#!/usr/bin/env python
from __future__ import division
import ipdb
import os
import random
import sys
import time
import parse
import topsort
import yaml
import tree
random.seed(time.time())
def main():
composition = {
"fm_test": {
"intro": {
"melody": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 2 6 5 1",
"octave": 8,
"duration": 10,
"grammars": { # Notes for this instrument to use in this piece
"u": ["G/2 G/2 | G/4 G/4 A/4 A/4 | A/2 A/2 | G | G | A | A | A3 (w)"],
"w": ["E | E | F | F | G/2 G/2 | G3 (u)"],
},
},
},
},
"verse1": {
"intro": {
"melody": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 1",
"octave": 8,
"duration": 10,
"grammars": { # Notes for this instrument to use in this piece
"u": ["G/2 G/2 | G/4 G/4 A/4 A/4 | A/2 A/2 | G | G | A | A | A3 (w)"],
"w": ["E | E | F | F | G/2 G/2 | G3 (u)"],
},
},
},
"body": {
"melody": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 1",
"octave": 8,
"duration": 10,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C | G/2 G/2 | G/2 G/2 | C | B, | F' | C | F | C | B | F | (w)"],
"w": ["E/4 A/4 D/4 G/4 | F/4 F/4 B2 | (u)"],
},
},
},
"outro": {
"melody": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 1",
"octave": 8,
"duration": 10,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C/4 C/4 C/4 C/4 | z2"],
},
},
},
},
"verse2": {
"body": {
"melody": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 1",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C | C | C | C | F/2 F/2 | F/2 F/2 | (u)", "D | D | G/2 A/2 | D | D | (u)"],
},
},
"harmony": { # Instrument 'melody'
"score_line": "i3 %(time)f %(duration)f 4000 %(octave)d.%(note)s 2 3 5 3",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C | C | z | C | C | z/2 F/4 F/2 F/2 | F/2 F/2 | z (u)", "D | D | G/2 A/2 | D | D | z (u)"],
},
},
"percussion": { # Instrument 'melody'
"score_line": "i1 %(time)f %(duration)f 7000 %(octave)d.%(note)s %(octave)d.%(note)s 0 6",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C/4 C/4 C/4 C/4 | F/2 F/2 | F/2 F/2 | (u)", "D/4 D/4 G/4 A/4 | D | D | (v)"],
"v": ["C | D | E | F | E | D | C | (u)",],
},
},
},
"outro": {
"percussion": { # Instrument 'melody'
"score_line": "i1 %(time)f %(duration)f 7000 %(octave)d.%(note)s %(octave)d.%(note)s 0 6",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["C/4 C/4 C/4 C/4"],
},
},
},
},
"sync_test": {
"body": {
"lead_instr": { # Instrument 'melody'
"score_line": "i1 %(time)f %(duration)f 7000 %(octave)d.%(note)s %(octave)d.%(note)s 0 6",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["D/4 D/4 D/4 D/4 (v)"],
"v": ["C/4 C/4 C/4 C/4 (u)"],
},
},
"follow_instr": { # Instrument 'melody'
"score_line": "i2 %(time)f %(duration)f 7000 %(octave)d.%(note)s 1",
"sync": "lead_instr",
"octave": 8,
"duration": 30,
"grammars": { # Notes for this instrument to use in this piece
"u": ["D/4 D/4 D/4 D/4 (u)"],
"v": ["C/4 C/4 C/4 C/4 (v)"],
},
},
},
},
}
print '''f1 0 512 10 1
f2 0 8192 10 .24 .64 .88 .76 .06 .5 .34 .08
f3 0 1025 10 1
t 0 100
'''
section_start = 0
# for section in ["verse1", "verse2"]:
for section in ["sync_test"]:
print "; Section " + section
subsection_start = section_start
section = composition[section]
for subsection in ["intro", "body", "outro"]:
try:
print "; Subsection " + subsection
subsection = section[subsection]
unordered_instrs = []
for instr in subsection:
subsection[instr]["name"] = instr
if not "sync" in subsection[instr].keys():
subsection[instr]["sync"] = None
unordered_instrs.append([subsection[instr]["sync"], instr])
ordered_instrs = topsort.topsort(unordered_instrs)
ordered_instrs.remove(None) # None used as a placeholder for sort order for instruments with no sync setting
instrs = []
syncs = {}
for instr in ordered_instrs:
print ";Instrument " + instr
instr = subsection[instr]
max_time = instr["duration"]
instr_score, syncs = render_instr(instr, syncs, max_time)
instrs.append(instr_score)
for line in generate_csound_score(instr_score, instr["score_line"], subsection_start):
print line
longest_score = max(instrs, key=lambda i: score_len(i))
subsection_start += score_len(longest_score)
section_start += score_len(longest_score)
except KeyError:
pass
def render_instr(instr, syncs, max_time):
for g in instr["grammars"]:
for i in range(len(instr["grammars"][g])):
instr["grammars"][g][i] = parse.parse(instr["grammars"][g][i])
score= []
try:
score, syncs = choose_phrase(instr, syncs, 0, max_time, None)
while True:
score_index_to_replace = None
for item in range(len(score)): # Optimize this by caching the index of the last node I replaced and startng there
if isinstance(score[item], tree.Tree): # Also, make this use the find_next_node() function (or whatever I called it)
score_index_to_replace = item
if score_index_to_replace is None:
raise ValueError("No more nodes to fill in")
time_remaining = max_time - score_len(score)
new_phrase, syncs = choose_phrase(instr, syncs, score_len(score), time_remaining, score)
score = score[:score_index_to_replace-1] + new_phrase + score[score_index_to_replace+1:]
except ValueError:
return (score, syncs)
def choose_phrase(instr, syncs, current_time, time_remaining, score):
'''Filters grammars for ones that match the sync option, and phrases that fit the time remaining in the score'''
time_filtered_grammars = {}
for grammar in instr["grammars"]:
fitting_phrases = get_phrases_that_fit(instr["grammars"][grammar], time_remaining)
if len(fitting_phrases) > 0:
time_filtered_grammars[grammar] = fitting_phrases
if len(time_filtered_grammars.keys()) == 0:
raise ValueError("No available grammars that will fit in the score")
grammar = None
if instr["sync"] is not None:
guiding_instr = instr["sync"]
sync_node = get_sync_node_at_time(syncs[guiding_instr], current_time)
if sync_node in time_filtered_grammars.keys():
grammar = sync_node
else:
grammar = random.choice(time_filtered_grammars.keys())
if score is None:
grammar = random.choice(time_filtered_grammars.keys())
elif instr["sync"] is None:
grammar = get_next_node(score);
phrases = time_filtered_grammars[grammar]
if instr["name"] not in syncs.keys():
syncs[instr["name"]] = []
syncs[instr["name"]].append({"node": grammar, "time": current_time})
return random.choice(phrases), syncs
def get_phrases_that_fit(grammar, time_remaining):
valid_phrases = []
for phrase in grammar:
if score_len(phrase) <= time_remaining:
valid_phrases.append(phrase)
return valid_phrases
def get_sync_node_at_time(syncs, t):
for s in range(len(syncs)):
if syncs[s]["time"] >= t:
return syncs[s]["node"]
def get_next_node(score):
for token in score:
if isinstance(token, tree.Tree):
return token.name
def score_len(score):
total = 0
for n in score:
if not isinstance(n, tree.Tree):
total += n.duration
return total
def generate_csound_score(score, score_line, t):
csound_note_values = {
"C": "00",
"C#": "01",
"D": "02",
"D#": "03",
"E": "04",
"F": "05",
"F#": "06",
"G": "07",
"G#": "08",
"A": "09",
"A#": "10",
"B": "11",
}
csound_score = []
for token in score:
if isinstance(token, parse.Chord): # Chords
for note in token.chord:
note = csound_note_values[note]
csound_score.append(score_line % {"time": t, "octave": token.octave, "note": note, "duration": token.duration})
elif isinstance(token, parse.Note): # Individual notes
note = csound_note_values[token.value]
csound_score.append(score_line % {"time": t, "octave": token.octave, "note": note, "duration": token.duration})
elif isinstance(token, tree.Tree):
continue
t += token.duration
return csound_score
if __name__ == "__main__": main()