mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 12:20:23 +00:00
41 lines
1.5 KiB
Python
41 lines
1.5 KiB
Python
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class ThisAmericanLifeIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?thisamericanlife\.org/(?:radio-archives/episode/|play_full\.php\?play=)(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'http://www.thisamericanlife.org/radio-archives/episode/487/harper-high-school-part-one',
|
|
'md5': '8f7d2da8926298fdfca2ee37764c11ce',
|
|
'info_dict': {
|
|
'id': '487',
|
|
'ext': 'm4a',
|
|
'title': '487: Harper High School, Part One',
|
|
'description': 'md5:ee40bdf3fb96174a9027f76dbecea655',
|
|
'thumbnail': 're:^https?://.*\.jpg$',
|
|
},
|
|
}, {
|
|
'url': 'http://www.thisamericanlife.org/play_full.php?play=487',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(
|
|
'http://www.thisamericanlife.org/radio-archives/episode/%s' % video_id, video_id)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': 'http://stream.thisamericanlife.org/{0}/stream/{0}_64k.m3u8'.format(video_id),
|
|
'protocol': 'm3u8_native',
|
|
'ext': 'm4a',
|
|
'acodec': 'aac',
|
|
'vcodec': 'none',
|
|
'abr': 64,
|
|
'title': self._html_search_meta(r'twitter:title', webpage, 'title', fatal=True),
|
|
'description': self._html_search_meta(r'description', webpage, 'description'),
|
|
'thumbnail': self._og_search_thumbnail(webpage),
|
|
}
|