【请大家一定添加我们微信客服:sx-980】
软件失效,资源失效,求资源,反馈问题,防失联等
【所有软件均来源于网络以及投稿人提供】
【如被骗自行负责,请勿相信任何软件内推广信息】
简介
建议使用PC端网页进行查看,手机你可能整不明包
下载
# -*- coding: utf-8 -*- import string import time import re import random import requests import pprint ''' ==========================================酷狗音乐解析主要API============================================= 1.get_rank_list() 获取音乐榜单列表 返回形式:(榜单名,榜单地址) 2.get_rank_info(url) url为榜单地址 返回歌曲的信息[{}] 3.get_search_musicData(search_txt,dis_num) search_txt:搜索内容,dis_num:总共加载几首歌曲,返回歌曲的信息[{}] =========================================内部函数========================================================= 4.get_musicInfo(hash,id) 通过传入的hash、id 返回歌名、播放地址、图片、歌词 字典形式 返回形式:music_info_dic={'audio_name':XXXX,'play_url':XXXX,'lyrics':XXX,'img':XXX} 5.klcTolrc(self,lyrics) 通过传入获取的krc歌词 变成lrc形式的歌词 ======================================================================================================== 可以用主函数测试 ''' class kuGouMusic (): def __init__(self): self.headers = { 'user-agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/102.0.0.0 Safari/537.36', } def __get_response(self, url, headers): res = requests.get (url, headers) return res # 获取酷狗音乐榜单 def get_rank_list(self): rank_url = 'https://www.kugou.com/yy/html/rank.html' res = self.__get_response (rank_url, self.headers) res.encoding = res.apparent_encoding rank_list_urls = re.findall (r'<a title="(.*?)" .*? hidefocus="true" href="(.*?)"', res.text)[1:] return rank_list_urls # 获取榜单的Hash、id def get_rank_info(self, url): res = self.__get_response (url, self.headers) res.encoding = res.apparent_encoding Hash_list = re.findall ('"Hash":"(.*?)"', res.text) album_id_list = re.findall ('"album_id":(.*?),', res.text) play_list = [] num=0 for Hash, album_id in zip (Hash_list, album_id_list): print (f'\r正在加歌单歌曲列表,....当前进度:{int ((num + 1) / len (Hash_list) * 100)}%', end='') play_list.append (self.__get_musicInfo (Hash, album_id)) num=num+1 return play_list # ======================以下是搜索部分=============# def get_search_musicData(self,word='任贤齐',page_mum=1): url = 'http://mobilecdn.kugou.com/api/v3/search/song' data_list=[] headers = { 'user-agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/102.0.0.0 Safari/537.36' } parameter={ 'keyword': word, 'page': page_mum, 'pagesize': '30', } res = requests.get(url, headers=headers,params=parameter).json() info_list=res['data']['info'] num=0 for info in info_list: print (f'\r正在加载搜索歌曲列表,....当前进度:{int ((num + 1) / len (info_list) * 100)}%', end='') data_list.append(self.__get_musicInfo(info['hash'],info['album_id'])) num=num+1 return data_list # ======================通过Hash,id获取到歌曲地址、LRC、封面、歌曲名共用部分=============# # 获取歌曲信息 hash 为字符串 id为整型数据 def __get_musicInfo(self, hash, id): music_url = 'https://wwwapi.kugou.com/yy/index.php' parameter = { 'r': 'play/getdata', 'hash': hash, 'dfid': self.__get_dfid (23), 'mid': self.__get_mid (23), 'album_id': id, '_': str (round (time.time () * 1000)) # 时间戳 } json_data = requests.get (music_url, headers=self.headers, params=parameter).json () music_info_dic = { 'audio_name': json_data['data']['audio_name'], 'play_url': json_data['data']['play_url'], 'lyrics': self.__klcTolrc(json_data['data']['lyrics']), 'img_conntent': json_data['data']['img'] } return music_info_dic # 获取a-z A-Z 0-9组成的随机23位数列 def __get_dfid(self, num): random_str = ''.join (random.sample ((string.ascii_letters + string.digits), num)) return random_str # 获取a-z 0-9组成的随机23位数列 def __get_mid(self, num): random_str = ''.join (random.sample ((string.ascii_letters[:26] + string.digits), num)) return random_str # krc转成lrc def __klcTolrc(self, lyrics): index = lyrics.find ('[00:00') lyrics = lyrics[index:] return lyrics if __name__ == '__main__': d = kuGouMusic () # url='https://www.kugou.com/yy/rank/home/1-8888.html?from=rank' # lis=d.get_rank_info(url) # pprint.pprint(lis) lis2=d.get_search_musicData() print(lis2)