Daniel J. Rothblatt, February 2016

This program generates a Markov model of a corpus of music transcribed in ABC notation. ABC notation is a simple, easily readable music transcription system used largely to transcribe folk music. As a result, the data provided in big_guy.abc is a corpus of English folk music.

• README.md: This file
• big_guy.abc: Music collection in ABC notation
• clean_keys.sed: Preprocesses keys to a uniform format
• clear_comments.sed: Removes comments from ABC files
• fix_meters.sed: Changes meter to a known meter if unknown meter
• helper.py: Auxiliary functions that prove useful in other libraries
• prepare.py: Separates data from metadata in each piece and prepares pieces for noodle.py
• rhythm.py: Computes duration of a musical phrase
• transpose.py: Transposes a piece to a target key
• noodle.py: Generates a piece of music from input by building a Markov model.
• music_generator.py: RUN THIS! Reads in a user-specified ABC file and outputs a piece of music.

This project was inspired by David Pesetsky and Jonah Katz of MIT. Their Identity Thesis for Music claims that music and language have fundamentally the same syntactic structure--the only difference being the types of elements that are treated as syntactic objects. A prediction that the Identity Thesis puts forward is that a Markov model is wholly inadequate for producing music: Just as a Markov model of the syntax of a language will always be able to generate ungrammatical sentences (because syntax is not a regular language), a Markov model of the structure of music should generate "ungrammatical" music. The question was, "What would ungrammatical music look like?" The tentative conclusion of this project is that "ungrammatical" music is atonal--pieces may modulate freely from key to key without ever returning to the tonic (initial key of a piece), because the linear nature of a Markov model prevents it from remembering what the tonic is. However, the project as it currently stands has major flaws that make this conclusion only a tentative one. (See "To Do" below.)

Run music_generator.py from the command line.

<command> ::= python3 music_generator.py <unary_flag>*
<unary_flag> ::= <file_flag> | <meter_flag> | <length_flag> |
<key_flag>
<file_flag> ::= (-i | --in | -o | --out) \w+.abc

Unary Flags
      - file: indicates input or output file
      - meter: specifies the meter of the piece (what type of note
      gets the beat, how many beats per measure). Note that the
      note that gets the beat should always be a power of 2
      greater than or equal to 1 (i.e., for all m,n: m/2^n)
      - length: specifies how many beats per measure. Should be an
      integer.
      - key: specifies the key of the piece.

I (Daniel Rothblatt) do not consider this project complete in the slightest. The major problems currently facing Noodle are listed here, as well as in music_generator.py:

• This project does not really generate bar lines (measure delimiters) in a sensible way yet. The author was under the assumption that the bar lines would be easier to insert than they in fact are.
• This project does not handle anything but note duration and pitch--additional work would enable it to keep track of things like dynamics and ties.