Part of the CineJSON family of film/TV object notation proposals.
Author: Alex Coppen (azcoppen@protonmail.com)
SubJSON is a data model and object notation/interchange syntax for film and TV subtitles. It is offered freely for anyone to use or adapt under a Creative Commons Attribution-ShareAlike 4.0 International License.
Subtitles are an under-rated part of movie data, but a source of enormous inefficiency. The most interesting part is a literary transcription of audio information provides a way to make movies searchable.
There are at least 8 different subtitle formats that are regularly in use. SubRip (.srt) is by far the most widely adopted for downloaded files.
- VOB (DVD)
- PGS (BluRay)
- SubRip (.srt)
- WebVTT (.vtt)
- Substation Alpha (.ass)
- Youtube Subtitles (.sbv)
- JSON (TED.com) Subtitles (.json)
- TTML (.dfxp)
Example .srt entry:
119
00:08:51,565 --> 00:08:54,943
So, when Andy Dufresne came to me in 1949
120
00:08:55,027 --> 00:08:58,363
and asked me to smuggle Rita Hayworth
into the prison for him,
121
00:08:58,447 --> 00:09:00,532
I told him, "No problem."
To mine or search subtitle transcription content, access needs to be in a format that is easily deserialized into a simple and usable object model. Subtitles, like screenplays, are formatted in a form of markup that describes the timing of an overlay, but nothing about the content inside. Accessing it as data allows more sophisticated interchange across programs and platforms, as well as enabling dynamic document generation/manipulation.
It makes no sense at all to have multiple country files for the same film. Notwithstanding translations happening at different times in different places, having 1 file per device or release seems considerably more practical.
The sheer amount of diferent versions is baffling:
http://www.yifysubtitles.com/movie-imdb/tt0111161
A release of a film in France may be different from that in the US. A Blu-Ray Special Edition may be different from the US DVD. A "Director's Cut" may be a totally different length.
Each translated file stores its cues in milliseconds. Unfortunately, each language file annotates the same dalogue and/or screen text in a less-than-forensic way - meaning entries do not match up across countries, and are usually 100-400ms off. Timecode "start" events are not synchronised at all.
The main document container is a simplistic header that provides metadata about the document, as well as information on visual presentation (templates, individual styles).
The timed overlay entries are stored as an array in the data attribute.
{
"id": "33834784375375",
"title": "The Shawshank Redemption",
"format": "SubRip",
"templates": {
"default": "__CONTENT__",
"italic": "<i>__CONTENT__<\/i>"
},
"styles": {
"default": "font-style: 10px; line-height: 1; color: #FFF;"
},
"data": []
}Screen events are an array of overlay objects, ordered ASC by their starting timecode (in milliseconds).
A timecoded subtitle overlay is represented as a common JSON object:
- trigger: The time in milliseconds the screen event cue occurs.
- lang: The ISO language code of the content to be displayed.
- styles: Which style classes the playback device should apply to the content.
- templates: Which display templates the playback device should apply to the content (queued).
- start: A mathematical breakdown of the time the screen event should be triggered.
- end: A mathematical breakdown of the time the screen event should finish and no longer display.
- duration: A mathematical breakdown of the screen event's lifetime.
- content: The LTR/RTL string that should be displayed.
- meta: Any custom data that can be added to the overlay event for the bbenefit of the playback device,
{
"trigger": 4736088,
"lang": "en",
"styles": [
"default"
],
"templates": [
"default"
],
"start": {
"time": 4736088,
"hour": 1,
"mins": 18,
"secs": 56,
"ms": 88
},
"end": {
"time": 4739256,
"hour": 1,
"mins": 18,
"secs": 59,
"ms": 256
},
"duration": {
"secs": 3.168,
"ms": 3168
},
"content": "There are a hundred different ways to skim off the top.",
"meta": {
"original": {
"start": "01:18:56,088",
"end": "01:18:59,256"
}
}
}Example assuming we have JQuery in our display environment (e.g. Node-based program like Electron, or web player like Video.js).
Example to pull all subtitles from an EN subtitle file, and print them to an ordered HTML list:
$( video_player_object ).load(function() {
$.getJSON( "subtitles/en/RAW_en_the_shawshank_redemption.json", function( file_contents ) {
var items = [];
$.each( file_contents.data, function( key, val ) {
items.push( "<li id='" + key + "'>" + val + "</li>" );
});
$( "<ul/>", {
"class": "my-new-list",
html: items.join( "" )
}).appendTo( "body" );
});
});Example to pull out all Spanish overlay data from a multi-language file that also has EN, FR, PT, and IT entries.
$( video_player_object ).load(function() {
$.getJSON( "subtitles/multi-language/RAW_multi_the_shawshank_redemption.json", function( file_contents ) {
var items = [];
$.each( file_contents.data, function( key, val ) {
call_some_function_when_time_hits(val.trigger, val.content);
// etc etc
});
});
});Once stored in a back-end NoSQL database (like MongoDB, CouchDB, Neo4j, or even HTML5 localStorage), we suddenly have a way to search the movie.
(assuming we already know the document ID is 4ecc05e55dd98a436ddcc47c).
db.subtitles.find({"_id" : ObjectId("4ecc05e55dd98a436ddcc47c"), 'data.lang':'fr'})db.subtitles.ensureIndex({content:1});
db.subtitles.find(content:"skim");db.subtitles.find({"_id" : ObjectId("4ecc05e55dd98a436ddcc47c")}).aggregate(
[
{
$group:
{
_id: "$item",
avgDuration: { $avg: "$duration" }
}
}
]
)Using Mongo's MapReduce functionality (https://docs.mongodb.com/manual/core/map-reduce/) to look at all occurrences of keywords:
db.subtitles.mapReduce(map, reduce, {out: "word_count"})Result:
db.word_count.find().sort({value:-1})
{ "_id" : "is", "value" : 3 }
{ "_id" : "bad", "value" : 2 }
{ "_id" : "good", "value" : 2 }
{ "_id" : "this", "value" : 2 }
{ "_id" : "neither", "value" : 1 }
{ "_id" : "or", "value" : 1 }
{ "_id" : "something", "value" : 1 }
{ "_id" : "that", "value" : 1 }