices currently active. The following procedure handles this:
INTEGER PROCEDURE OtherVoiceCheck(INTEGER Cn,Cp,v,NumParts,Species); BEGIN INTEGER Val,k,CurBass,Other0,Other1,Int0,Int1,ActPitch,IntBass,LastCp; BOOLEAN AllSkip; IF v=1 THEN RETURN(0); ! two part or bass voice, so nothing to check; ARRCLR(IntervalsWithBass); ! our current chord; Val:=0; CurBass:=Bass(Cn,v);
Since we assume that either the first voice or the cantus firmus is the bass, we cannot allow other voices to cross below the current bass and thereby render invalid all the previous calculations of voice leading and chord type.
IF Cp <= CurBass THEN Val:=Val+CrossBelowBassPenalty;
We must also ensure that the raised leading tone in the bass (if present) does not confuse the consonance checker into thinking a diminished fourth is a major third.
IntBass:=((Cp-CurBass) MOD 12); IF IntBass=MajorThird AND NOT InMode(CurBass,Mode) THEN Val:=Val+AugmentedIntervalPenalty; ActPitch:=(Cp MOD 12); IF Val >= infinity OR (v=NumParts AND Dissonance[IntBass]) THEN RETURN(Val); ! logic here is that only the last part can be non-1st species and may therefore have various dissonances that don't want to be calculated as chord tones ; AllSkip:=ASkip(Cp-Us(Cn-1,v)); AddInterval(IntBass); LastCp:=Us(Cn-1,v); FOR k:=0 STEP 1 UNTIL v-1 DO BEGIN ! check our pitch against each other voice; Other0:=Other(Cn,v,k); Other1:=Other(Cn-1,v,k); IF NOT ASkip(Other0-Other1) THEN AllSkip:=FALSE; AddInterval(Other0-CurBass); ! add up tones in chord; ! avoid unison with other voice; IF NOT LastNote(Cn,v) AND Other0=Cp THEN Val:=Val+UnisonPenalty; ! keep upper voices closer together than lower; IF Other0 <> CurBass AND ABS(Cp-Other0) >= Octave+Fifth THEN Val:=Val+UpperVoicesTooFarApartPenalty; ! check for direct motion to perfect consonace between these two voices; Int0:=ABS(Other0-Cp) MOD 12; Int1:=ABS(Other1-LastCp) MOD 12; IF Int1=Int0 THEN IF Int0=unison THEN Val:=Val+ParallelUnisonPenalty ELSE IF Int0=fifth THEN Val:=Val+ParallelFifthPenalty; IF Cn>2 AND Int0=Unison AND (ABS(Us(Cn-2,v)-Other(Cn-2,v,k)) MOD 12)=Unison THEN Val:=Val+ParallelUnisonPenalty; IF Val >= infinity THEN RETURN(Val); ! penalize tritones between voices; IF Int0=Tritone THEN Val:=Val+VerticalTritonePenalty;
The following block of code attempts to ensure that a 6-5 chord is properly prepared and resolved. If there is a dissonance between these two voices, it is either a fourth or tritone (6-3 or 5-3 chord), a second (6-5 with 6th above), or a seventh (6-5 with fifth above). The dissonance in the 6-3 chord is considered unobjectionable. In the implementation here, the 6-5 chord is prepared by holding the fifth while the sixth is added (a sort of suspension below), then "resolve" the fifth by moving downward by step.
IF Species=5 THEN BEGIN IF Dissonance[Int1] AND Int1 <> fourth THEN ! must be a 6-5 chord; BEGIN INTEGER ourLastInt; ourLastInt:=((LastCp-Bass(Cn-1,v)) MOD 12); IF ourLastInt <> unison ! if unison, 6-6 somewhere else?; THEN IF (ourLastInt=fifth) THEN IF (ASkip(Cp-LastCp) OR Cp >= LastCp) THEN Val:=Val+UnresolvedSixFivePenalty ELSE ELSE IF (ASkip(Other0-Other1) OR Other0 >= Other1) THEN Val:=Val+UnresolvedSixFivePenalty; END; IF Dissonance[Int0] AND Int0 <> fourth AND IntBass <> unison THEN IF (IntBass=fifth AND (Cp-LastCp) <> unison) OR (IntBass <> fifth AND (Other0-Other1) <> unison) THEN Val:=Val+UnpreparedSixFivePenalty; END; ! penalize direct motion to perfect consonance except at the cadence; IF NOT LastNote(Cn,v) AND DirectMotionToPerfectConsonance(LastCp,Cp,Other1,Other0) THEN Val:=Val+InnerVoicesInDirectToPerfectPenalty; ! if we have an unraised leading tone it is possible that some other; ! voice has the raised form thereof (since the voices can move at very ! different paces, one voice's next to last note may be long before ! another's); IF ActPitch=10 AND ! if 11 we've aready checked; (Other0 MOD 12)=11 ! They have the raised form; THEN Val:=Val+DoubledLeadingTonePenalty; ! similarly for motion to a tritone; IF MotionType(LastCp,Cp,Other1,Other0)=DirectMotion AND Int0=Tritone THEN Val:=Val+InnerVoicesInDirectToTritonePenalty; ! look for a common diminished fourth (when a raised leading tone is in ! the bass, a "major third" above it is actually a diminished fourth. ! Similarly, an augmented fifth can be formed in other cases; IF ActPitch=3 AND (Other0 MOD 12)=11 THEN Val:=Val+AugmentedIntervalPenalty; ! try to encourage voices not to move in parallel too much; IF MotionType(LastCp,Cp,Other1,Other0) <> ContraryMotion THEN Val:=Val+NotContraryToOthersPenalty; END;
Now we must check the current contents of the chord being formed:
! check for doubled third; IF intervalsWithBass>1 THEN Val:=Val+ThirdDoubledPenalty; ! check for doubled sixth; IF IntervalsWithBass=0 AND IntervalsWithBass>1 THEN Val:=Val+DoubledSixthPenalty; ! check for too many voices at octaves; IF IntervalsWithBass>2 THEN Val:=Val+TripledBassPenalty; ! check for doubled fifth; IF IntervalsWithBass>1 THEN Val:=Val+DoubledFifthPenalty; ! check that chord contains at least one third or sixth; IF v=NumParts AND (NOT LastNote(Cn,v)) AND IntervalsWithBass=0 AND IntervalsWithBass=0 THEN Val:=Val+NotTriadPenalty; ! discourage all voices from skipping at once; IF v=NumParts AND AllSkip THEN Val:=Val+AllVoicesSkipPenalty; ! except in 5th species, disallow 6-5 chords altogether; IF IntervalsWithBass>0 AND IntervalsWithBass>0 AND Species <> 5 THEN Val:=Val+SixFiveChordPenalty; RETURN(Val); END;
The underlying method is a recursive search. It starts from its current pitch and tries in succession all possible melodic intervals from that pitch (if necessary), looking for any such interval whose associated cumulative penalty is less than the current best overall penalty. The cumulative penalty is the sum of all the penalties associated with each of the notes in the counterpoint melody. The overall best penalty is the lowest cumulative penalty of any complete counterpoint solution found to that point in the counterpoint. At the beginning of time this overall best penalty is infinity (no solutions have been found), but once any solution at all is found, the overall penalty is reset to that new number. If a position is reached that makes further progress impossible, the searcher backs up one note and tries some other interval. The intervals are chosen in an order that maximizes the chance of finding a good interval quickly. If any solution at all exists, we are guaranteed to find it. Given enough time, we are also guaranteed to find the "best" solution according to the rules. By checking only those branches whose penalty is less than the current minimum, we can drastically reduce the number of branches that must be checked, but total execution times can still be high. For a two part counterpoint with a short cantus firmus it is not unreasonable to carry out such a search, but more complex cases drag to a halt. Since our initial goal was to write five to eight part mixed species counterpoint, we obviously need more intelligence guiding the search.
The next method tried was inspired by an article in Science about computer circuit design by
simulated annealing. Like the other methods discussed below, this method gains much of its
efficiency by accepting less than optimal solutions Ð it is not guaranteed to find the best solution,
and may in some cases not find any solution at all. In this algorithm, our initial counterpoint is just
any random collection of notes. Our annealing "temperature" is the number of semitones each of
these notes can move. Time is represented by successive passes over the counterpoint applying the
same rules as were applied in the recursive search case, but here we look for the local minimum
penalty (whereas in the recursive case we grabbed the first acceptable branch and started down it).
Each note independently moves to its local minimum and the next pass is started. This method is extremely fast, and works well in first species counterpoint. It does not always converge on a very good solution, but we originally thought that we could run it across several random collections, and thereby increase our chance of getting something reasonable. In practice, however, these successive runs do not improve much. But more important, beyond first species the annealing process sometimes cannot find any acceptable solution. This problem is most easily observed in second species where a note can be dissonant if it is a passing tone. As the rules are structured, we do not penalize a step to a dissonance because one more step (if possible) will resolve the dissonance correctly. Nor do we ever look ahead to see if such a resolution is possible. If there is no possible resolution after all (if a passing tone is impossible at that point), the annealer has no good way to back out. After many fruitless attempts to get around this problem we finally jettisoned the entire notion.
Bernard Mont-Reynaud suggested changing the search to be a best-first search. In this version we compute the penalty associated with each possible melodic interval from the current pitch, then continue recursively using the best of these results first. If forced to abandon a branch, we back up and try the next best interval until a complete solution is found. Once we have a solution we drop back to the very beginning and try a different beginning interval. This rather odd looking practice grew from experience watching many hundreds of runs Ð generally most of the wasted effort (branches checked that led nowhere) seemed to be attributable to the fact that these new branches made no real difference in the global appearance of the given melody Ð much time was being spent optimizing something that had already given all it had to give. By trying a new starting interval, we maximize the chance of finding a truly different solution. Once again we abandon any branch if its accumulated penalty is above that of the best complete solution found so far. Although first through fourth species are quickly solved, fifth species is still a problem. The number of choices increases in this species not only because it generally has more notes in the counterpoint melody, but also because we have a number of possible rhythmic values to assign each note. Several optimizations were added to reduce this problem: we search only for those solutions that are markedly better than the current one (as determined by the variable PenaltyRatio), and once a solution has been found we don't spend too much time on any other single interval (controlled by the variables MaxBranch and Branches). Compute times for multi-voice fifth species can be high even with all this machinery.
To illustrate how the rules help the program decide whether one melody is better than another, take the following three acceptable counterpoints to the bottom line (the cantus firmus). The number under each note is the penalty associated with that note. Despite the fact that the first (top) solution starts out with a higher penalty, it ends up with a lower overall penalty. This case illustrates that a simple best first search is not entirely adequate in all cases.
We can point out in detail how each note in these lines gets its penalty:
/ | / | | | | / | | / | / | | \ \ repeated | | / | / | repeated \ \ pitch | | direct | direct | pitch | | (1) | | motion | motion | (1) repeated | direct | (1) | (1) | pitch | motion | direct skip | (1) skip (1) | motion preceded | followed upper | (1) by same | by same neighbor | direction | direction (the g) | (1) | (3) (1) | | lower direct neighbor motion (the fs) (1) two repeated notes (2) perfect consonance (2) repeated pitch (2) | | | | | | | | | / repeated / / / | | direct | / pitch / / direct | | motion | / (1) / / motion | | (1) | compound interval (1) / / (1) / / perfect skip followed by same direction (3) / | / / consonance two skips (1) / repeated / / (2)
skip preceded by same direction (1) | pitch / direct lower | (1) | motion neighbor | perfect | (1) (1) | consonance | repeated repeated | (2) | pitch pitch repeated pitch (1) | (1) (1) skip followed by same direction (3) | perfect consonance (2) repeated pitch (1)
In the first attempt at multi-part counterpoint we solved one voice at a time and built up the entire ensemble by layering. This worked well for three voices because the first voice added was always pretty good, and the second added voice still had enough degrees of freedom to find an acceptable solution. As more voices were added however, the later layers became less and less acceptable. It became clear that the entire ensemble has to be calculated together, that all the voices must be examined to decide the current overall best configuration. The search routines that follow implement this form of search.
First we need a place to save the last complete solution while we search for a better one:
INTEGER ARRAY BestFit[1:MostNotes,1:MostVoices];
and variables to hold the current maximum penalty (MaxPenalty), the current actual best (lowest) penalty (BestFitPenalty), the ratio that determines how much better a new solution has to be to be worth pursuing (PenaltyRatio), a flag to tell the searcher when to quit (AllDone), the current branch counter (Branches), and the maximum number of branches we are willing to search before giving up (MaxBranch).
INTEGER BestFitPenalty,MaxPenalty,Branches,MaxBranch; BOOLEAN AllDone; REAL PenaltyRatio;
We will examine only the best continuations from any given point, making no effort to save every possible continuation. The macro NumFields determines how many continuations we save at each branch.
Each continuation consists of its associated penalty and the melodic intervals of each voice that make up the next note of the continuation. At each branch we search for the NumFields best continuations and save them in an array. The following code implements this mechanism.
DEFINE Field="(MostVoices+1)"; DEFINE EndF="Field*NumFields"; PROCEDURE ClearSpace(INTEGER ARRAY Sp); ARRCLR(Sp,infinity); INTEGER PROCEDURE SaveIndx(INTEGER indx; INTEGER ARRAY sp); BEGIN ! if INDX is less than current NUMFIELD-th worst, ! find its position in SP, insert space for its ! data, and return a pointer to the block. The ! blocks are stored "backwards" for ARRBLT; INTEGER i; FOR i:=EndF STEP -Field UNTIL 0 DO IF Sp[i]>indx THEN DONE; IF i>0 ! 0 is the end of the list. If i>0 then we insert INDX; THEN BEGIN ARRBLT(Sp,Sp[Field],i); Sp[i]:=Indx; ! SP[i]=penalty for block starting at I. SP[i-1]=index into ! melodic interval array for voice 1, SP[i-2] for voice 2 and ! so on. The searcher starts at SP[EndF] and works backwards ! through the stored continuations as it searches for a satisfactory ! overall solution; END; RETURN(i); END;
The next procedure takes a newly completed solution and saves it. At this time we also check for problems associated with raised leading tones. If the musica ficta rules were at all clear, we could also add them to the solution. The raised leading tone code can run into problems Ð a future version may try to be smarter about cadential passages. PROCEDURE SaveResults(INTEGER CurrentPenalty,Penalty,v1,Species); BEGIN INTEGER i,v,LastPitch,v,Cn,k; FOR v:=1 STEP 1 UNTIL v1 DO ! check all voices for raised leading tone; BEGIN Cn:=TotalNotes[v]; LastPitch:=(Us(Cn-1,v) MOD 12); ! must be raised if any are; IF (NOT InMode(LastPitch,Mode)) ! it is a raised leading tone; THEN FOR k:=2 STEP 1 WHILE TRUE DO ! look backwards through voice's notes; BEGIN "outer" INTEGER Pitch; IF k >= (Cn-1) THEN DONE; ! ran off start!!; Pitch:=(Us(Cn-k,v) MOD 12); ! current pitch; IF (Pitch<8 AND Pitch <> 0) OR ! not 6-7-1 scale degree anymore; (ASkip(Us(Cn-k+1,v)-Us(Cn-k,v))) ! skip breaks drive to cadence; THEN DONE; ! so exit; Pitch:=ABS(Us(Cn-k,v)-Us(Cn-k-1,v)); ! interval with raised leading tone; IF pitch=Fourth OR Pitch=Fifth OR Pitch=unison OR Pitch=Octave THEN DONE; ! don't create illegal melody; FOR i:=0 STEP 1 UNTIL v1 DO ! do others have unraised form?; IF i <> v AND (Other(Cn-k,v,i) MOD 12)=11 THEN DONE "outer"; IF ((Us(Cn-1,v)-Us(Cn-k,v))=MinorThird) OR ((Us(Cn-1,v)-Us(Cn-k,v))=MinorSecond) THEN ! raise it and maybe 6th degree too; SetUs(Cn-k,Us(Cn-k,v)+1,v); END "outer"; END; END;
Musica ficta are not entirely trivial to add to the program. If we were doing so, this would be the place to do it. The ficta-finder would run through all the voices here looking for tritones. If one is found a decision has to be made whether to alter (flatten or sharpen) the current note, or leave it alone. In a simple version tried during the development of this program, we used the following sequence:
Step through all voices Step through all notes of each voice If current pitch is B (11) and it is approached and left by either a unison, a rising major second, a rising major second, a descending minor second, or a descending minor third (to avoid creating illegal augmented or diminished intervals melodically), and it forms a tritone with some other voice, Then If melodic interval is not a unison Then flatten the B Else While pitch is B, look to see if the flattened form will create tritones (E in other voices), and check to see that the interval on either side of the repeated B's is the correct form of interval. If all this is true, flatten all the B's.
This algorithm is moderately conservative (it makes no attempt to sharpen F's for example), but still makes a mess of certain passages. For example, take the simple passage:
Here our algorithm flattens the second B but not the first (correctly avoiding a tritone melodically), but perhaps the correct ficta is an F-sharp. The algorithm also ignores the possible presence of the unaltered pitch in other voices as in:
In any case, since ficta are not a primary concern of the author, the entire problem was shunted aside and left as an exercise for the interested reader.
BestFitPenalty:=CurrentPenalty+Penalty; MaxPenalty:=(BestFitPenalty*PenaltyRatio) MIN MaxPenalty; AllDone:=TRUE;
We have now updated the value of the best fit penalty and the maximum penalty that a new solution should have, and have signalled the searcher to back up to the start again. Next we save the current state of the CtrPt array (which holds the soution transposed to C) in the BestFit array, adding back the original transposition:
FOR v:=1 STEP 1 UNTIL v1 DO FOR i:=1 STEP 1 UNTIL TotalNotes[v] DO BestFit[i,v]:=CtrPt[i,v]+BasePitch;
Now we print out the results for debugging and what not.
PRINT(" ! [",BestFitPenalty,"] "); FOR v:=1 STEP 1 UNTIL v1 DO BEGIN PRINT(" "); FOR i:=1 STEP 1 UNTIL TotalNotes[v] DO BEGIN PRINT(BestFit[i,v]," "); IF i <> 1 THEN PRINT("[",Check(i,BestFit[i,v]-BasePitch,v,v1, (IF v=v1 THEN Species ELSE 1)),"] ") ELSE PRINT(" "); END; END; PRINT(" "); END;
We need an array containing the legal melodic intervals in the order in which we want them to be checked. Whether this order is of great importance or not depends on how the searching is carried out. In the current scheme, the order doesn't matter much.
PRELOAD_WITH 1,-1,2,-2,3,-3,0,4,-4,5,7,-5,8,12,-7,-12; INTEGER ARRAY Indx[1:16];
We need a procedure that examines the possible continuations and puts the best such continuations in order in a local array (on the stack). Each time the searcher lurches forward a note, a new recursive call is made allocating this and other arrays of local data. We can back up without losing previous state merely by returning from a call.
RECURSIVE PROCEDURE Look( INTEGER CurPen,CurVoice,NumParts,Species; REFERENCE INTEGER Lim; INTEGER ARRAY Pens,Is,CurNotes); BEGIN INTEGER penalty,Pit,i; FOR is[CurVoice]:=1 STEP 1 UNTIL 16 DO BEGIN pit:=Indx[is[CurVoice]]+Ctrpt[CurNotes[CurVoice]-1,CurVoice];
Pit is the proposed new pitch for the current voice (CurVoice). For each such pitch we check how it affects the global penalty of the rest of the active voices
penalty:=CurPen+Check(CurNotes[CurVoice],Pit,CurVoice,NumParts, (IF CurVoice=NumParts THEN Species ELSE 1)); SetUs(CurNotes[CurVoice],Pit,CurVoice); IF penalty<Lim THEN IF CurVoice<NumParts THEN BEGIN
We have a possible continuation for CurVoice. Now we need to find whether other voices can
continue also given this pitch. Since the voices may not be moving together, we check new pitches only for those voices that are getting a new note at the current time.
FOR i:=CurVoice+1 STEP 1 UNTIL NumParts DO IF CurNotes[i] <> 0 THEN DONE; IF i <= NumParts ! there is another voice needing a note; THEN Look(Penalty,i,NumParts,Species,Lim,Pens,Is,CurNotes); END ELSE
If the current voice is the last voice, then we save the current proposed overall continuation in the Pens array, and continue looking for more. When we finish this loop we will have all the best solutions in order in the Pens array.
BEGIN INTEGER x,i; x:=SaveIndx(Penalty,Pens); IF x>0 THEN FOR i:=1 STEP 1 UNTIL NumParts DO Pens[x-i]:=is[i] ELSE Lim:=Lim MIN Penalty; END; END; END;
Next we need a procedure that calls Look (given above), and coordinates all the voices as the ensemble grinds toward the cadence.
RECURSIVE PROCEDURE BestFitFirst(INTEGER CurTime,CurrentPenalty,NumParts,Species); BEGIN INTEGER i,j,Curmin,Lim,ChoiceIndex,NextTime,OurTime; INTEGER ARRAY Pens[0:Field*NumFields],Is[1:NumParts],CurNotes[1:MostVoices]; IF AllDone OR CurrentPenalty>MaxPenalty THEN RETURN;
AllDone is true after we have found a solution. At that point we exit all the calls currently active and start afresh with a new maximum penalty.
ChoiceIndex:=EndF; AllDone:=FALSE; ARRCLR(Pens,infinity); ARRCLR(Is); ARRCLR(CurNotes); Branches:=Branches+1; IF Branches>MaxBranch THEN RETURN; IF (Branches MOD 10) =0 THEN MaxPenalty:=MaxPenalty*PenaltyRatio;
If we just let the searcher run until a solution is found, it happens quite frequently that the search gets completely bogged down in a blind alley, spending immense amounts of time beating up against an impossible cadence situation. The branch counter gives us a way to jump out of such a situation. In addition, as we spend more and more time on a given attempt, we gradually reduce the acceptable maximum penalty somewhat like a person getting more and more frustrated as more effort is poured into a fruitless search.
CurMin:=infinity; Lim:=BestFitPenalty-CurrentPenalty; NextTime:=infinity; FOR i:=1 STEP 1 UNTIL NumParts DO BEGIN OurTime:=Onset[VIndex(CurTime,i)+1,i]; IF OurTime <> 0 THEN NextTime:=NextTime MIN OurTime; END;
We now know what the next note is overall. We set up the CurNotes array marking all voices that have an onset at the minimum next time. Each of these voices will then take part in the search for the best overall continuation.
FOR i:=1 STEP 1 UNTIL NumParts DO IF Onset[j:=Vindex(NextTime,i),i]=NextTime THEN CurNotes[i]:=j; FOR i:=1 STEP 1 UNTIL NumParts DO IF CurNotes[i] <> 0 THEN DONE; Look(0,i,NumParts,Species,Lim,Pens,Is,CurNotes);
Now we have the NumFields best continuations saved in the Pens array along with the associated penalties. We go through these continuations one at a time, trying to find one that will get us all the way to a cadence.
CurMin:=Pens[ChoiceIndex]; IF CurMin >= Infinity THEN RETURN; WHILE NOT AllDone DO BEGIN IF CurTime<TotalTime THEN IF (curmin+currentpenalty >= MaxPenalty) THEN RETURN ELSE ELSE IF (curmin+currentpenalty >= BestFitPenalty) THEN RETURN;
We are still below the global maximum penalties, so set up the next continuation and give it a whirl.
FOR i:=1 STEP 1 UNTIL NumParts DO IF CurNotes[i] <> 0 THEN SetUs(CurNotes[i],indx[Pens[ChoiceIndex-i]]+Us(CurNotes[i]-1,i),i); IF NextTime<TotalTime THEN BestFitFirst(NextTime,CurrentPenalty+CurMin,NumParts,Species) ELSE ! we've reached the end; SaveResults(CurrentPenalty,CurMin,NumParts,Species);
The variable ChoiceIndex points to where we are in the continuations. By decrementing it by field size, we move on to the next best continuation.
ChoiceIndex:=ChoiceIndex-Field; IF ChoiceIndex <= 0 THEN RETURN; CurMin:=Pens[ChoiceIndex]; IF CurMin=infinity THEN RETURN; IF CurTime=0 THEN MaxPenalty:=BestFitPenalty*PenaltyRatio; END; END;
The only thing left that is of any interest is the code that decides what rhythms to employ in fifth
species. For simplicity's sake, we just load up an array with the legal rhythmic patterns and choose
among them randomly. This approach obviously leaves much to be desired. Musical styles are
differentiated more by rhythmic practices than melodic, and on a smaller level, it is the fluid
handling of rhythm that makes a group of sounds a piece of music rather than a pedagogical exercise.
However, this entire program ignores issues of phrasing or larger melodic and rhythmic structures. If we decide to carry this effort further in that direction, a much more complex decision and search mechanism will be required. Just as Fux's book was one step on the way toward Parnassus, this program, simple minded, even toy-like in many ways, may lead toward good music somewhere down the road. In any case, messing around with counterpoint is a pleasure in itself, so the author needed no further justification.
INTEGER Seed; ! random number sequence seed; INTEGER ARRAY RhyPat[0:10,0:8],RhyNotes[0:10]; ! array of legal rhythmic patterns; SIMPLE PROCEDURE FillRhyPat; BEGIN INTEGER i; RhyPat[0,1]:=wholeNote; ! for first species "rhythms"; RhyNotes:=1; FOR i:=1 STEP 1 UNTIL 2 DO RhyPat[1,i]:=CASE i OF (0,halfnote,halfnote); RhyNotes:=2; FOR i:=1 STEP 1 UNTIL 3 DO RhyPat[2,i]:=CASE i OF (0,halfnote,quarternote,quarternote); RhyNotes:=3; FOR i:=1 STEP 1 UNTIL 4 DO RhyPat[3,i]:=quarternote; RhyNotes:=4; FOR i:=1 STEP 1 UNTIL 3 DO RhyPat[4,i]:=CASE i OF (0,quarternote,quarternote,halfnote); RhyNotes:=3; FOR i:=1 STEP 1 UNTIL 4 DO RhyPat[5,i]:=CASE i OF (0,quarternote,eighthnote,eighthnote,halfnote); RhyNotes:=4; FOR i:=1 STEP 1 UNTIL 5 DO RhyPat[6,i]:=CASE i OF (0,quarternote,eighthnote,eighthnote,quarternote,quarternote); RhyNotes:=5; FOR i:=1 STEP 1 UNTIL 4 DO RhyPat[7,i]:=CASE i OF (0,halfnote,quarternote,eighthnote,eighthnote); RhyNotes:=4; FOR i:=1 STEP 1 UNTIL 6 DO RhyPat[8,i]:=CASE i OF (0,quarternote,eighthnote,eighthnote,quarternote,eighthnote,eighthnote); RhyNotes:=6; FOR i:=1 STEP 1 UNTIL 5 DO RhyPat[9,i]:=CASE i OF (0,quarternote,quarternote,quarternote,eighthnote,eighthnote); RhyNotes:=5; RhyPat[10,1]:=wholeNote; ! for first species "rhythms"; RhyNotes:=1; END; REQUIRE FillRhyPat INITIALIZATION;
Now a few simple procedures to clear and access the arrays associated with the rhythm finder.
PROCEDURE UsedRhy(INTEGER n); RhyPat[n,0]:=RhyPat[n,0]+1; INTEGER PROCEDURE CurRhy(INTEGER n); RETURN(RhyPat[n,0]); PROCEDURE CleanRhy; BEGIN INTEGER i; FOR i:=1 STEP 1 UNTIL 9 DO RhyPat[i,0]:=0; END; INTEGER PROCEDURE GoodRhy; BEGIN INTEGER i; i:=10*RAN(seed); IF (CurRhy(i)>CurRhy(1 MAX (i-1))) THEN RETURN(1 MAX (i-1)); IF (CurRhy(i) <= CurRhy(9 MIN (i+1))) THEN RETURN(9 MIN (i+1)); RETURN(i); END;
The rest of the program merely stuffs the cantus firmus into the CtrPt array and sets the starting
points of the various voices. We append several examples of counterpoint written by the program.
Under each counterpoint note is the associated penalty. Currently the program has no provision for starting a melody with a rest, nor does it reward invertible counterpoint and imitation. It tends to let voices get entangled in each other, and makes no decisions about overall melodic shapes. Certain implied vertical clashes are not noticed (especially those involving tritones). Given these caveats the program does quite well in a reasonably short time.
I thank Jonathan Berger for his very valuable advice and criticisms.
Here is the rest of the program (not originally included in the paper because it is so dependent on the operating system I was using):
; BOOLEAN PROCEDURE AnySpecies(INTEGER ourMode,StartPitch); BEGIN "Any" INTEGER i,j,k,m,v,OldSpecies; ARRCLR(BestFit); ARRCLR(CtrPt); PenaltyRatio:=0.7; CurrentMode:=OurMode; TotalNotes:=CantusFirmusLength; BasePitch:=TCF[CantusFirmusLength] MOD 12; BestFitPenalty:=MaxPenalty:=infinity; ! IF CurV=1 THEN MaxPenalty:=bad!-1; FOR i:=1 STEP 1 UNTIL TotalNotes DO TCF[i]:=TCF[i]-BasePitch; PRINT(" Cantus firmus: "); FOR i:=1 STEP 1 UNTIL TotalNotes DO PRINT(TCF[i]+BasePitch," "); OldSpecies:=Species; FOR v:=1 STEP 1 UNTIL Curv DO BEGIN "loop" IF v<>curV THEN Species:=1 ELSE Species:=OldSpecies; IF Species=1 THEN BEGIN FOR i:=1 STEP 1 UNTIL TotalNotes-1 DO Dur[i,v]:=wholenote; END ELSE IF Species=2 THEN BEGIN TotalNotes:=CantusFirmusLength*2-1; FOR i:=1 STEP 1 UNTIL TotalNotes-1 DO Dur[i,v]:=halfnote; END ELSE IF Species=3 THEN BEGIN TotalNotes:=CantusFirmusLength*4-3; FOR i:=1 STEP 1 UNTIL TotalNotes-1 DO Dur[i,v]:=QuarterNote; END ELSE IF Species=4 THEN BEGIN TotalNotes:=CantusFirmusLength*2-1; FOR i:=1 STEP 1 UNTIL TotalNotes-1 DO Dur[i,v]:=HalfNote; END ELSE BEGIN "5" CleanRhy; m:=0; FOR i:=1 STEP 1 UNTIL CantusFirmusLength-1 DO BEGIN j:=GoodRhy; UsedRhy(j); FOR k:=1 STEP 1 UNTIL RhyNotes[j] DO Dur[k+m,v]:=RhyPat[j,k]; m:=m+RhyNotes[j]; END; TotalNotes:=m+1; END "5"; Dur[TotalNotes,v]:=wholenote; Onset[1,v]:=0; FOR k:=2 STEP 1 UNTIL TotalNotes DO Onset[k,v]:=Onset[k-1,v]+Dur[k-1,v]; IF V=1 THEN Ctrpt[1,1]:=StartPitch-BasePitch ELSE IF StartPitch-BasePitch>TCF THEN IF ((StartPitch-BasePitch) MOD 12)=fifth THEN Ctrpt[1,v]:=TCF - (12*(v-1)) ELSE Ctrpt[1,v]:=StartPitch-BasePitch - (12*(v-1)) ELSE IF ((StartPitch-BasePitch) MOD 12)=fifth THEN Ctrpt[1,v]:=TCF + (12*(v-1)) ELSE Ctrpt[1,v]:=StartPitch-BasePitch + (24*(v-1)); IF v=1 THEN MaxPenalty:=2*RealBad! ELSE MaxPenalty:=infinity; Search(v); IFC Display THENC S_Prepare; S_Add; S_End; S_Plot; ! INCHWL; ENDC END "loop"; END "Any"; INTEGER k,brk,eof,chn; REQUIRE "Billib.Rel[Lib,Bil]" LIBRARY; EXTERNAL PROCEDURE Crawl; EXTERNAL INTEGER _ERRP_; ! where SAIL looks for its error handler; SIMPLE INTEGER PROCEDURE SAILError(INTEGER loc; STRING msg,rsp); BEGIN PRINT(Msg); Crawl; RETURN(3 LSH 18); END; SIMPLE PROCEDURE LogIt(INTEGER Chan; STRING S); BEGIN OUT(Chn,S); IF Chan=-1 THEN OUTSTR(S) ELSE OUT(Chan,S); END; EXTERNAL INTEGER ARRAY Gogtab[0:'300]; DEFINE $$PROU='255; _ERRP_:=LOCATION(SAILError); GOGTAB[$$PROU]:=LOCATION(LogIt); OPEN(Chn:=GETCHAN,"DSK",0,0,31,0,brk,eof); ENTER(Chn,"FUX.LOG",Eof); Seed:=431; Seed:=RAN(Seed); IFC Display THENC INTMAP(INTMAIL_INX,Post,0); ! interrupt on interjob mail and drop into POST; ENABLE(INTMAIL_INX); INTMAP(INTTTI_INX,EscI,0); ! Interrupt on ESC-I and drop into EscI Procedure; ENABLE(INTTTI_INX); S_Short:=FALSE; ! make a full score; S_Init; ! get an EDMS job, accept all defaults; ENDC FOR Species:=2 STEP 1 UNTIL 5 DO ! Species:=5; BEGIN PRINT(" SPECIES ",Species," "); CantusFirmusLength:=11; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (50,53,52,50,55,53,57,55,53,52,50)); AnySpecies(Dorian,57); CantusFirmusLength:=11; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (50,53,52,50,55,53,57,55,53,52,50)); AnySpecies(Dorian,38); CantusFirmusLength:=10; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (52,48,50,48,45,57,55,52,53,52)); AnySpecies(Phrygian,59); CantusFirmusLength:=10; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (52,48,50,48,45,57,55,52,53,52)); AnySpecies(Phrygian,40); CantusFirmusLength:=12; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (53,55,57,53,50,52,53,60,57,53,55,53)); AnySpecies(Lydian,65); CantusFirmusLength:=12; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (53,55,57,53,50,52,53,60,57,53,55,53)); AnySpecies(Lydian,41); CantusFirmusLength:=14; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (43,48,47,43,48,52,50,55,52,48,50,47,45,43)); AnySpecies(Mixolydian,55); CantusFirmusLength:=14; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (43,48,47,43,48,52,50,55,52,48,50,47,45,43)); AnySpecies(Mixolydian,43); CantusFirmusLength:=12; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (45,48,47,50,48,52,53,52,50,48,47,45)); AnySpecies(Aeolian,57); CantusFirmusLength:=12; FOR k:=1 STEP 1 UNTIL CantusFirmusLength DO TCF[k]:=(CASE k-1 OF (45,48,47,50,48,52,53,52,50,48,47,45)); AnySpecies(Aeolian,45); END;