Change file structure
This commit is contained in:
parent
dcd275e3b1
commit
71e967808b
22
main.py
22
main.py
@ -4,14 +4,14 @@
|
||||
# author: David-123
|
||||
|
||||
|
||||
from modules.inputs import rinput
|
||||
from modules.information import print_info
|
||||
from modules.multi_download import mdl
|
||||
from modules.one_download import download_one_lyric
|
||||
from modules.settings import settings_menu
|
||||
from modules.save_load_settings import load_settings
|
||||
from modules.clear_screen import clear
|
||||
from modules.load_file_song import get_lyric_from_folder
|
||||
from modules.utils.inputs import rinput
|
||||
from modules.utils.information import print_info
|
||||
from modules.functions.multi_download import mdl
|
||||
from modules.functions.one_download import download_one_lyric
|
||||
from modules.submenus.settings import settings_menu
|
||||
from modules.functions.save_load_settings import load_settings
|
||||
from modules.utils.clear_screen import clear
|
||||
from modules.functions.load_file_song import get_lyric_from_folder
|
||||
|
||||
|
||||
class MainProcess(object):
|
||||
@ -30,11 +30,11 @@ class MainProcess(object):
|
||||
r = rinput("请选择:")
|
||||
|
||||
if r == "1":
|
||||
download_one_lyric(self.settings.lyric_path)
|
||||
download_one_lyric(self)
|
||||
elif r == "2":
|
||||
mdl(self.settings.lyric_path)
|
||||
mdl(self)
|
||||
elif r == "3":
|
||||
get_lyric_from_folder(self.settings.lyric_path)
|
||||
get_lyric_from_folder(self)
|
||||
elif r == "0":
|
||||
exit(0)
|
||||
elif r == "i":
|
||||
|
0
modules/functions/__init__.py
Normal file
0
modules/functions/__init__.py
Normal file
@ -1,298 +1,306 @@
|
||||
import binascii
|
||||
import json
|
||||
import os
|
||||
import struct
|
||||
from base64 import b64decode
|
||||
from multiprocessing import Process, Queue
|
||||
from queue import Empty
|
||||
|
||||
from progress.bar import Bar
|
||||
from Cryptodome.Cipher import AES
|
||||
from mutagen import File, flac
|
||||
from mutagen.id3 import ID3, TPE1, APIC, COMM, TIT2, TALB
|
||||
|
||||
from modules.clear_screen import clear
|
||||
from modules.get_song import get_song_lyric
|
||||
from modules.inputs import cinput, rinput
|
||||
|
||||
|
||||
def load_information_from_song(path):
|
||||
"""从音乐文件中的 Comment 字段获取 163 key 并解密返回歌曲信息"""
|
||||
file = File(path) # 使用 mutagen 获取歌曲信息
|
||||
if os.path.splitext(path)[-1] == ".mp3": # 当文件为 mp3 时使用 ID3 格式读取
|
||||
if file.tags.get("COMM::XXX"):
|
||||
if file.tags["COMM::XXX"].text[0][:7] == "163 key":
|
||||
ciphertext = file.tags["COMM::XXX"].text[0][22:]
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
elif os.path.splitext(path)[-1] == ".flac": # 当文件为 flac 时使用 FLAC 格式读取
|
||||
if file.tags.get("DESCRIPTION"):
|
||||
if file.tags["DESCRIPTION"][0][:7] == "163 key":
|
||||
ciphertext = file.tags["DESCRIPTION"][0][22:]
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
|
||||
def unpad(s): # 创建清理针对于网易云的 AES-128-ECB 解密后末尾占位符的函数
|
||||
if type(s[-1]) == int:
|
||||
end = s[-1]
|
||||
else:
|
||||
end = ord(s[-1])
|
||||
return s[0:-end]
|
||||
# return s[0:-(s[-1] if type(s[-1]) == int else ord(s[-1]))] 更加清晰的理解 ↑
|
||||
|
||||
cryptor = AES.new(b"#14ljk_!\\]&0U<'(", AES.MODE_ECB) # 使用密钥创建解密器
|
||||
|
||||
# 下方这一行将密文 ciphertext 转换为 bytes 后进行 base64 解码, 得到加密过的 AES 密文
|
||||
# 再通过上方创建的 AES 128-ECB 的解密器进行解密, 然后使用 unpad 清除末尾无用的占位符后得到结果
|
||||
try:
|
||||
r = unpad((cryptor.decrypt(b64decode(bytes(ciphertext, "utf-8"))).decode("utf-8")))
|
||||
except ValueError:
|
||||
return "decrypt_failed"
|
||||
|
||||
if r:
|
||||
if r[:5] == "music":
|
||||
return json.loads(r[6:])
|
||||
else:
|
||||
return "not_a_normal_music"
|
||||
else:
|
||||
return "decrypt_failed"
|
||||
|
||||
|
||||
def load_and_decrypt_from_ncm(file_path, targetdir): # nondanee的源代码, 根据需求更改了某些东西
|
||||
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(file_path, 'rb')
|
||||
header = f.read(8)
|
||||
assert binascii.b2a_hex(header) == b'4354454e4644414d'
|
||||
f.seek(2, 1)
|
||||
key_length = f.read(4)
|
||||
key_length = struct.unpack('<I', bytes(key_length))[0]
|
||||
key_data = f.read(key_length)
|
||||
key_data_array = bytearray(key_data)
|
||||
for i in range(0, len(key_data_array)):
|
||||
key_data_array[i] ^= 0x64
|
||||
key_data = bytes(key_data_array)
|
||||
cryptor = AES.new(core_key, AES.MODE_ECB)
|
||||
key_data = unpad(cryptor.decrypt(key_data))[17:]
|
||||
key_length = len(key_data)
|
||||
key_data = bytearray(key_data)
|
||||
key_box = bytearray(range(256))
|
||||
c = 0
|
||||
last_byte = 0
|
||||
key_offset = 0
|
||||
for i in range(256):
|
||||
swap = key_box[i]
|
||||
c = (swap + last_byte + key_data[key_offset]) & 0xff
|
||||
key_offset += 1
|
||||
if key_offset >= key_length:
|
||||
key_offset = 0
|
||||
key_box[i] = key_box[c]
|
||||
key_box[c] = swap
|
||||
last_byte = c
|
||||
meta_length = f.read(4)
|
||||
meta_length = struct.unpack('<I', bytes(meta_length))[0]
|
||||
meta_data = f.read(meta_length)
|
||||
meta_data_array = bytearray(meta_data)
|
||||
for i in range(0, len(meta_data_array)):
|
||||
meta_data_array[i] ^= 0x63
|
||||
meta_data = bytes(meta_data_array)
|
||||
comment = meta_data
|
||||
meta_data = 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 = f.read(4)
|
||||
crc32 = struct.unpack('<I', bytes(crc32))[0]
|
||||
f.seek(5, 1)
|
||||
image_size = f.read(4)
|
||||
image_size = struct.unpack('<I', bytes(image_size))[0]
|
||||
image_data = f.read(image_size)
|
||||
file_name = f.name.split("/")[-1].split(".ncm")[0] + '.' + meta_data['format']
|
||||
m = open(os.path.join(targetdir, file_name), 'wb')
|
||||
chunk = bytearray()
|
||||
while True:
|
||||
chunk = bytearray(f.read(0x8000))
|
||||
chunk_length = len(chunk)
|
||||
if not chunk:
|
||||
break
|
||||
for i in range(1, chunk_length + 1):
|
||||
j = i & 0xff
|
||||
chunk[i - 1] ^= key_box[(key_box[j] + key_box[(key_box[j] + j) & 0xff]) & 0xff]
|
||||
m.write(chunk)
|
||||
m.close()
|
||||
f.close()
|
||||
|
||||
# 对解密后的文件进行信息补全
|
||||
if meta_data["format"] == "mp3": # 针对 mp3 使用 ID3 进行信息补全
|
||||
audio = ID3(os.path.join(targetdir, os.path.splitext(file_path.split("/")[-1])[0] + ".mp3"))
|
||||
artists = []
|
||||
for i in meta_data["artist"]:
|
||||
artists.append(i[0])
|
||||
audio["TPE1"] = TPE1(encoding=3, text=artists) # 插入歌手
|
||||
audio["APIC"] = APIC(encoding=3, mime='image/jpg', type=3, desc='', data=image_data) # 插入封面
|
||||
audio["COMM::XXX"] = COMM(encoding=3, lang='XXX', desc='', text=[comment.decode("utf-8")]) # 插入 163 key 注释
|
||||
audio["TIT2"] = TIT2(encoding=3, text=[meta_data["musicName"]]) # 插入歌曲名
|
||||
audio["TALB"] = TALB(encoding=3, text=[meta_data["album"]]) # 插入专辑名
|
||||
audio.save()
|
||||
elif meta_data["format"] == "flac": # 针对 flac 使用 FLAC 进行信息补全
|
||||
audio = flac.FLAC(os.path.join(targetdir, os.path.splitext(file_path.split("/")[-1])[0] + ".flac"))
|
||||
artists = []
|
||||
for i in meta_data["artist"]:
|
||||
artists.append(i[0])
|
||||
audio["artist"] = artists[:] # 插入歌手
|
||||
audio["title"] = [meta_data["musicName"]] # 插入歌曲名
|
||||
audio["album"] = [meta_data["album"]] # 插入专辑名
|
||||
audio["description"] = comment.decode("utf-8") # 插入 163 key 注释
|
||||
audio.save()
|
||||
|
||||
return meta_data
|
||||
|
||||
|
||||
def process_work(path, filename, target, q_err: Queue, q_info: Queue):
|
||||
try:
|
||||
result = load_and_decrypt_from_ncm(path, target)
|
||||
except AssertionError:
|
||||
q_err.put(f"\t- 文件 \"{filename}\" 破解失败!")
|
||||
else:
|
||||
q_info.put(result)
|
||||
|
||||
|
||||
def get_lyric_from_folder(lyric_path: str):
|
||||
clear()
|
||||
path = cinput("请输入歌曲的保存文件夹(绝对路径):")
|
||||
if not os.path.exists(path):
|
||||
input("路径不存在.\n按回车返回...")
|
||||
return
|
||||
|
||||
print("正在遍历目录,请稍后...")
|
||||
musics = []
|
||||
ncm_files = []
|
||||
fails = 0
|
||||
for i in os.listdir(path): # 遍历目录,查找目标文件
|
||||
ext = os.path.splitext(i)[-1]
|
||||
if ext in ['.mp3', '.flac']: # 对于 mp3 和 flac 文件, 使用对应读取解密方式
|
||||
result = load_information_from_song(os.path.join(path, i))
|
||||
if result == "not_support":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 未包含 163 key ,跳过")
|
||||
elif result == "decrypt_failed":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 内 163 key 解密失败,跳过")
|
||||
elif result == "not_a_normal_music":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 内 163 key 不是一个普通音乐文件,这可能是一个电台曲目")
|
||||
else:
|
||||
musics.append({"id": result['musicId'], "name": result["musicName"], "artists": result["artist"]})
|
||||
elif ext == ".ncm": # 对于 ncm 先加入到列表,等待解密
|
||||
ncm_files.append(i)
|
||||
else:
|
||||
pass
|
||||
|
||||
if ncm_files:
|
||||
while True:
|
||||
print(f"\n发现{len(ncm_files)}个ncm加密文件!")
|
||||
print("请问解密后的文件保存在哪里?\n"
|
||||
"[1] 保存在相同文件夹内\n[2] 保存在程序设定的下载文件夹中\n[3] 保存在自定义文件夹内\n[q] 取消解密,下载歌词时将忽略这些文件")
|
||||
select = rinput("请选择: ")
|
||||
if select == 'q':
|
||||
target_path = "NOT_DECRYPT"
|
||||
break
|
||||
elif select == '1':
|
||||
target_path = path
|
||||
break
|
||||
elif select == '2':
|
||||
target_path = lyric_path
|
||||
break
|
||||
elif select == '3':
|
||||
target_path = input("请输入: ").strip()
|
||||
break
|
||||
else:
|
||||
print("输入无效!按回车继续...")
|
||||
|
||||
if target_path != "NOT_DECRYPT": # 开始进行逐个文件解密
|
||||
errors = [] # 初始化变量
|
||||
q_err = Queue() # 错误信息队列
|
||||
q_info = Queue() # 返回信息队列
|
||||
max_process = 20 # 最大进程数
|
||||
current_process = 0 # 当前正在活动的进程数
|
||||
passed = 0 # 总共结束的进程数
|
||||
with Bar("正在破解", max=len(ncm_files)) as bar:
|
||||
total = len(ncm_files)
|
||||
finished = 0 # 已经分配的任务数量
|
||||
while True: # 进入循环,执行 新建进程->检测队列->检测任务完成 的循环
|
||||
if current_process <= max_process and finished < total: # 分配进程
|
||||
Process(target=process_work,
|
||||
args=(os.path.join(path, ncm_files[finished]),
|
||||
ncm_files[finished],
|
||||
target_path,
|
||||
q_err,
|
||||
q_info)).start()
|
||||
finished += 1
|
||||
while True: # 错误队列检测
|
||||
try:
|
||||
errors.append(q_err.get_nowait())
|
||||
passed += 1 # 总任务完成数
|
||||
current_process -= 1 # 检测到进程完毕将进程-1
|
||||
bar.next() # 推动进度条
|
||||
fails += 1 # 错误数量+1
|
||||
except Empty:
|
||||
break
|
||||
while True: # 信息队列检测
|
||||
try:
|
||||
r = q_info.get_nowait()
|
||||
musics.append({"id": r['musicId'], "name": r["musicName"], "artists": r["artist"]})
|
||||
passed += 1
|
||||
current_process -= 1
|
||||
bar.next()
|
||||
except Empty:
|
||||
break
|
||||
if passed >= len(ncm_files):
|
||||
break
|
||||
if errors:
|
||||
print("解密过程中发现了以下错误:")
|
||||
for i in errors:
|
||||
print(i)
|
||||
|
||||
# 汇报索引结果
|
||||
print(f"\n索引完毕!共找到{fails + len(musics) + len(ncm_files)}个目标文件\n{len(musics)}个文件已载入\n{fails}个文件失败")
|
||||
if ncm_files:
|
||||
if target_path == "NOT_DECRYPT":
|
||||
print(f"{len(ncm_files)}个文件放弃加载")
|
||||
while True:
|
||||
print("\n你希望如何保存这些歌曲的歌词?\n[1]保存到刚刚输入的绝对路径中\n[2]保存到程序设定的下载路径中")
|
||||
r = rinput("请选择: ")
|
||||
if r == "1":
|
||||
lyric_path = path
|
||||
break
|
||||
elif r == "2":
|
||||
break
|
||||
else:
|
||||
try:
|
||||
input("无效选择, 若取消请按 ^C ,继续请按回车")
|
||||
clear()
|
||||
except KeyboardInterrupt:
|
||||
return
|
||||
|
||||
clear()
|
||||
for i in range(0, len(musics)): # 根据索引结果获取歌词
|
||||
print("\n进度: %d/%d" % (i + 1, len(musics)))
|
||||
if get_song_lyric(musics[i], lyric_path, allinfo=True) == "dl_err_connection":
|
||||
input("下载发生错误!可能是连接被拒绝!请检查网络后再试\n按回车键继续任务(该任务会被跳过)...")
|
||||
if ncm_files:
|
||||
if target_path != "NOT_DECRYPT":
|
||||
agree = rinput("是否删除原ncm文件? (y/n)")
|
||||
if agree == "y":
|
||||
for i in range(0, len(ncm_files)):
|
||||
print("删除进度: %d/%d\n -> %s\033[F" % (i + 1, len(ncm_files), ncm_files[i]), end="")
|
||||
os.remove(os.path.join(path, ncm_files[i]))
|
||||
else:
|
||||
print("取消.", end="")
|
||||
input("\n\033[K按回车返回...")
|
||||
return
|
||||
import binascii
|
||||
import json
|
||||
import os
|
||||
import struct
|
||||
from base64 import b64decode
|
||||
from multiprocessing import Process, Queue
|
||||
from queue import Empty
|
||||
|
||||
from progress.bar import Bar
|
||||
from Cryptodome.Cipher import AES
|
||||
from mutagen import File, flac
|
||||
from mutagen.id3 import ID3, TPE1, APIC, COMM, TIT2, TALB
|
||||
|
||||
from modules.utils.clear_screen import clear
|
||||
from modules.functions.get_song import get_song_lyric
|
||||
from modules.utils.inputs import cinput, rinput
|
||||
|
||||
|
||||
def load_information_from_song(path):
|
||||
"""从音乐文件中的 Comment 字段获取 163 key 并解密返回歌曲信息"""
|
||||
file = File(path) # 使用 mutagen 获取歌曲信息
|
||||
if os.path.splitext(path)[-1] == ".mp3": # 当文件为 mp3 时使用 ID3 格式读取
|
||||
if file.tags.get("COMM::XXX"):
|
||||
if file.tags["COMM::XXX"].text[0][:7] == "163 key":
|
||||
ciphertext = file.tags["COMM::XXX"].text[0][22:]
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
elif os.path.splitext(path)[-1] == ".flac": # 当文件为 flac 时使用 FLAC 格式读取
|
||||
if file.tags.get("DESCRIPTION"):
|
||||
if file.tags["DESCRIPTION"][0][:7] == "163 key":
|
||||
ciphertext = file.tags["DESCRIPTION"][0][22:]
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
else:
|
||||
return "not_support"
|
||||
|
||||
def unpad(s): # 创建清理针对于网易云的 AES-128-ECB 解密后末尾占位符的函数
|
||||
if type(s[-1]) == int:
|
||||
end = s[-1]
|
||||
else:
|
||||
end = ord(s[-1])
|
||||
return s[0:-end]
|
||||
# return s[0:-(s[-1] if type(s[-1]) == int else ord(s[-1]))] 更加清晰的理解 ↑
|
||||
|
||||
cryptor = AES.new(b"#14ljk_!\\]&0U<'(", AES.MODE_ECB) # 使用密钥创建解密器
|
||||
|
||||
# 下方这一行将密文 ciphertext 转换为 bytes 后进行 base64 解码, 得到加密过的 AES 密文
|
||||
# 再通过上方创建的 AES 128-ECB 的解密器进行解密, 然后使用 unpad 清除末尾无用的占位符后得到结果
|
||||
try:
|
||||
r = unpad((cryptor.decrypt(b64decode(bytes(ciphertext, "utf-8"))).decode("utf-8")))
|
||||
except ValueError:
|
||||
return "decrypt_failed"
|
||||
|
||||
if r:
|
||||
if r[:5] == "music":
|
||||
return json.loads(r[6:])
|
||||
else:
|
||||
return "not_a_normal_music"
|
||||
else:
|
||||
return "decrypt_failed"
|
||||
|
||||
|
||||
def load_and_decrypt_from_ncm(file_path, targetdir): # nondanee的源代码, 根据需求更改了某些东西
|
||||
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(file_path, 'rb')
|
||||
header = f.read(8)
|
||||
assert binascii.b2a_hex(header) == b'4354454e4644414d'
|
||||
f.seek(2, 1)
|
||||
key_length = f.read(4)
|
||||
key_length = struct.unpack('<I', bytes(key_length))[0]
|
||||
key_data = f.read(key_length)
|
||||
key_data_array = bytearray(key_data)
|
||||
for i in range(0, len(key_data_array)):
|
||||
key_data_array[i] ^= 0x64
|
||||
key_data = bytes(key_data_array)
|
||||
cryptor = AES.new(core_key, AES.MODE_ECB)
|
||||
key_data = unpad(cryptor.decrypt(key_data))[17:]
|
||||
key_length = len(key_data)
|
||||
key_data = bytearray(key_data)
|
||||
key_box = bytearray(range(256))
|
||||
c = 0
|
||||
last_byte = 0
|
||||
key_offset = 0
|
||||
for i in range(256):
|
||||
swap = key_box[i]
|
||||
c = (swap + last_byte + key_data[key_offset]) & 0xff
|
||||
key_offset += 1
|
||||
if key_offset >= key_length:
|
||||
key_offset = 0
|
||||
key_box[i] = key_box[c]
|
||||
key_box[c] = swap
|
||||
last_byte = c
|
||||
meta_length = f.read(4)
|
||||
meta_length = struct.unpack('<I', bytes(meta_length))[0]
|
||||
meta_data = f.read(meta_length)
|
||||
meta_data_array = bytearray(meta_data)
|
||||
for i in range(0, len(meta_data_array)):
|
||||
meta_data_array[i] ^= 0x63
|
||||
meta_data = bytes(meta_data_array)
|
||||
comment = meta_data
|
||||
meta_data = 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 = f.read(4)
|
||||
crc32 = struct.unpack('<I', bytes(crc32))[0]
|
||||
f.seek(5, 1)
|
||||
image_size = f.read(4)
|
||||
image_size = struct.unpack('<I', bytes(image_size))[0]
|
||||
image_data = f.read(image_size)
|
||||
file_name = f.name.split("/")[-1].split(".ncm")[0] + '.' + meta_data['format']
|
||||
m = open(os.path.join(targetdir, file_name), 'wb')
|
||||
chunk = bytearray()
|
||||
while True:
|
||||
chunk = bytearray(f.read(0x8000))
|
||||
chunk_length = len(chunk)
|
||||
if not chunk:
|
||||
break
|
||||
for i in range(1, chunk_length + 1):
|
||||
j = i & 0xff
|
||||
chunk[i - 1] ^= key_box[(key_box[j] + key_box[(key_box[j] + j) & 0xff]) & 0xff]
|
||||
m.write(chunk)
|
||||
m.close()
|
||||
f.close()
|
||||
|
||||
# 对解密后的文件进行信息补全
|
||||
if meta_data["format"] == "mp3": # 针对 mp3 使用 ID3 进行信息补全
|
||||
audio = ID3(os.path.join(targetdir, os.path.splitext(file_path.split("/")[-1])[0] + ".mp3"))
|
||||
artists = []
|
||||
for i in meta_data["artist"]:
|
||||
artists.append(i[0])
|
||||
audio["TPE1"] = TPE1(encoding=3, text=artists) # 插入歌手
|
||||
audio["APIC"] = APIC(encoding=3, mime='image/jpg', type=3, desc='', data=image_data) # 插入封面
|
||||
audio["COMM::XXX"] = COMM(encoding=3, lang='XXX', desc='', text=[comment.decode("utf-8")]) # 插入 163 key 注释
|
||||
audio["TIT2"] = TIT2(encoding=3, text=[meta_data["musicName"]]) # 插入歌曲名
|
||||
audio["TALB"] = TALB(encoding=3, text=[meta_data["album"]]) # 插入专辑名
|
||||
audio.save()
|
||||
elif meta_data["format"] == "flac": # 针对 flac 使用 FLAC 进行信息补全
|
||||
audio = flac.FLAC(os.path.join(targetdir, os.path.splitext(file_path.split("/")[-1])[0] + ".flac"))
|
||||
artists = []
|
||||
for i in meta_data["artist"]:
|
||||
artists.append(i[0])
|
||||
audio["artist"] = artists[:] # 插入歌手
|
||||
audio["title"] = [meta_data["musicName"]] # 插入歌曲名
|
||||
audio["album"] = [meta_data["album"]] # 插入专辑名
|
||||
audio["description"] = comment.decode("utf-8") # 插入 163 key 注释
|
||||
audio.save()
|
||||
|
||||
return meta_data
|
||||
|
||||
|
||||
def process_work(path, filename, target, q_err: Queue, q_info: Queue):
|
||||
try:
|
||||
result = load_and_decrypt_from_ncm(path, target)
|
||||
except AssertionError:
|
||||
q_err.put(f"\t- 文件 \"{filename}\" 破解失败!")
|
||||
else:
|
||||
q_info.put(result)
|
||||
|
||||
|
||||
def get_lyric_from_folder(self):
|
||||
global ncm_files_num
|
||||
clear()
|
||||
path = cinput(
|
||||
f"[NeteaseMusicLyricDownloader] {self.version}\n"
|
||||
"[自动获取]\n"
|
||||
"请输入歌曲的保存文件夹(绝对路径):")
|
||||
if not os.path.exists(path):
|
||||
input("路径不存在.\n按回车返回...")
|
||||
return
|
||||
|
||||
print("正在遍历目录,请稍后...")
|
||||
musics = []
|
||||
ncm_files = []
|
||||
fails = 0
|
||||
for i in os.listdir(path): # 遍历目录,查找目标文件
|
||||
ext = os.path.splitext(i)[-1]
|
||||
if ext in ['.mp3', '.flac']: # 对于 mp3 和 flac 文件, 使用对应读取解密方式
|
||||
result = load_information_from_song(os.path.join(path, i))
|
||||
if result == "not_support":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 未包含 163 key ,跳过")
|
||||
elif result == "decrypt_failed":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 内 163 key 解密失败,跳过")
|
||||
elif result == "not_a_normal_music":
|
||||
fails += 1
|
||||
print(f"文件 \"{i}\" 内 163 key 不是一个普通音乐文件,这可能是一个电台曲目")
|
||||
else:
|
||||
musics.append({"id": result['musicId'], "name": result["musicName"], "artists": result["artist"]})
|
||||
elif ext == ".ncm": # 对于 ncm 先加入到列表,等待解密
|
||||
ncm_files.append(i)
|
||||
else:
|
||||
pass
|
||||
|
||||
if ncm_files:
|
||||
while True:
|
||||
print(f"\n发现{len(ncm_files)}个ncm加密文件!")
|
||||
print("请问解密后的文件保存在哪里?\n"
|
||||
"[1] 保存在相同文件夹内\n[2] 保存在程序设定的下载文件夹中\n[3] 保存在自定义文件夹内\n[q] 取消解密,下载歌词时将忽略这些文件")
|
||||
select = rinput("请选择: ")
|
||||
if select == 'q':
|
||||
target_path = "NOT_DECRYPT"
|
||||
break
|
||||
elif select == '1':
|
||||
target_path = path
|
||||
break
|
||||
elif select == '2':
|
||||
target_path = self.settings.lyric_path
|
||||
break
|
||||
elif select == '3':
|
||||
target_path = input("请输入: ").strip()
|
||||
break
|
||||
else:
|
||||
print("输入无效!按回车继续...")
|
||||
|
||||
if target_path != "NOT_DECRYPT": # 开始进行逐个文件解密
|
||||
errors = [] # 初始化变量
|
||||
q_err = Queue() # 错误信息队列
|
||||
q_info = Queue() # 返回信息队列
|
||||
max_process = 20 # 最大进程数
|
||||
current_process = 0 # 当前正在活动的进程数
|
||||
passed = 0 # 总共结束的进程数
|
||||
with Bar("正在破解", max=len(ncm_files)) as bar:
|
||||
total = len(ncm_files)
|
||||
finished = 0 # 已经分配的任务数量
|
||||
while True: # 进入循环,执行 新建进程->检测队列->检测任务完成 的循环
|
||||
if current_process <= max_process and finished < total: # 分配进程
|
||||
Process(target=process_work,
|
||||
args=(os.path.join(path, ncm_files[finished]),
|
||||
ncm_files[finished],
|
||||
target_path,
|
||||
q_err,
|
||||
q_info)).start()
|
||||
finished += 1
|
||||
while True: # 错误队列检测
|
||||
try:
|
||||
errors.append(q_err.get_nowait())
|
||||
passed += 1 # 总任务完成数
|
||||
current_process -= 1 # 检测到进程完毕将进程-1
|
||||
bar.next() # 推动进度条
|
||||
fails += 1 # 错误数量+1
|
||||
except Empty:
|
||||
break
|
||||
while True: # 信息队列检测
|
||||
try:
|
||||
r = q_info.get_nowait()
|
||||
musics.append({"id": r['musicId'], "name": r["musicName"], "artists": r["artist"]})
|
||||
passed += 1
|
||||
current_process -= 1
|
||||
bar.next()
|
||||
except Empty:
|
||||
break
|
||||
if passed >= len(ncm_files):
|
||||
break
|
||||
if errors:
|
||||
print("解密过程中发现了以下错误:")
|
||||
for i in errors:
|
||||
print(i)
|
||||
|
||||
# 汇报索引结果
|
||||
ncm_files_num = 0
|
||||
if ncm_files:
|
||||
if target_path == "NOT_DECRYPT":
|
||||
ncm_files_num = len(ncm_files)
|
||||
print(f"\n索引完毕!共找到{fails + len(musics) + ncm_files_num}个目标文件\n{len(musics)}个文件已载入\n{fails}个文件失败")
|
||||
if ncm_files:
|
||||
if target_path == "NOT_DECRYPT":
|
||||
print(f"{len(ncm_files)}个文件放弃加载")
|
||||
while True:
|
||||
print("\n你希望如何保存这些歌曲的歌词?\n[1]保存到刚刚输入的绝对路径中\n[2]保存到程序设定的下载路径中")
|
||||
r = rinput("请选择: ")
|
||||
if r == "1":
|
||||
lyric_path = path
|
||||
break
|
||||
elif r == "2":
|
||||
break
|
||||
else:
|
||||
try:
|
||||
input("无效选择, 若取消请按 ^C ,继续请按回车")
|
||||
clear()
|
||||
except KeyboardInterrupt:
|
||||
return
|
||||
|
||||
clear()
|
||||
for i in range(0, len(musics)): # 根据索引结果获取歌词
|
||||
print("\n进度: %d/%d" % (i + 1, len(musics)))
|
||||
if get_song_lyric(musics[i], lyric_path, allinfo=True) == "dl_err_connection":
|
||||
input("下载发生错误!可能是连接被拒绝!请检查网络后再试\n按回车键继续任务(该任务会被跳过)...")
|
||||
if ncm_files:
|
||||
if target_path != "NOT_DECRYPT":
|
||||
agree = rinput("是否删除原ncm文件? (y/n)")
|
||||
if agree == "y":
|
||||
for i in range(0, len(ncm_files)):
|
||||
print("删除进度: %d/%d\n -> %s\033[F" % (i + 1, len(ncm_files), ncm_files[i]), end="")
|
||||
os.remove(os.path.join(path, ncm_files[i]))
|
||||
else:
|
||||
print("取消.", end="")
|
||||
input("\n\033[K按回车返回...")
|
||||
return
|
@ -1,36 +1,38 @@
|
||||
import re
|
||||
from modules.clear_screen import clear
|
||||
from modules.inputs import rinput
|
||||
from modules.get_song import get_song_lyric
|
||||
|
||||
|
||||
def mdl(path: str):
|
||||
"""多个歌词文件的下载
|
||||
|
||||
``path: str`` 传入歌词文件保存的路径"""
|
||||
clear()
|
||||
ids = []
|
||||
print("输入歌曲id,用回车分开,输入s停止")
|
||||
while True:
|
||||
r = rinput()
|
||||
if r == 's':
|
||||
break
|
||||
else:
|
||||
try:
|
||||
int(r)
|
||||
except ValueError:
|
||||
tmp = re.search(r"song\?id=[0-9]*", r)
|
||||
if tmp:
|
||||
r = tmp.group()[8:]
|
||||
else:
|
||||
print("不合法的形式.\n")
|
||||
continue
|
||||
ids.append(r)
|
||||
print("\t#%d id:%s - 已添加!" % (len(ids), r))
|
||||
clear()
|
||||
for i in range(0, len(ids)):
|
||||
print("进度: %d/%d" % (i+1, len(ids)))
|
||||
r = get_song_lyric(ids[i], path)
|
||||
if r == "dl_err_connection":
|
||||
input("下载发生错误!可能是连接被拒绝!请检查网络后再试\n按回车键继续任务(该任务会被跳过)...")
|
||||
input("按回车键返回...")
|
||||
import re
|
||||
from modules.utils.clear_screen import clear
|
||||
from modules.utils.inputs import rinput
|
||||
from modules.functions.get_song import get_song_lyric
|
||||
|
||||
|
||||
def mdl(self):
|
||||
"""多个歌词文件的下载
|
||||
|
||||
``path: str`` 传入歌词文件保存的路径"""
|
||||
clear()
|
||||
ids = []
|
||||
print(f"[NeteaseMusicLyricDownloader] {self.version}\n"
|
||||
"[手动-多个下载]\n"
|
||||
"输入歌曲id,用回车分开,输入s停止")
|
||||
while True:
|
||||
r = rinput()
|
||||
if r == 's':
|
||||
break
|
||||
else:
|
||||
try:
|
||||
int(r)
|
||||
except ValueError:
|
||||
tmp = re.search(r"song\?id=[0-9]*", r)
|
||||
if tmp:
|
||||
r = tmp.group()[8:]
|
||||
else:
|
||||
print("不合法的形式.\n")
|
||||
continue
|
||||
ids.append(r)
|
||||
print("\t#%d id:%s - 已添加!" % (len(ids), r))
|
||||
clear()
|
||||
for i in range(0, len(ids)):
|
||||
print("进度: %d/%d" % (i+1, len(ids)))
|
||||
r = get_song_lyric(ids[i], self.settings.lyric_path)
|
||||
if r == "dl_err_connection":
|
||||
input("下载发生错误!可能是连接被拒绝!请检查网络后再试\n按回车键继续任务(该任务会被跳过)...")
|
||||
input("按回车键返回...")
|
@ -1,15 +1,18 @@
|
||||
import re
|
||||
from modules.inputs import rinput
|
||||
from modules.get_song import get_song_lyric
|
||||
from modules.clear_screen import clear
|
||||
from modules.utils.inputs import rinput
|
||||
from modules.functions.get_song import get_song_lyric
|
||||
from modules.utils.clear_screen import clear
|
||||
|
||||
|
||||
def download_one_lyric(path: str):
|
||||
def download_one_lyric(self):
|
||||
"""单次下载歌词
|
||||
|
||||
``path: str`` 存储歌词的路径"""
|
||||
clear()
|
||||
song_id = rinput("请输入歌曲id:")
|
||||
song_id = rinput(
|
||||
f"[NeteaseMusicLyricDownloader] {self.version}\n"
|
||||
"[手动-单个下载]\n"
|
||||
"请输入歌曲id:")
|
||||
try:
|
||||
int(song_id)
|
||||
except ValueError:
|
||||
@ -20,6 +23,6 @@ def download_one_lyric(path: str):
|
||||
input("不合法的形式.\n按回车键返回...")
|
||||
return
|
||||
|
||||
if get_song_lyric(song_id, path) == "dl_err_connection":
|
||||
if get_song_lyric(song_id, self.settings.lyric_path) == "dl_err_connection":
|
||||
input("下载发生错误!可能是连接被拒绝!请检查网络后再试\n按回车键返回...")
|
||||
input("按回车键返回...")
|
0
modules/submenus/__init__.py
Normal file
0
modules/submenus/__init__.py
Normal file
@ -1,9 +1,9 @@
|
||||
"""集合设置参数"""
|
||||
|
||||
import os
|
||||
from modules.clear_screen import clear
|
||||
from modules.inputs import rinput, cinput
|
||||
from modules.save_load_settings import save_settings
|
||||
from modules.utils.clear_screen import clear
|
||||
from modules.utils.inputs import rinput, cinput
|
||||
from modules.functions.save_load_settings import save_settings
|
||||
|
||||
|
||||
def settings_menu(self):
|
||||
@ -36,7 +36,7 @@ def __remove_output_files(self):
|
||||
clear()
|
||||
print(f"[NeteaseMusicLyricDownloader] {self.version}\n"
|
||||
"[设置菜单 - 删除文件]\n"
|
||||
"[0] 返回上级\n[1] 清除歌词文件\n[2] 清除歌曲文件")
|
||||
"[0] 返回上级\n[1] 清除歌词文件\n[2] 清除歌曲文件\n[a] 清除所有文件")
|
||||
r = rinput("请选择:") # 选择清除的文件格式
|
||||
if r == "0":
|
||||
return
|
||||
@ -46,14 +46,20 @@ def __remove_output_files(self):
|
||||
elif r == "2":
|
||||
dellist = [".mp3", ".flac"]
|
||||
break
|
||||
elif r == "a":
|
||||
dellist = ["ALL"]
|
||||
break
|
||||
else:
|
||||
input("输入无效!\n按回车键继续...")
|
||||
files = []
|
||||
for i in os.listdir(self.settings.lyric_path): # 列出所有文件
|
||||
if os.path.splitext(i)[-1] in dellist: # 匹配文件
|
||||
if dellist[0] == "ALL":
|
||||
files = os.listdir(self.settings.lyric_path)
|
||||
break
|
||||
elif os.path.splitext(i)[-1] in dellist: # 匹配文件
|
||||
files.append(i) # 将匹配到的文件加入到列表, 等待删除
|
||||
if len(files) != 0:
|
||||
if len(files) > 50:
|
||||
if len(files) > 30:
|
||||
special_text = "\033[F"
|
||||
else:
|
||||
special_text = "\n"
|
0
modules/utils/__init__.py
Normal file
0
modules/utils/__init__.py
Normal file
@ -1,12 +1,12 @@
|
||||
"""包含一个函数,用来清空命令行的信息,自动判别系统"""
|
||||
import os
|
||||
|
||||
|
||||
def clear():
|
||||
name = os.name
|
||||
if name == "nt":
|
||||
os.system("cls")
|
||||
elif name == "posix":
|
||||
os.system("clear")
|
||||
else:
|
||||
"""包含一个函数,用来清空命令行的信息,自动判别系统"""
|
||||
import os
|
||||
|
||||
|
||||
def clear():
|
||||
name = os.name
|
||||
if name == "nt":
|
||||
os.system("cls")
|
||||
elif name == "posix":
|
||||
os.system("clear")
|
||||
else:
|
||||
os.system("clear")
|
@ -1,24 +1,24 @@
|
||||
"""该程序的自述信息,调用即输出"""
|
||||
from modules.clear_screen import clear
|
||||
|
||||
|
||||
def print_info(self):
|
||||
"""调用即输出,无返回值"""
|
||||
clear()
|
||||
print(f"""[NeteaseMusicLyricDownloader]
|
||||
版本: {self.version}
|
||||
本软件开源,项目地址:https://github.com/1826013250/NeteaseMusicLyricDownloader
|
||||
作者:David-123
|
||||
联系方式:
|
||||
\tQQ:1826013250
|
||||
\tE-mail:1826013250@qq.com(mainly)
|
||||
\t mc1826013250@gmail.com
|
||||
|
||||
特别感谢:
|
||||
\t- nondanee - ncmdump https://github.com/nondanee/ncmdump
|
||||
\t- QCloudHao - 保存了完整的ncmdump源代码 https://github.com/QCloudHao/ncmdump
|
||||
\t- chuyaoxin - 提供了对ncmdump以及ncm文件的详细解说 https://www.cnblogs.com/cyx-b/p/13443003.html
|
||||
|
||||
若程序遇到bug请提交至github上的issue""")
|
||||
input("按回车键返回...")
|
||||
return
|
||||
"""该程序的自述信息,调用即输出"""
|
||||
from modules.utils.clear_screen import clear
|
||||
|
||||
|
||||
def print_info(self):
|
||||
"""调用即输出,无返回值"""
|
||||
clear()
|
||||
print(f"""[NeteaseMusicLyricDownloader]
|
||||
版本: {self.version}
|
||||
本软件开源,项目地址:https://github.com/1826013250/NeteaseMusicLyricDownloader
|
||||
作者:David-123
|
||||
联系方式:
|
||||
\tQQ:1826013250
|
||||
\tE-mail:1826013250@qq.com(mainly)
|
||||
\t mc1826013250@gmail.com
|
||||
|
||||
特别感谢:
|
||||
\t- nondanee - ncmdump https://github.com/nondanee/ncmdump
|
||||
\t- QCloudHao - 保存了完整的ncmdump源代码 https://github.com/QCloudHao/ncmdump
|
||||
\t- chuyaoxin - 提供了对ncmdump以及ncm文件的详细解说 https://www.cnblogs.com/cyx-b/p/13443003.html
|
||||
|
||||
若程序遇到bug请提交至github上的issue""")
|
||||
input("按回车键返回...")
|
||||
return
|
@ -1,11 +1,11 @@
|
||||
"""该模块提供几个自定义处理输入函数"""
|
||||
|
||||
|
||||
def rinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格并全部小写的str"""
|
||||
return input(string).strip().lower()
|
||||
|
||||
|
||||
def cinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格的str"""
|
||||
return input(string).strip()
|
||||
"""该模块提供几个自定义处理输入函数"""
|
||||
|
||||
|
||||
def rinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格并全部小写的str"""
|
||||
return input(string).strip().lower()
|
||||
|
||||
|
||||
def cinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格的str"""
|
||||
return input(string).strip()
|
@ -1,11 +1,11 @@
|
||||
"""该模块提供几个自定义处理输入函数"""
|
||||
|
||||
|
||||
def rinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首位空格并全部小写的str"""
|
||||
return input(string).strip().lower()
|
||||
|
||||
|
||||
def cinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格的str"""
|
||||
"""该模块提供几个自定义处理输入函数"""
|
||||
|
||||
|
||||
def rinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首位空格并全部小写的str"""
|
||||
return input(string).strip().lower()
|
||||
|
||||
|
||||
def cinput(string: str = ''):
|
||||
"""当调用该函数时,同input()一样,但是返回一个去除首尾空格的str"""
|
||||
return input(string).strip()
|
Loading…
Reference in New Issue
Block a user