[ie/facebook] Improve subtitles extraction (#8296)

Authored by: kclauhk
pull/8677/head
kclauhk 1 year ago committed by GitHub
parent 1732eccc0a
commit 9cafb9ff17
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -16,6 +16,7 @@ from ..utils import (
determine_ext, determine_ext,
error_to_compat_str, error_to_compat_str,
float_or_none, float_or_none,
format_field,
get_element_by_id, get_element_by_id,
get_first, get_first,
int_or_none, int_or_none,
@ -420,6 +421,29 @@ class FacebookIE(InfoExtractor):
r'data-sjs>({.*?ScheduledServerJS.*?})</script>', webpage)] r'data-sjs>({.*?ScheduledServerJS.*?})</script>', webpage)]
post = traverse_obj(post_data, ( post = traverse_obj(post_data, (
..., 'require', ..., ..., ..., '__bbox', 'require', ..., ..., ..., '__bbox', 'result', 'data'), expected_type=dict) or [] ..., 'require', ..., ..., ..., '__bbox', 'require', ..., ..., ..., '__bbox', 'result', 'data'), expected_type=dict) or []
automatic_captions, subtitles = {}, {}
subs_data = traverse_obj(post, (..., 'video', ..., 'attachments', ..., lambda k, v: (
k == 'media' and str(v['id']) == video_id and v['__typename'] == 'Video')))
is_video_broadcast = get_first(subs_data, 'is_video_broadcast', expected_type=bool)
captions = get_first(subs_data, 'video_available_captions_locales', 'captions_url')
if url_or_none(captions): # if subs_data only had a 'captions_url'
locale = self._html_search_meta(['og:locale', 'twitter:locale'], webpage, 'locale', default='en_US')
subtitles[locale] = [{'url': captions}]
# or else subs_data had 'video_available_captions_locales', a list of dicts
for caption in traverse_obj(captions, (
{lambda x: sorted(x, key=lambda c: c['locale'])}, lambda _, v: v['captions_url'])
):
lang = caption.get('localized_language') or ''
subs = {
'url': caption['captions_url'],
'name': format_field(caption, 'localized_country', f'{lang} (%s)', default=lang),
}
if caption.get('localized_creation_method') or is_video_broadcast:
automatic_captions.setdefault(caption['locale'], []).append(subs)
else:
subtitles.setdefault(caption['locale'], []).append(subs)
media = traverse_obj(post, (..., 'attachments', ..., lambda k, v: ( media = traverse_obj(post, (..., 'attachments', ..., lambda k, v: (
k == 'media' and str(v['id']) == video_id and v['__typename'] == 'Video')), expected_type=dict) k == 'media' and str(v['id']) == video_id and v['__typename'] == 'Video')), expected_type=dict)
title = get_first(media, ('title', 'text')) title = get_first(media, ('title', 'text'))
@ -463,6 +487,8 @@ class FacebookIE(InfoExtractor):
webpage, 'view count', default=None)), webpage, 'view count', default=None)),
'concurrent_view_count': get_first(post, ( 'concurrent_view_count': get_first(post, (
('video', (..., ..., 'attachments', ..., 'media')), 'liveViewerCount', {int_or_none})), ('video', (..., ..., 'attachments', ..., 'media')), 'liveViewerCount', {int_or_none})),
'automatic_captions': automatic_captions,
'subtitles': subtitles,
} }
info_json_ld = self._search_json_ld(webpage, video_id, default={}) info_json_ld = self._search_json_ld(webpage, video_id, default={})

Loading…
Cancel
Save