mirror of
https://codeberg.org/polarisfm/youtube-dl.git
synced 2024-12-27 18:16:44 +00:00
69 lines
2.5 KiB
Python
69 lines
2.5 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import float_or_none
|
|
|
|
|
|
class AudioBoomIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?audioboom\.com/(?:boos|posts)/(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'https://audioboom.com/boos/4279833-3-09-2016-czaban-hour-3?t=0',
|
|
'md5': '63a8d73a055c6ed0f1e51921a10a5a76',
|
|
'info_dict': {
|
|
'id': '4279833',
|
|
'ext': 'mp3',
|
|
'title': '3/09/2016 Czaban Hour 3',
|
|
'description': 'Guest: Nate Davis - NFL free agency, Guest: Stan Gans',
|
|
'duration': 2245.72,
|
|
'uploader': 'SB Nation A.M.',
|
|
'uploader_url': r're:https?://(?:www\.)?audioboom\.com/channel/steveczabanyahoosportsradio',
|
|
}
|
|
}, {
|
|
'url': 'https://audioboom.com/posts/4279833-3-09-2016-czaban-hour-3?t=0',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
clip = None
|
|
|
|
clip_store = self._parse_json(
|
|
self._search_regex(
|
|
r'data-new-clip-store=(["\'])(?P<json>{.*?"clipId"\s*:\s*%s.*?})\1' % video_id,
|
|
webpage, 'clip store', default='{}', group='json'),
|
|
video_id, fatal=False)
|
|
if clip_store:
|
|
clips = clip_store.get('clips')
|
|
if clips and isinstance(clips, list) and isinstance(clips[0], dict):
|
|
clip = clips[0]
|
|
|
|
def from_clip(field):
|
|
if clip:
|
|
return clip.get(field)
|
|
|
|
audio_url = from_clip('clipURLPriorToLoading') or self._og_search_property(
|
|
'audio', webpage, 'audio url')
|
|
title = from_clip('title') or self._og_search_title(webpage)
|
|
description = from_clip('description') or self._og_search_description(webpage)
|
|
|
|
duration = float_or_none(from_clip('duration') or self._html_search_meta(
|
|
'weibo:audio:duration', webpage))
|
|
|
|
uploader = from_clip('author') or self._og_search_property(
|
|
'audio:artist', webpage, 'uploader', fatal=False)
|
|
uploader_url = from_clip('author_url') or self._html_search_meta(
|
|
'audioboo:channel', webpage, 'uploader url')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': audio_url,
|
|
'title': title,
|
|
'description': description,
|
|
'duration': duration,
|
|
'uploader': uploader,
|
|
'uploader_url': uploader_url,
|
|
}
|