Akarsh-scratch.txt

LOOKING INTO A MIDI FILE FOR REPRESENTATION:
============================================

* To find the tonal range of a track:

mididump.py /tmp/bwv1041a-solo.mid  | grep midi.NoteO.*Event | sed 's/.*data=\[\(.*\)\].*$/\1/' | awk -F',' '{ print $1 }' | sort -n | uniq

* Find possible note velocities:

mididump.py /tmp/bwv1041a-solo.mid  | grep midi.NoteO.*Event | sed 's/.*data=\[\(.*\)\].*$/\1/' | awk -F',' '{ print $2 }' | sort -n | uniq

* Find possible tick intervals:

mididump.py /tmp/bwv1041a-solo.mid  | grep midi.NoteO.*Event | sed 's/.*tick=\([0-9]*\).*$/\1/' | sort -n | uniq

* Count the number of 0 tick interals:

mididump.py /tmp/bwv1041a-solo.mid  | grep midi.NoteO.*Event | sed 's/.*tick=\([0-9]*\).*$/\1/' | sort -n | grep 0 | wc -l

CASE OF BWV1041A-SOLO.MID:
==========================

* Pruned down to one track

* Ignore any chords on the violin

* => 33 note range (including non-used ones for simplicity), plus OFF state = 34 states for 1-grams

* Least tick resolution = 60. i.e. one "unit" of time is 60 ticks. Approximate anything else by nearest multiple of 60.

* Largest tick count = 1080 => max of 18 units.

* Lowest note = 56

OUTLINE OF THE MIDI -> UNIGRAMS CODE:
=====================================

For now, we will keep it simple and assume single channel, no polyphony

+ Specify the following constants:
  1. FILE_NAME: Name of the MIDI file (=bwv1041a-solo.mid for example)
  2. BASE_TICK: Time unit in ticks (=60 for bwv1041a-solo.mid)
  3. N_NOTES: Number of notes used in the track (=32 for bwv1041a-solo.mid)
  4. LOWEST_NOTE: Number corresponding to the lowest note (=56 for bwv1041a-solo.mid)

+ Read the states from the file
  
  Because of no polyphony, each NoteOffEvent == set state to 0 if current state + LOWEST_NOTE - 1 matches the note given in the NoteOffEvent. Otherwise, ignore.
  
  Each NoteOnEvent == set state to data[0] - LOWEST_NOTE + 1

  Divide number of ticks by BASE_TICK, round to nearest int, and write the state into the file those many times

OUTLINE OF UNIGRAMS -> N-GRAMS CODE:
====================================

+ Specify the following constants:
  1. FILE_NAME: Name of the State List file
  2. N_UNIGRAMS: Number of possible unigram states
  3. N_GRAMS: n in Markov n-grams (=2 for starters)

+ Compute state space size 'N'

  N_STATES = (N_UNIGRAMS)^(N_GRAMS)

+ Create an N-gram state variable, initialize to 0

  unsigned long state = 0 # Hopefully, will not overflow!!!

+ Read unigram file and add to the n-gram state

  state *= N_UNIGRAMS
  state += unigram
  state %= N_STATES

  Write the n-gram state