Skip to content

Instantly share code, notes, and snippets.

@zhangzhishan
Last active April 27, 2020 10:08
Show Gist options
  • Save zhangzhishan/bda71f5a33c53e95af5d to your computer and use it in GitHub Desktop.
Save zhangzhishan/bda71f5a33c53e95af5d to your computer and use it in GitHub Desktop.
import os
import os.path
import re
import eyed3
import urllib2
import urllib
from urllib import urlencode
import sys
import os
reload(sys)
sys.setdefaultencoding('utf8')
music_path = r"E:\music"
lrc_path = r"e:\lrc"
os.remove('nolrc.txt')
os.remove('lrcxml.txt')
the_file = open('lrcxml.txt','a')
nolrc_file = open('nolrc.txt','a')
for root,dirs,files in os.walk(music_path):
for filepath in files:
the_path = os.path.join(root,filepath)
if (the_path.find("mp3") != -1):
print the_path
the_music = eyed3.load(the_path)
the_teg = the_music.tag._getAlbum()
the_artist = the_music.tag._getArtist()
the_title = the_music.tag._getTitle()
# print the_teg
# print the_title
# print the_artist
b = the_title.replace(' ','+')
# print b
a = the_artist.replace(' ','+')
#print urlencode(str(b))
if isinstance(a,unicode):
a = a.encode('utf8')
song_url = "http://box.zhangmen.baidu.com/x?op=12&count=1&title="+b+"$$"+a+"$$$$ "
the_file.write(song_url+'\n')
page = urllib2.urlopen(song_url).read()
print page
theid = 0
lrcid = re.compile('<lrcid>(.*?)</lrcid>',re.S).findall(page)
have_lrc = True
if lrcid != []:
theid = lrcid[0]
else:
nolrc_file.write(the_title+'\n')
have_lrc = False
print theid
if have_lrc:
firstid = int(theid)/100
lrcurl = "http://box.zhangmen.baidu.com/bdlrc/"+str(firstid)+"/"+theid+".lrc"
print lrcurl
lrc = urllib2.urlopen(lrcurl).read()
if(lrc.find('html')== -1):
lrcfile = open(lrc_path+"\\"+the_title+".lrc",'w')
lrcfile.writelines(lrc)
lrcfile.close()
else:
nolrc_file.write(the_title+'\n')
the_file.close()
nolrc_file.close()
print "end!"
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment