mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-05 01:44:33 +01:00
81 lines
3.5 KiB
Python
81 lines
3.5 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
int_or_none,
|
|
parse_iso8601,
|
|
sanitized_Request,
|
|
)
|
|
|
|
|
|
class AudiMediaIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?audimedia\.tv/(?:en|de)/vid/(?P<id>[^/?#]+)'
|
|
_TEST = {
|
|
'url': 'https://audimedia.tv/en/vid/60-seconds-of-audi-sport-104-2015-wec-bahrain-rookie-test',
|
|
'md5': '79a8b71c46d49042609795ab59779b66',
|
|
'info_dict': {
|
|
'id': '1565',
|
|
'ext': 'mp4',
|
|
'title': '60 Seconds of Audi Sport 104/2015 - WEC Bahrain, Rookie Test',
|
|
'description': 'md5:60e5d30a78ced725f7b8d34370762941',
|
|
'upload_date': '20151124',
|
|
'timestamp': 1448354940,
|
|
'duration': 74022,
|
|
'view_count': int,
|
|
}
|
|
}
|
|
# extracted from https://audimedia.tv/assets/embed/embedded-player.js (dataSourceAuthToken)
|
|
_AUTH_TOKEN = 'e25b42847dba18c6c8816d5d8ce94c326e06823ebf0859ed164b3ba169be97f2'
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
raw_payload = self._search_regex(r'<script[^>]+class="amtv-embed"[^>]+id="([^"]+)"', webpage, 'raw payload')
|
|
_, stage_mode, video_id, lang = raw_payload.split('-')
|
|
|
|
# TODO: handle s and e stage_mode (live streams and ended live streams)
|
|
if stage_mode not in ('s', 'e'):
|
|
request = sanitized_Request(
|
|
'https://audimedia.tv/api/video/v1/videos/%s?embed[]=video_versions&embed[]=thumbnail_image&where[content_language_iso]=%s' % (video_id, lang),
|
|
headers={'X-Auth-Token': self._AUTH_TOKEN})
|
|
json_data = self._download_json(request, video_id)['results']
|
|
formats = []
|
|
|
|
stream_url_hls = json_data.get('stream_url_hls')
|
|
if stream_url_hls:
|
|
formats.extend(self._extract_m3u8_formats(
|
|
stream_url_hls, video_id, 'mp4',
|
|
entry_protocol='m3u8_native', m3u8_id='hls', fatal=False))
|
|
|
|
stream_url_hds = json_data.get('stream_url_hds')
|
|
if stream_url_hds:
|
|
formats.extend(self._extract_f4m_formats(
|
|
stream_url_hds + '?hdcore=3.4.0',
|
|
video_id, f4m_id='hds', fatal=False))
|
|
|
|
for video_version in json_data.get('video_versions'):
|
|
video_version_url = video_version.get('download_url') or video_version.get('stream_url')
|
|
if not video_version_url:
|
|
continue
|
|
formats.append({
|
|
'url': video_version_url,
|
|
'width': int_or_none(video_version.get('width')),
|
|
'height': int_or_none(video_version.get('height')),
|
|
'abr': int_or_none(video_version.get('audio_bitrate')),
|
|
'vbr': int_or_none(video_version.get('video_bitrate')),
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': json_data['title'],
|
|
'description': json_data.get('subtitle'),
|
|
'thumbnail': json_data.get('thumbnail_image', {}).get('file'),
|
|
'timestamp': parse_iso8601(json_data.get('publication_date')),
|
|
'duration': int_or_none(json_data.get('duration')),
|
|
'view_count': int_or_none(json_data.get('view_count')),
|
|
'formats': formats,
|
|
}
|