|
Automated Music Composition using Motifs
& Transformations (A class project for CSCI575) By - Anwar Hossain Click here
to go to my personal website Limitations & Future Improvements There have been
a lot of talks, dreams, and attempts to automate Music Composition process.
Music composition is a creative process and thus we expect to see unexpected
twist in it. Over history there have been different genres and sometimes few
dominating ones. And those dominating ones controlled the shaping of what kind
of composition will take place. In recent time, in the movies we see the
‘Motif-based’ or ‘Theme-based’ music have its own very powerful and interesting
area. Composers working on this type of music starts with one or more than one
motif (a small piece of music) and twists its different parameters to bring
different moods in different parts of the movie. In my project, I basically
attempted to automate this process in a simple, scalable framework. The system
starts with some motifs and generates many more motifs based on different
combination of transformations on the motifs. Then, a constrained selection
process patches selected motifs together to make a bigger piece. As another
benefit, the system may also help us better understand the correlation between
motifs and how some composers approaches to create different piece out of
those. Middle ages: Around 500
A.D., western civilization began to emerge from the period known as "The
Dark Ages" .For the next ten centuries, the newly emerging Christian
Church would dominate . While music composition is a creative process,
it is not just random notes. There is this mysterious term ‘nice music’ which
is difficult to formularize. But, again the history of music is very
interesting and surely goes through lot of twists. The Renaissance: Generally
considered to be from ca.1420 to 1600, the Renaissance was a time of great cultural awakening throughout The Baroque Age: Named after the
popular ornate architectural style of the time, the Baroque period (ca.1600 to
1750) saw composers beginning to rebel against the styles that were prevalent
during the High Renaissance. Many monarchs employed composers at their courts,
where they were little more than servants expected to churn out music for any
desired occasions The greatest composer of the period, Johann Sebastian Bach,
was such a servant. Yet the best composers of the time were able to break new
musical ground, and in so doing succeeded in creating an entirely new style of
music. The Classical Period: From roughly
1750 to 1820, musicians moved away from the heavily ornamented styles of the
Baroque and the Rococo, and instead embraced a clean, uncluttered style they
thought reminiscent of Classical Greece. The Romantic Era: As the many
socio-political revolutions of the late eighteenth-century established new
social orders and new ways of life and thought, so composers of the period
broke new musical ground by adding a new emotional depth to the prevailing
classical forms. Throughout the remainder of the nineteenth-century (from ca.
1820 to 1900), artists of all kinds became intent in expressing their
subjective, personal emotions. The Twentieth-Century: By the turn of
the century and for the next few decades, artists of all nationalities were
searching for exciting and different modes of expression. Composers such as
Arnold Schoenberg explored unusual and unorthodox harmonies and tonal schemes.
In addition to new and eclectic styles of musical trends, the twentieth century
boasts numerous composers whose harmonic and melodic styles an average listener
can still easily appreciate and enjoy. Components of the MotifMax System: The system was
developed in an object-oriented fashion. Each motif was designated as an object
or class. Each motif class also had a number of transformation methods. The
motifs are fed from an easily updateable text file containing motif
information. For the experiment I used the following motif file “MotifB.txt” – StartMotif------ 480 E 3 100 480 C 4 110 200 B 3 120 100 A 3 127 EndMotif------- StartMotif------ 480 E 3 100 240 G 3 128 240 A 3 128 240 B 3 110 480 E 3 110 EndMotif------- StartMotif------ 480 F 3 100 480 A 3 128 480 D 4 128 480 F 3 110 EndMotif------- StartMotif------ 480 A 3 100 480 C 4 128 480 E 4 128 480 B 3 110 EndMotif------- StartMotif------ 550 G 4 100 550 E 4 128 550 D 4 128 550 C 4 128 550 B 3 110 550 C 4 128 550 D 4 128 550 E 4 128 550 F 4 128 EndMotif------- END The above
example file has 5 motifs. Each motif data is surrounded by the 2 tags StartMotif------ and EndMotif-------
. The file ends with the tag END.
In each motif there are 4 columns and they are (from left to right)– - Note Length (e.g. 480) - Note Name (e.g. E) - Note Octave (e.g. 3) - Note Volume (e.g. 100) Transformation
function is the main playground for changing a motif to a different variation
of it which sometimes can be as simple as shifting pitch or changing scale or
it could be as interesting as changing different note duration and inversion. In
the current version of the system only a cut-down version of these functions
have been implemented – Shifting Pitch: This is a very
common practice in music. Now, in some literature and situations especially in
vocal music, the pitch shifting is done by time stretching. But in the MotifMax
system we do not deal with voice data rather we have known pitch class. So
basically we shift pitch the way someone would do in piano. This transformation
functions takes “Offset” as a parameter which can be either a positive or
negative integer. All the notes in the motif are shifted by notes specified by
Offset parameter. So, if a note in the motif were, C3, then the pitch will be
E3 after shifting if the offset is +2, and it will be B2 if the offset is -1. By
using the offset parameter 7, a whole scale shift can occur. Scale Time: This is another
common technique in music. This function basically means changing the tempo of
the music. The system allows limited range for this change to keep the
generated motif not so unrealistically fast or slow. The function takes
“Factor” as a parameter. If the factor is more than 1, then the time is scaled
up and the tempo becomes slower. If the factor is less than 1, then the scale
is timed down making the tempo faster. This could bring the excited, sad etc
mood in a piece of motif. Change Volume: The
transformation function is self explanatory from its name. The volume of all
the notes in a motif is changed by some positive or negative offset. This could
indicate highetended tension or mood if the volume is increased. Inverse: This
transformation is the one which is not obvious to detect. The notes and related
information are just reversed. So, the very first note in a motif becomes the
very last note and versa. The very first motif in our experimental file is - StartMotif------ 480 E 3 100 480 C 4 110 200 B 3 120 100 A 3 127 EndMotif------- After inverse,
it becomes – StartMotif------ 100 A 3 127 200 B 3 120 480 C 4 110 480 E 3 100 EndMotif------- This phase of
the system is more of a control block. This part decides whether a motif
(including the newly generated ones by transformation) should be selected or
not and if yes, in which location it should be selected. This phase also takes
a parameter “IterationCount” as input. The music generation process stops when
the number of motifs in the generated piece becomes same as “IterationCount”. These rules or
constrained determine whether a music piece is a rigid one or atonal. In this
aspect, we can refer to the paper “Music Generation from Statistical Models” by
Darrell Conklin. In that paper the author mentions several approaches for music
generation by computer. One of them was Statistical Modeling technique. These
models can classify whether a music is of certain genre or by certain
composers. Then the author proposes – using random walk or stochastic sampling
or some other sampling methods pieces (in our case motifs) can be chosen.
Whether the motif will be retained or not can be decided by the model based on
whether that generated piece is classify as right for intended music type. In my
implementation I wanted to take a little bit different approach. I wanted to use
“Evolution”. This feature is
yet to implement in the project. But this is the part which is supposed to take
care of the basic question we discussed in the introduction of so called “nice
music”. To implement this, the first thing we need is “Fitness” function to
quantify the how nice a piece of music is. The function will give a score for a
piece of music based on its rhythm and melody structure. We would use
Genetic Algorithm for the evolution. So, in brief the steps for the music
composition would be as follows – 1.
Create
your kernel or seed motifs in the text file 2.
Apply
different transformations in different combinations to create lot of other
motifs from the kernels 3.
Select
a motif based on the Flexi rule sets from the pool of motifs and add it to the
already generated piece. 4.
Repeat
steps 3 several times to get intended length for piece of music 5.
Repeat
steps 3 & 4 to create multiple piece which will be used as the base
population for the evolution process 6.
Define
a fitness function to quantify how nice a piece is 7.
Select
2 pieces of music from the base population 8.
Do
some random transformation on them, and possibly cross over and generate new
pieces and add it to the base population 9.
Apply
the fitness function and score all pieces in the population 10.
Remove
the piece with the smallest score 11.
Repeat
7 to 10 until no pieces gets removed or certain iteration number is reached. 12.
These
will end up in results having several pieces of nice music Some sample
output is added below. Click on the link to download the midi files and play
it. Limitation & Future Improvement: The number of
transformation functions implemented needs to be enhanced. Multi-track music is
not track-synchronization at this point. Evolution method needs to be added. Since the
evolution is yet to implement, at this point, we basically run the program,
generate music, keep the one we like and discard the others. Even though this
process may sound a bit manual, but we surely can think of application of it.
When we are trying to compose music using some motifs, this program can play
around those motifs in lot of different ways and might play something that we
otherwise would not think about and may like it. Since the system automatically
saves the music in
-
Visual
Studio C++ -
Txt2Midi
by Guenter Nagler Download Source. Download Executables. |
