diff options
Diffstat (limited to 'youtube_dl/extractor/audiomack.py')
-rw-r--r-- | youtube_dl/extractor/audiomack.py | 69 |
1 files changed, 69 insertions, 0 deletions
diff --git a/youtube_dl/extractor/audiomack.py b/youtube_dl/extractor/audiomack.py new file mode 100644 index 000000000..622b20989 --- /dev/null +++ b/youtube_dl/extractor/audiomack.py @@ -0,0 +1,69 @@ +# coding: utf-8 +from __future__ import unicode_literals + +from .common import InfoExtractor +from .soundcloud import SoundcloudIE +from ..utils import ExtractorError + +import time + + +class AudiomackIE(InfoExtractor): + _VALID_URL = r'https?://(?:www\.)?audiomack\.com/song/(?P<id>[\w/-]+)' + IE_NAME = 'audiomack' + _TESTS = [ + # hosted on audiomack + { + 'url': 'http://www.audiomack.com/song/roosh-williams/extraordinary', + 'info_dict': + { + 'id': 'roosh-williams/extraordinary', + 'ext': 'mp3', + 'title': 'Roosh Williams - Extraordinary' + } + }, + # hosted on soundcloud via audiomack + { + 'add_ie': ['Soundcloud'], + 'url': 'http://www.audiomack.com/song/xclusiveszone/take-kare', + 'info_dict': { + 'id': '172419696', + 'ext': 'mp3', + 'description': 'md5:1fc3272ed7a635cce5be1568c2822997', + 'title': 'Young Thug ft Lil Wayne - Take Kare', + 'uploader': 'Young Thug World', + 'upload_date': '20141016', + } + }, + ] + + def _real_extract(self, url): + video_id = self._match_id(url) + + api_response = self._download_json( + "http://www.audiomack.com/api/music/url/song/%s?_=%d" % ( + video_id, time.time()), + video_id) + + if "url" not in api_response: + raise ExtractorError("Unable to deduce api url of song") + realurl = api_response["url"] + + # Audiomack wraps a lot of soundcloud tracks in their branded wrapper + # - if so, pass the work off to the soundcloud extractor + if SoundcloudIE.suitable(realurl): + return {'_type': 'url', 'url': realurl, 'ie_key': 'Soundcloud'} + + webpage = self._download_webpage(url, video_id) + artist = self._html_search_regex( + r'<span class="artist">(.*?)</span>', webpage, "artist") + songtitle = self._html_search_regex( + r'<h1 class="profile-title song-title"><span class="artist">.*?</span>(.*?)</h1>', + webpage, "title") + title = artist + " - " + songtitle + + return { + 'id': video_id, + 'title': title, + 'url': realurl, + } |