[extractor/genius] Add extractors (#5221)

Closes #5209
Authored by: bashonly
This commit is contained in:
bashonly 2022-11-04 15:37:45 +00:00 committed by GitHub
parent ed13a772d7
commit 68a9a450d4
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 131 additions and 0 deletions

View file

@ -636,6 +636,10 @@ from .gazeta import GazetaIE
from .gdcvault import GDCVaultIE
from .gedidigital import GediDigitalIE
from .generic import GenericIE
from .genius import (
GeniusIE,
GeniusLyricsIE,
)
from .gettr import (
GettrIE,
GettrStreamingIE,

127
yt_dlp/extractor/genius.py Normal file
View file

@ -0,0 +1,127 @@
from .common import InfoExtractor
from ..utils import (
ExtractorError,
js_to_json,
smuggle_url,
str_or_none,
traverse_obj,
unescapeHTML,
)
class GeniusIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?genius\.com/videos/(?P<id>[^?/#]+)'
_TESTS = [{
'url': 'https://genius.com/videos/Vince-staples-breaks-down-the-meaning-of-when-sparks-fly',
'md5': '64c2ad98cfafcfda23bfa0ad0c512f4c',
'info_dict': {
'id': '6313303597112',
'ext': 'mp4',
'title': 'Vince Staples Breaks Down The Meaning Of “When Sparks Fly”',
'description': 'md5:bc15e00342c537c0039d414423ae5752',
'tags': 'count:1',
'uploader_id': '4863540648001',
'duration': 388.416,
'upload_date': '20221005',
'timestamp': 1664982341,
'thumbnail': r're:^https?://.*\.jpg$',
},
}, {
'url': 'https://genius.com/videos/Breaking-down-drakes-certified-lover-boy-kanye-beef-way-2-sexy-cudi',
'md5': 'b8ed87a5efd1473bd027c20a969d4060',
'info_dict': {
'id': '6271792014001',
'ext': 'mp4',
'title': 'md5:c6355f7fa8a70bc86492a3963919fc15',
'description': 'md5:1774638c31548b31b037c09e9b821393',
'tags': 'count:3',
'uploader_id': '4863540648001',
'duration': 2685.099,
'upload_date': '20210909',
'timestamp': 1631209167,
'thumbnail': r're:^https?://.*\.jpg$',
},
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
metadata = self._search_json(
r'<meta content="', webpage, 'metadata', display_id, transform_source=unescapeHTML)
video_id = traverse_obj(
metadata, ('video', 'provider_id'),
('dfp_kv', lambda _, x: x['name'] == 'brightcove_video_id', 'values', 0), get_all=False)
if not video_id:
raise ExtractorError('Brightcove video id not found in webpage')
config = self._search_json(r'var\s*APP_CONFIG\s*=', webpage, 'config', video_id, default={})
account_id = config.get('brightcove_account_id', '4863540648001')
player_id = traverse_obj(
config, 'brightcove_standard_web_player_id', 'brightcove_standard_no_autoplay_web_player_id',
'brightcove_modal_web_player_id', 'brightcove_song_story_web_player_id', default='S1ZcmcOC1x')
return self.url_result(
smuggle_url(
f'https://players.brightcove.net/{account_id}/{player_id}_default/index.html?videoId={video_id}',
{'referrer': url}), 'BrightcoveNew', video_id)
class GeniusLyricsIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?genius\.com/(?P<id>[^?/#]+)-lyrics[?/#]?'
_TESTS = [{
'url': 'https://genius.com/Lil-baby-heyy-lyrics',
'playlist_mincount': 2,
'info_dict': {
'id': '8454545',
'title': 'Heyy',
'description': 'Heyy by Lil Baby',
},
}, {
'url': 'https://genius.com/Outkast-two-dope-boyz-in-a-cadillac-lyrics',
'playlist_mincount': 1,
'info_dict': {
'id': '36239',
'title': 'Two Dope Boyz (In a Cadillac)',
'description': 'Two Dope Boyz (In a Cadillac) by OutKast',
},
}, {
'url': 'https://genius.com/Playboi-carti-rip-lyrics',
'playlist_mincount': 1,
'info_dict': {
'id': '3710582',
'title': 'R.I.P.',
'description': 'R.I.P. by Playboi Carti',
},
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
json_string = self._search_json(
r'window\.__PRELOADED_STATE__\s*=\s*JSON\.parse\(', webpage, 'json string',
display_id, transform_source=js_to_json, contains_pattern=r'\'{(?s:.+)}\'')
song_info = self._parse_json(json_string, display_id)
song_id = str_or_none(traverse_obj(song_info, ('songPage', 'song')))
if not song_id:
raise ExtractorError('Song id not found in webpage')
title = traverse_obj(
song_info, ('songPage', 'trackingData', lambda _, x: x['key'] == 'Title', 'value'),
get_all=False, default='untitled')
artist = traverse_obj(
song_info, ('songPage', 'trackingData', lambda _, x: x['key'] == 'Primary Artist', 'value'),
get_all=False, default='unknown artist')
media = traverse_obj(
song_info, ('entities', 'songs', song_id, 'media'), expected_type=list, default=[])
entries = []
for m in media:
if m.get('type') in ('video', 'audio') and m.get('url'):
if m.get('provider') == 'spotify':
self.to_screen(f'{song_id}: Skipping Spotify audio embed')
else:
entries.append(self.url_result(m['url']))
return self.playlist_result(entries, song_id, title, f'{title} by {artist}')