1
0
mirror of https://github.com/sehugg/8bitworkshop.git synced 2024-11-25 18:33:11 +00:00
8bitworkshop/tools/midi2song.py

183 lines
5.4 KiB
Python
Raw Normal View History

2017-03-31 17:25:18 +00:00
#!/usr/bin/python
2017-04-12 02:46:06 +00:00
import sys, string, math, argparse
2017-03-31 17:25:18 +00:00
import mido
2017-04-12 02:46:06 +00:00
parser = argparse.ArgumentParser()
parser.add_argument('-s', '--start', type=int, default=21, help="first MIDI note")
parser.add_argument('-n', '--num', type=int, default=21+63, help="number of notes")
parser.add_argument('-v', '--voices', type=int, default=3, help="number of voices")
parser.add_argument('-T', '--transpose', type=int, default=0, help="transpose by half-steps")
parser.add_argument('-t', '--tempo', type=int, default=48, help="tempo")
parser.add_argument('-o', '--one', action="store_true", help="one voice per channel")
parser.add_argument('-z', '--compress', action="store_true", help="compress song (experimental)")
parser.add_argument('-H', '--hex', action="store_true", help="hex output")
parser.add_argument('midifile', help="MIDI file")
parser.add_argument('midichannels', nargs='?', help="comma-separated list of MIDI channels, or -")
args = parser.parse_args()
min_note = args.start
max_note = min_note + args.num
max_voices = args.voices
one_voice_per_channel = args.one
tempo = args.tempo
compress = args.compress
transpose = args.transpose
coutput = not args.hex
2017-04-09 13:33:38 +00:00
# for 2600
#max_voices = 2
#coutput = 0
# for 2600 wavetable
#max_voices = 4
#one_voice_per_channel = 0
2017-03-31 17:25:18 +00:00
2017-04-12 02:46:06 +00:00
fn = args.midifile
2017-03-31 17:25:18 +00:00
mid = mido.MidiFile(fn)
2017-04-09 13:33:38 +00:00
def hex1(n):
return '%02x'%n
2017-03-31 17:25:18 +00:00
def hex2(n):
return '0x%02x'%n
def only_notes(s):
for ch in s:
if ord(ch) == 0xff:
return False
return True
def find_common_substrings(s):
results = {}
for l in range(64, 6, -1):
for i in range(0,len(s)-l*2):
match = s[i:i+l]
if not only_notes(match):
continue
count = 0
j = i+l
while j < len(s):
p = s.find(match, j)
if p > 0:
count += 1
j = p+l
else:
break
if count:
n = count*(l-1)-1
if not results.get(i) or n > results[i][0]:
results[i] = (n,l)
return results
def get_best_substring(ss):
best = (0,0,0)
2018-11-22 19:11:56 +00:00
for k,v in list(ss.items()):
2017-03-31 17:25:18 +00:00
if v[0] > best[2]:
best = (k,v[1],v[0])
return best
def offset2str(ofs):
return chr(ofs & 0xff) + chr((ofs >> 8) & 0xff)
#return chr(0xc0 | (ofs & 0x3f)) + chr(0xc0 | ((ofs >> 6) & 0x3f))
g_code = 0xc1
g_subs = []
def replace_substrings(s, bss):
global g_code
i,l,n = bss
count = (n+1)/(l-1)
match = s[i:i+l]
2018-11-22 19:11:56 +00:00
print((i,l,n,count,repr(match)))
2017-03-31 17:25:18 +00:00
r = s[0:i]
while i<len(s):
r += chr(g_code)
p = s.find(match,i+l)
if p < 0:
p = len(s)
r += s[i+l:p]
i = p
g_subs.append(match + chr(0xff))
g_code += 1
2018-11-22 19:11:56 +00:00
print((len(s), len(r)+n+l))
2017-03-31 17:25:18 +00:00
assert len(s) == len(r)+n+l
return r
def channels_for_track(track):
channels = set()
for msg in track:
if msg.type == 'note_on':
channels.add(msg.channel)
return list(channels)
2017-04-12 02:46:06 +00:00
if not args.midichannels:
2018-11-22 19:11:56 +00:00
print(mid)
print((mid.length, 'seconds'))
2017-03-31 17:25:18 +00:00
for i, track in enumerate(mid.tracks):
2018-11-22 19:11:56 +00:00
print(('Track {}: {} ({}) {}'.format(i, track.name, len(track), channels_for_track(track))))
2017-03-31 17:25:18 +00:00
#for msg in track:
# print(msg)
else:
gtime = 0
curtime = 0
nnotes = 0
nvoices = 0
curchans = 0
2018-11-22 19:11:56 +00:00
channels = [int(x) for x in args.midichannels.split(',')]
print ('')
print(("// %s %s" % (mid, channels)))
2017-03-31 17:25:18 +00:00
output = []
for msg in mid:
gtime += msg.time * tempo
if msg.type == 'note_on' and msg.channel in channels:
note = msg.note + transpose
vel = msg.velocity
t = int(math.ceil(gtime))
if vel > 0:
while curtime < t:
dt = min(63, t-curtime)
curtime += dt
if nnotes > 0:
nvoices = 0
curchans = 0
output.append(dt+128)
if note >= min_note and note <= max_note and nvoices < max_voices:
if not (one_voice_per_channel and (curchans & (1<<msg.channel))):
n = note - min_note
output.append(n)
nnotes += 1
nvoices += 1
curchans |= 1<<msg.channel
output.append(0xff)
2017-04-09 13:33:38 +00:00
if coutput:
2018-11-22 19:11:56 +00:00
print((','.join([hex2(x) for x in output])))
2017-04-09 13:33:38 +00:00
else:
2018-11-22 19:11:56 +00:00
bighex = ''.join([hex1(x) for x in output])
2017-04-09 13:33:38 +00:00
for i in range(0,len(bighex)+32,32):
2018-11-22 19:11:56 +00:00
print(('\thex', bighex[i:i+32]))
2017-03-31 17:25:18 +00:00
if compress:
# replace common substrings
bout = ''.join(chr(i) for i in output)
for iter in range(0,32):
ss = find_common_substrings(bout)
bss = get_best_substring(ss)
2018-11-22 19:11:56 +00:00
print(bss)
2017-03-31 17:25:18 +00:00
if bss[1] == 0:
break
bout = replace_substrings(bout, bss)
2018-11-22 19:11:56 +00:00
print((repr(bout)))
2017-03-31 17:25:18 +00:00
# build substring table
ofs = (len(g_subs)+1)*2
zout = offset2str(ofs)
ofs += len(bout)
for ss in g_subs:
ofs += len(ss)
zout += offset2str(ofs)
# output strings
zout += bout
for ss in g_subs:
zout += ss
# print output
2018-11-22 19:11:56 +00:00
print((','.join([hex2(ord(x)) for x in zout])))
print(("// compressed %d -> %d bytes" % (len(output), len(zout))))