mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-16 03:53:29 +01:00
1f343eaabb
webpage. I added the parameter webpage, so now it's similar to the way automatic captions are handled. This is an improvement needed for websites like TED.
92 lines
3.7 KiB
Python
92 lines
3.7 KiB
Python
from .common import InfoExtractor
|
|
|
|
from ..utils import (
|
|
compat_str,
|
|
ExtractorError,
|
|
)
|
|
|
|
|
|
class SubtitlesInfoExtractor(InfoExtractor):
|
|
@property
|
|
def _have_to_download_any_subtitles(self):
|
|
return any([self._downloader.params.get('writesubtitles', False),
|
|
self._downloader.params.get('writeautomaticsub')])
|
|
|
|
def _list_available_subtitles(self, video_id, webpage):
|
|
""" outputs the available subtitles for the video """
|
|
sub_lang_list = self._get_available_subtitles(video_id, webpage)
|
|
auto_captions_list = self._get_available_automatic_caption(video_id, webpage)
|
|
sub_lang = ",".join(list(sub_lang_list.keys()))
|
|
self.to_screen(u'%s: Available subtitles for video: %s' %
|
|
(video_id, sub_lang))
|
|
auto_lang = ",".join(auto_captions_list.keys())
|
|
self.to_screen(u'%s: Available automatic captions for video: %s' %
|
|
(video_id, auto_lang))
|
|
|
|
def extract_subtitles(self, video_id, webpage):
|
|
"""
|
|
returns {sub_lang: sub} ,{} if subtitles not found or None if the
|
|
subtitles aren't requested.
|
|
"""
|
|
if not self._have_to_download_any_subtitles:
|
|
return None
|
|
available_subs_list = {}
|
|
if self._downloader.params.get('writeautomaticsub', False):
|
|
available_subs_list.update(self._get_available_automatic_caption(video_id, webpage))
|
|
if self._downloader.params.get('writesubtitles', False):
|
|
available_subs_list.update(self._get_available_subtitles(video_id, webpage))
|
|
|
|
if not available_subs_list: # error, it didn't get the available subtitles
|
|
return {}
|
|
if self._downloader.params.get('allsubtitles', False):
|
|
sub_lang_list = available_subs_list
|
|
else:
|
|
if self._downloader.params.get('subtitleslangs', False):
|
|
requested_langs = self._downloader.params.get('subtitleslangs')
|
|
elif 'en' in available_subs_list:
|
|
requested_langs = ['en']
|
|
else:
|
|
requested_langs = [list(available_subs_list.keys())[0]]
|
|
|
|
sub_lang_list = {}
|
|
for sub_lang in requested_langs:
|
|
if not sub_lang in available_subs_list:
|
|
self._downloader.report_warning(u'no closed captions found in the specified language "%s"' % sub_lang)
|
|
continue
|
|
sub_lang_list[sub_lang] = available_subs_list[sub_lang]
|
|
|
|
subtitles = {}
|
|
for sub_lang, url in sub_lang_list.items():
|
|
subtitle = self._request_subtitle_url(sub_lang, url)
|
|
if subtitle:
|
|
subtitles[sub_lang] = subtitle
|
|
return subtitles
|
|
|
|
def _request_subtitle_url(self, sub_lang, url):
|
|
""" makes the http request for the subtitle """
|
|
try:
|
|
sub = self._download_webpage(url, None, note=False)
|
|
except ExtractorError as err:
|
|
self._downloader.report_warning(u'unable to download video subtitles for %s: %s' % (sub_lang, compat_str(err)))
|
|
return
|
|
if not sub:
|
|
self._downloader.report_warning(u'Did not fetch video subtitles')
|
|
return
|
|
return sub
|
|
|
|
def _get_available_subtitles(self, video_id, webpage):
|
|
"""
|
|
returns {sub_lang: url} or {} if not available
|
|
Must be redefined by the subclasses
|
|
"""
|
|
pass
|
|
|
|
def _get_available_automatic_caption(self, video_id, webpage):
|
|
"""
|
|
returns {sub_lang: url} or {} if not available
|
|
Must be redefined by the subclasses that support automatic captions,
|
|
otherwise it will return {}
|
|
"""
|
|
self._downloader.report_warning(u'Automatic Captions not supported by this server')
|
|
return {}
|