This tools takes a movie file and a subtitles file and prints ffmpeg commands that will chop the entire movie up into clips, one clip for each span of time that a subtitle appears on the screen. It also produces commands that will extract clips for all durations when no subtitles are shown on the screen.
Note that this program does not actually execute any ffmpeg commands, it only prints commands that should be executed. This makes it easier to verify that the commands are sane before any expensive extraction operations are executed.
I'm inspired by the old Arnold Schwarzenegger soundboards people used to use in prank phone calls.
I like the idea of doing what was done with soundboards with movies. I imagine friends gathering around a projector taking turns playing the role of a character by puppeteering their cinematic appearances. That, or joining public video chats, or chat roulettes, and broadcasting a puppeted feed of the character. So this program is a first step toward those ends.
Also, some funny compilations can come out of chopping up a video. Here's one where a character in Moonrise Kingdom narrates the credits, identifying instruments playing in an orchestral arrangement. The original credits include breaks in the narration while each instrument plays a bit of a solo, but this movie strips all elegance from the piece and lets us see all the instruments listed as quickly as they are named. It is a silly piece that wouldn't be worth noting if not for the fact that it was so easy to create after the clips had all already been split up by their subtitles. A real diamond found in a rough that was only waiting to be cut with the right tool!
Watch it: Moonrise Kingdom Orchestral Credits (Abridged)
One interesting thing I found while writing this program was a "typo" in one of the Moonrise Kingdom subtitle files. I've included a variety of the subtitle files in this repo for reference.
There's a strange character that appears in subtitle sequence numbers 15 and 89. The character becomes obvious when diffing two subtitle files:
diff subtitles/moonrise.srt subtitles/moonrise_marked.srt
71c76
< The flutes, the oboes,
---
> The flutes, the oboes,
415c420
< two bedrolls, plus an air rifle.
---
> two bedrolls, plus an air rifle.
The characters "fl" in moonrise.srt are replaced by a strange unicode character "fl" in moonrise_marked.srt. These strange unicode characters are known to appear in scammy URLs trying to mimic legitimate websites, "flaypal.com" vs "flaypal.com," but why should they appear in a subtitles file? The answer, I believe, lies in cartography.
The word on the street is that ancient, and medieval, and renaissance, and colonial, and modern, and even digital map makers would sometimes intentionally include tiny errors in their maps to protect their copyright. If these errors ever appeared in another unauthorized map then they could make the case that the other map copied all of their work, including the bogus errors, from their own maps. Though I know of no formal word to describe these errors (though I'd bet German has a name for them) I'm convinced whoever originally created the subtitle file inserted this error intentionally to see how far their work would spread. The pirate movie subtitle community surely must be as proud and territorial as any cartography community.
I only noticed and discovered this error because I wasn't originally parsing the file with the proper text encoding and the program would mysteriously crash when it encountered this character. Lesson learned: don't worry about encoding until you have to! You may reveal wonderful conspiracies!
This program should be executed as follows:
python subtitles.py [movie_file] [subtitle_file]
It will print a series of ffmpeg commands that will extract each clip.
ffmpeg -ss 00:00:00.000 -i "2012 Moonrise Kingdom.mp4" -t 00:01:30.376 -sameq -async 1 lines/000000_zzz.mp4
ffmpeg -ss 00:01:30.376 -i "2012 Moonrise Kingdom.mp4" -t 00:00:01.250 -sameq -async 1 lines/000001_in_order_to_show_you_how.mp4
ffmpeg -ss 00:01:31.626 -i "2012 Moonrise Kingdom.mp4" -t 00:00:00.083 -sameq -async 1 lines/000001_zzz.mp4
ffmpeg -ss 00:01:31.709 -i "2012 Moonrise Kingdom.mp4" -t 00:00:02.500 -sameq -async 1 lines/000002_a_big_symphonyorchestra_is_put_together.mp4
ffmpeg -ss 00:01:34.209 -i "2012 Moonrise Kingdom.mp4" -t 00:00:00.083 -sameq -async 1 lines/000002_zzz.mp4
If you want execute the commands you should redirect the output to a file and execute it:
python subtitles.py [movie_file] [subtitle_file] > extract.sh
chmod +x extract.sh
./extract.sh
Note that the commands alternate between extracting a clip that has a subtitle and a clip that has no subtitle. This ensures that the entire movie is broken into clips, whether that portion of the movie was showing a subtitle or not. The clips without subtitles are usually longer scenic sequences without dialogue.
Clips with subtitles will include the sequence number of the subtitle and the subtitle text in their filename.
Clips without subtitles will have a sequence number equal to one less than the subtitle appearing after them and will all be named "zzz," as if the actors are taking naps between dialogue.
This program expects the subtitle file to be in the SubRup format. What follows is my own explanation of the file format, but a more exact definition may be found on the SubRip wikipedia article.
Here is an example of an actual subtitle entry:
1
00:01:30,376 --> 00:01:31,626
Hello World.
Welcome to my movie.
Here are the abstract components of the entry:
[subtitle_sequence_number]
[timestamp_start] --> [timestamp_end]
Subtitle text
Possible spanning multiple lines.
[a blank line]
The subtitle_sequence_number should be an int and is used to keep the extracted scenes in order. The timestamp_start and timestamp_end represent when the subtitle should appear on the screen and when it should disappear. These time are separated by the --> symbol.
Each timestamp is of the form:
[hour]:[minute]:[second],[millisecond]
Here is a sample of actual subtitles for the beginning of Moonrise Kingdom:
1
00:01:30,376 --> 00:01:31,626
In order to show you how
2
00:01:31,709 --> 00:01:34,209
a big symphony
orchestra is put together,
3
00:01:34,292 --> 00:01:37,376
Benjamin Britten
has written a big piece of music,
4
00:01:37,459 --> 00:01:39,209
which is made up
of smaller pieces
5
00:01:39,292 --> 00:01:42,292
that show you all
the separate parts of the orchestra.
6
00:01:42,376 --> 00:01:45,167
These smaller pieces
are called variations,
7
00:01:45,251 --> 00:01:48,751
which means different ways
of playing the same tune.
8
00:01:48,876 --> 00:01:52,084
First of all,
he lets us hear the tune or the theme,
9
00:01:52,167 --> 00:01:53,292
which is a beautiful melody
10
00:01:53,376 --> 00:01:56,751
by the much older
British composer Henry Purcell.