HLS WebVTT subtitle downloading support #6144
Conversation
|
i want you to know that in this commit avformat/hlsenc: Add WebVtt support in hls has been added to ffmpeg. |
|
libavcodec still doesn't have any support for Anyway, developers: is this any good? Should I adapt this to take advantage of #6392? @dstftw, @jaimeMF? |
| self.to_screen( | ||
| '[hlswebvtt] %s: Downloading manifest' % | ||
| (info_dict['id'])) | ||
| data = self.ydl.urlopen(url).read() |
There was a problem hiding this comment.
This won't work in case of live HLS WebVTT streams because you constantly get new subtitle segments at the same playlist URL.
It's a shame X-TIMESTAMP-MAP support patch hasn't been merged to ffmpeg yet after 3 years, but in my use-case (vlive.tv) it's not required, so dumping HLS WebVTT via ffmpeg works quite good.
_extract_m3u8_formats is renamed to _extract_m3u8_formats_and_subtitles and extended to properly handle subtitle references; a wrapper with the old name is provided for compatibility. _parse_m3u8_formats is likewise renamed and extended, but without adding the compatibility wrapper; the test suite is adjusted to test the enhanced method instead.
|
This code is awesome -- I had just started writing something just like this -- is there a reason it's not in master? |
|
You tell me. |
Authored by fstirlitz Modified from: ytdl-org/youtube-dl#6144 Closes: #73 Fixes: ytdl-org/youtube-dl#6106 ytdl-org/youtube-dl#14977 ytdl-org/youtube-dl#21438 ytdl-org/youtube-dl#23609 ytdl-org/youtube-dl#28132 Might also fix (untested): ytdl-org/youtube-dl#15424 ytdl-org/youtube-dl#18267 ytdl-org/youtube-dl#23899 ytdl-org/youtube-dl#24375 ytdl-org/youtube-dl#24595 ytdl-org/youtube-dl#27899 Related: ytdl-org/youtube-dl#22379 ytdl-org/youtube-dl#24517 ytdl-org/youtube-dl#24886 ytdl-org/youtube-dl#27215 Notes: * The functions `extractor.common._extract_..._formats` are still kept for compatibility * Only some extractors have currently been moved to using `_extract_..._formats_and_subtitles` * Direct subtitle manifests (without a master) are not supported and are wrongly identified as containing video formats * AES support is untested * The fragmented TTML subtitles extracted from DASH/ISM are valid, but are unsupported by `ffmpeg` and most video players * Their XML fragments can be dumped using `ffmpeg -i in.mp4 -f data -map 0 -c copy out.ttml`. Once the unnecessary headers are stripped out of this, it becomes a valid self-contained ttml file * The ttml subs downloaded from DASH manifests can also be directly opened with <https://github.com/SubtitleEdit> * Fragmented WebVTT files extracted from DASH/ISM are also unsupported by most tools * Unlike the ttml files, the XML fragments of these cannot be dumped using `ffmpeg` * The webtt subs extracted from DASH can be parsed by <https://github.com/gpac/gpac> * But validity of the those extracted from ISM are untested
Authored by fstirlitz Modified from: ytdl-org/youtube-dl#6144 Closes: #73 Fixes: ytdl-org/youtube-dl#6106 ytdl-org/youtube-dl#14977 ytdl-org/youtube-dl#21438 ytdl-org/youtube-dl#23609 ytdl-org/youtube-dl#28132 Might also fix (untested): ytdl-org/youtube-dl#15424 ytdl-org/youtube-dl#18267 ytdl-org/youtube-dl#23899 ytdl-org/youtube-dl#24375 ytdl-org/youtube-dl#24595 ytdl-org/youtube-dl#27899 Related: ytdl-org/youtube-dl#22379 ytdl-org/youtube-dl#24517 ytdl-org/youtube-dl#24886 ytdl-org/youtube-dl#27215 Notes: * The functions `extractor.common._extract_..._formats` are still kept for compatibility * Only some extractors have currently been moved to using `_extract_..._formats_and_subtitles` * Direct subtitle manifests (without a master) are not supported and are wrongly identified as containing video formats * AES support is untested * The fragmented TTML subtitles extracted from DASH/ISM are valid, but are unsupported by `ffmpeg` and most video players * Their XML fragments can be dumped using `ffmpeg -i in.mp4 -f data -map 0 -c copy out.ttml`. Once the unnecessary headers are stripped out of this, it becomes a valid self-contained ttml file * The ttml subs downloaded from DASH manifests can also be directly opened with <https://github.com/SubtitleEdit> * Fragmented WebVTT files extracted from DASH/ISM are also unsupported by most tools * Unlike the ttml files, the XML fragments of these cannot be dumped using `ffmpeg` * The webtt subs extracted from DASH can be parsed by <https://github.com/gpac/gpac> * But validity of the those extracted from ISM are untested
This should resolve #6106.
The patches are fairly invasive. Since downloading HLS WebVTT subtitles seems to be relatively heavyweight bandwidth-wise, downloading them right away in the extractor is out of the question; but there was no other way to download these subtitles in an offline-viewable form. Hence the subtitle downloading code was hacked to make use of the downloader infrastructure and a custom HLS WebVTT downloader was written.
The code was tested with Python 3.4.3 and 2.7.10 on a few ComCarCoff videos.