8bitworkshop/tools/midi2song.py

#!/usr/bin/python

import sys, string, math, argparse
import mido

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

# for 2600
#max_voices = 2
#coutput = 0
# for 2600 wavetable
#max_voices = 4
#one_voice_per_channel = 0

fn = args.midifile

mid = mido.MidiFile(fn)

def hex1(n):
    return '%02x'%n
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)
    for k,v in list(ss.items()):
      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]
    print((i,l,n,count,repr(match)))
    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
    print((len(s), len(r)+n+l))
    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)

if not args.midichannels:
    print(mid)
    print((mid.length, 'seconds'))
    for i, track in enumerate(mid.tracks):
        print(('Track {}: {} ({}) {}'.format(i, track.name, len(track), channels_for_track(track))))
        #for msg in track:
        #    print(msg)
else:
    gtime = 0
    curtime = 0
    nnotes = 0
    nvoices = 0
    curchans = 0
    channels = [int(x) for x in args.midichannels.split(',')]
    print ('')
    print(("// %s %s" % (mid, channels)))
    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)
    if coutput:
        print((','.join([hex2(x) for x in output])))
    else:
        bighex = ''.join([hex1(x) for x in output])
        for i in range(0,len(bighex)+32,32):
            print(('\thex', bighex[i:i+32]))
    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)
          print(bss)
          if bss[1] == 0:
              break
          bout = replace_substrings(bout, bss)
          print((repr(bout)))
        # 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
        print((','.join([hex2(ord(x)) for x in zout])))
        print(("// compressed %d -> %d bytes" % (len(output), len(zout))))