python版本网易云音乐ncm文件格式转换(mirror from https://github.com/nondanee/ncmdump)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

138 lines
3.8 KiB

# -*- coding: utf-8 -*-
"""
Created on Sun Jul 15 01:05:58 2018
@author: Nzix
"""
import binascii
import struct
import base64
import json
import os
from Crypto.Cipher import AES
from mutagen import mp3, flac, id3
def dump(input_path, output_path = None, skip = True):
output_path = (lambda path, meta : os.path.splitext(path)[0] + '.' + meta['format']) if not output_path else output_path
core_key = binascii.a2b_hex('687A4852416D736F356B496E62617857')
meta_key = binascii.a2b_hex('2331346C6A6B5F215C5D2630553C2728')
unpad = lambda s : s[0:-(s[-1] if type(s[-1]) == int else ord(s[-1]))]
f = open(input_path,'rb')
# magic header
header = f.read(8)
assert binascii.b2a_hex(header) == b'4354454e4644414d'
# key data
f.seek(2, 1)
key_length = f.read(4)
key_length = struct.unpack('<I', bytes(key_length))[0]
key_data = bytearray(f.read(key_length))
key_data = bytes(bytearray([byte ^ 0x64 for byte in key_data]))
cryptor = AES.new(core_key, AES.MODE_ECB)
key_data = unpad(cryptor.decrypt(key_data))[17:]
key_length = len(key_data)
# S-box (standard RC4 Key-scheduling algorithm)
key = bytearray(key_data)
S = bytearray(range(256))
j = 0
for i in range(256):
j = (j + S[i] + key[i % key_length]) % 256
S[i], S[j] = S[j], S[i]
# meta data
meta_length = f.read(4)
meta_length = struct.unpack('<I', bytes(meta_length))[0]
meta_data = bytearray(f.read(meta_length))
meta_data = bytes(bytearray([byte ^ 0x63 for byte in meta_data]))
meta_data = base64.b64decode(meta_data[22:])
cryptor = AES.new(meta_key, AES.MODE_ECB)
meta_data = unpad(cryptor.decrypt(meta_data)).decode('utf-8')[6:]
meta_data = json.loads(meta_data)
# crc32
crc32 = f.read(4)
crc32 = struct.unpack('<I', bytes(crc32))[0]
# album cover
f.seek(5, 1)
image_size = f.read(4)
image_size = struct.unpack('<I', bytes(image_size))[0]
image_data = f.read(image_size)
# media data
output_path = output_path(input_path, meta_data)
if skip and os.path.exists(output_path): return
m = open(output_path,'wb')
data = bytearray(f.read())
# stream cipher (modified RC4 Pseudo-random generation algorithm)
i = 0
j = 0
for k, _ in enumerate(data):
i = (i + 1) % 256
j = (i + S[i]) % 256 # in RC4, is j = (j + S[i]) % 256
# S[i], S[j] = S[j], S[i] # no swapping
data[k] ^= S[(S[i] + S[j]) % 256]
m.write(data)
m.close()
f.close()
# media tag
if meta_data['format'] == 'flac':
audio = flac.FLAC(output_path)
# audio.delete()
image = flac.Picture()
image.type = 3
image.mime = 'image/jpeg'
image.data = image_data
audio.clear_pictures()
audio.add_picture(image)
elif meta_data['format'] == 'mp3':
audio = mp3.MP3(output_path)
# audio.delete()
image = id3.APIC()
image.type = 3
image.mime = 'image/jpeg'
image.data = image_data
audio.tags.add(image)
audio.save()
audio = mp3.EasyMP3(output_path)
audio['title'] = meta_data['musicName']
audio['album'] = meta_data['album']
audio['artist'] = '/'.join([artist[0] for artist in meta_data['artist']])
audio.save()
return output_path
if __name__ == '__main__':
import sys
if len(sys.argv) > 1:
files = sys.argv[1:]
else:
files = [name for name in os.listdir('.') if os.path.splitext(name)[-1] == '.ncm']
if sys.version[0] == '2':
files = [path.decode(sys.stdin.encoding) for path in files]
if not files:
print('please input file path!')
for path in files:
try:
dump(path)
print(os.path.split(path)[-1])
except Exception as e:
print(e)
pass