Witiko / markov-midi

An unsupervised learning algorithm that generates MIDI tracks statistically similar to the input.

This tool cuts input standard MIDI files into tracks, feeds these track to a Markov chain, and generates a new random track.

Requirements

The script relies on Bash, Lua 5.2, and the midicsv package that provides the midicsv and csvmidi binaries for the conversion of standard MIDI files to CSV and vise-versa.

Usage

Run the markov-chain.sh script as follows:

./markov-chain.sh {LEFT_CONTEXT|-} {MAX_OUTPUT_LENGTH|-} {DAMPING|-} {SEED|-} [WEIGHT~]MID_FILE[=TRACKS] [...]

where:

• LEFT_CONTEXT stands for the number of previous MIDI commands that are kept as the left context when building the Markov chain. (Default: 3)

• MAX_OUTPUT_LENGTH specifies the number of lines after which the output is cut off. (Default: Infinity)

• DAMPING specifies the damping factor. 1-DAMPING is the probability that the random walker will move to a completely random node. (Default: 1)

• MID_FILE is the pathname to an input standard MIDI file. If the input file names contain ~ or =, you will need to explicitly specify 1~MID_FILE=* to prevent misparsing.

• WEIGHT how much weight will the specified tracks from the song carry within the produced Markov chain. (Default: 1)

• SEED is an integer seed for the random number generator. Supposing all other parameters remain constant, supplying the same seed will produce the same song. (Default: The current UNIX time)

• TRACKS are used to specify the MIDI tracks to use from the given input file. (Default: *). The parameter uses the following syntax:

  <root>  ::=  <expr> | '*'
  <expr>  ::=  <expr>,<expr> | <range> | <atom>
  <range> ::=  <atom>-<atom>
  <atom>  ::=  [0-9]+
  

The script dumps the generated song to the track.csv CSV file accepted by the csvmidi binary, and to the stdout as a standard MIDI file.

License

MIT