[voicerepublic] fix extraction

pull/2/head
Remita Amine 5 years ago
parent 9cf30dc017
commit be96f9924f

@ -1,17 +1,12 @@
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import (
compat_str,
compat_urlparse,
)
from ..compat import compat_str
from ..utils import (
ExtractorError,
determine_ext,
int_or_none,
sanitized_Request,
urljoin,
)
@ -26,8 +21,7 @@ class VoiceRepublicIE(InfoExtractor):
'ext': 'm4a',
'title': 'Watching the Watchers: Building a Sousveillance State',
'description': 'Secret surveillance programs have metadata too. The people and companies that operate secret surveillance programs can be surveilled.',
'thumbnail': r're:^https?://.*\.(?:png|jpg)$',
'duration': 1800,
'duration': 1556,
'view_count': int,
}
}, {
@ -38,63 +32,31 @@ class VoiceRepublicIE(InfoExtractor):
def _real_extract(self, url):
display_id = self._match_id(url)
req = sanitized_Request(
compat_urlparse.urljoin(url, '/talks/%s' % display_id))
# Older versions of Firefox get redirected to an "upgrade browser" page
req.add_header('User-Agent', 'youtube-dl')
webpage = self._download_webpage(req, display_id)
webpage = self._download_webpage(url, display_id)
if '>Queued for processing, please stand by...<' in webpage:
raise ExtractorError(
'Audio is still queued for processing', expected=True)
config = self._search_regex(
r'(?s)return ({.+?});\s*\n', webpage,
'data', default=None)
data = self._parse_json(config, display_id, fatal=False) if config else None
if data:
title = data['title']
description = data.get('teaser')
talk_id = compat_str(data.get('talk_id') or display_id)
talk = data['talk']
duration = int_or_none(talk.get('duration'))
talk = self._parse_json(self._search_regex(
r'initialSnapshot\s*=\s*({.+?});',
webpage, 'talk'), display_id)['talk']
title = talk['title']
formats = [{
'url': compat_urlparse.urljoin(url, talk_url),
'url': urljoin(url, talk_url),
'format_id': format_id,
'ext': determine_ext(talk_url) or format_id,
'vcodec': 'none',
} for format_id, talk_url in talk['links'].items()]
else:
title = self._og_search_title(webpage)
description = self._html_search_regex(
r"(?s)<div class='talk-teaser'[^>]*>(.+?)</div>",
webpage, 'description', fatal=False)
talk_id = self._search_regex(
[r"id='jc-(\d+)'", r"data-shareable-id='(\d+)'"],
webpage, 'talk id', default=None) or display_id
duration = None
player = self._search_regex(
r"class='vr-player jp-jplayer'([^>]+)>", webpage, 'player')
formats = [{
'url': compat_urlparse.urljoin(url, talk_url),
'format_id': format_id,
'ext': determine_ext(talk_url) or format_id,
'vcodec': 'none',
} for format_id, talk_url in re.findall(r"data-([^=]+)='([^']+)'", player)]
} for format_id, talk_url in talk['media_links'].items()]
self._sort_formats(formats)
thumbnail = self._og_search_thumbnail(webpage)
view_count = int_or_none(self._search_regex(
r"class='play-count[^']*'>\s*(\d+) plays",
webpage, 'play count', fatal=False))
return {
'id': talk_id,
'id': compat_str(talk.get('id') or display_id),
'display_id': display_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'duration': duration,
'view_count': view_count,
'description': talk.get('teaser'),
'thumbnail': talk.get('image_url'),
'duration': int_or_none(talk.get('archived_duration')),
'view_count': int_or_none(talk.get('play_count')),
'formats': formats,
}

Loading…
Cancel
Save