Build Status npm-version Dependency Status devDependency Status

Implementation of the WebVTT spec in JavaScript. Can be used
in NodeJS, on the browser, and many other places. Mozilla uses this implementation for parsing and processing WebVTT
files in Firefox/Gecko.

Table of Contents

Spec Compliance


Our processing model portion of the specification makes use of a custom property, hasBeenReset. It allows us to detect
when a VTTCue is dirty, i.e. one of its properties that affects display has changed and so we need to recompute its display
state. This allows us to reuse a cue's display state if it has already been computed and nothing has changed to effect its



The parser has a simple API:

var parser = new WebVTT.Parser(window, stringDecoder);
parser.onregion = function(region) {};
parser.oncue = function(cue) {};
parser.onflush = function() {};
parser.onparsingerror = function(e) {};

The Parser constructor is passed a window object with which it will create new
VTTCues and VTTRegions as well as an optional StringDecoder object which
it will use to decode the data that the parse() function receives. For ease of
use, a StringDecoder is provided via WebVTT.StringDecoder(). If a custom
StringDecoder object is passed in it must support the API specified by the
#whatwg string encoding spec.


Hands data in some format to the parser for parsing. The passed data format
is expected to be decodable by the StringDecoder object that it has. The parser
decodes the data and reassembles partial data (streaming), even across line breaks.

var parser = new WebVTT.Parser(window, WebVTT.StringDecoder());
parser.parse("00:32.500 --> 00:33.500 align:start size:50%\n");
parser.parse("<v.loud Mary>That's awesome!");


Indicates that no more data is expected and will force the parser to parse any
unparsed data that it may have. Will also trigger onflush.


Callback that is invoked for every region that is correctly parsed. Returns a VTTRegion

parser.onregion = function(region) {


Callback that is invoked for every cue that is fully parsed. In case of streaming parsing oncue is
delayed until the cue has been completely received. Returns a VTTCue object.

parser.oncue = function(cue) {


Is invoked in response to flush() and after the content was parsed completely.

parser.onflush = function() {


Is invoked when a parsing error has occured. This means that some part of the
WebVTT file markup is badly formed. See ParsingError for more

parser.onparsingerror = function(e) {

WebVTT.convertCueToDOMTree(window, cuetext)

Parses the cue text handed to it into a tree of DOM nodes that mirrors the internal WebVTT node structure of
the cue text. It uses the window object handed to it to construct new HTMLElements and returns a tree of DOM
nodes attached to a top level div.

var div = WebVTT.convertCueToDOMTree(window, cuetext);

WebVTT.processCues(window, cues, overlay)

Converts the cuetext of the cues passed to it to DOM trees—by calling convertCueToDOMTree—and
then runs the processing model steps of the WebVTT specification on the divs. The processing model applies the necessary
CSS styles to the cue divs to prepare them for display on the web page. During this process the cue divs get added
to a block level element (overlay). The overlay should be a part of the live DOM as the algorithm will use the
computed styles (only of the divs to do overlap avoidance.

var divs = WebVTT.processCues(window, cues, overlay);


A custom JS error object that is reported through the
onparsingerror callback. It has a name, code, and
message property, along with all the regular properties that come with a
JavaScript error object.

  "name": "ParsingError",
  "code": "SomeCode",
  "message": "SomeMessage"

There are two error codes that can be reported back currently:

  • 0 BadSignature
  • 1 BadTimeStamp

Note: Exceptions other then ParsingError will be thrown and not reported.


A DOM shim for the VTTCue. See the spec
for more information. Our VTTCue shim also includes properties of its abstract base class

var cue = new window.VTTCue(0, 1, "I'm a cue.");

Note: Since this polfyill doesn't implement the track specification directly the onenter
and onexit events will do nothing and do not exist on this shim.

Extended API

There is also an extended version of this shim that gives a few convenience methods
for converting back and forth between JSON and VTTCues. If you'd like to use these
methods then us vttcue-extended.js instead of vttcue.js. This isn't normally
built into the vtt.js distributable so you will have to build a custom distribution
instead of using bower.


Converts a cue to JSON.

var json = cue.toJSON();


Create and initialize a VTTCue from JSON.

var cue = VTTCue.fromJSON(json);


Initializes a VTTCue from an options object where the properties in the option
objects are the same as the properties on the VTTCue.

var cue = VTTCue.create(options);


A DOM shim for the VTTRegion. See the spec
for more information.

var region = new window.VTTRegion(0, 1, "I'm a region.");
cue.region = region;

Extended API

There is also an extended version of this shim that gives a few convenience methods
for converting back and forth between JSON and VTTRegions. If you'd like to use these
methods then us vttregion-extended.js instead of vttregion.js. This isn't normally
built into the vtt.js distributable so you will have to build a custom distribution
instead of using bower.


Creates and initializes a VTTRegion from JSON.

var region = VTTRegion.fromJSON(json);


Creates a VTTRegion from an options object where the properties on the options
object are the same as the properties on the VTTRegion.

var region = VTTRegion.create(options);


In order to use the vtt.js in a browser, you need to get the built distribution of vtt.js. The distribution
contains polyfills for TextDecoder, VTTCue,
and VTTRegion since not all browsers currently
support them.

Building Yourself

Building a browser-ready version of the library is done using grunt (if you haven't installed
grunt globally, you can run it from ./node_modules/.bin/grunt after running npm install):

$ grunt build
$ Running "uglify:dist" (uglify) task
$ File "dist/vtt.min.js" created.

$ Running "concat:dist" (concat) task
$ File "dist/vtt.js" created.

$ Done, without errors.

Your newly built vtt.js now lives in dist/vtt.min.js, or alternatively, dist/vtt.js for an
unminified version.


You can also get the a prebuilt distribution from Bower. Either run the shell

$ bower install vtt.js

Or include vtt.js as a dependency in your bower.json file. vtt.js should now
live in bower_components/vtt.js/vtt.min.js. There is also an unminified
version included with bower at bower_components/vtt.js/vtt.js.


To use vtt.js you can just include the script on an HTML page like so:

  <meta charset="utf-8">
  <title>vtt.js in the browser</title>
  <script src="bower_components/vtt.js/vtt.min.js"></script>
    var vtt = "WEBVTT\n\nID\n00:00.000 --> 00:02.000\nText",
        parser = new WebVTT.Parser(window, WebVTT.StringDecoder()),
        cues = [],
        regions = [];
    parser.oncue = function(cue) {
    parser.onregion = function(region) {

    var div = WebVTT.convertCueToDOMTree(window, cues[0].text);
    var divs = WebVTT.processCues(window, cues, document.getElementById("overlay"));
  <div id="overlay" style="position: relative; width: 300px; height: 150px"></div>


You have a couple of options if you'd like to run the library from Node.


vtt.js is on npm. Just do:

npm install vtt.js

Require it and use it:

var vtt = require("vtt.js"),
    WebVTT = vtt.WebVTT,
    VTTCue = vtt.VTTCue,
    VTTRegion = vtt.VTTRegion;

var parser = new WebVTT.Parser(window);
// etc

var elements = WebVTT.processCues(window, cues, overlay);
var element = WebVTT.convertCueToDOMTree(window, cuetext);

var cue = new VTTCue(0, 1, "I'm a cue.");
var region = new VTTRegion();

See the API for more information on how to use it.

Note: If you use this method you will have to provide your own window object
or a shim of one with the necessary functionality for either the parsing or processing
portion of the spec. The only shims that are provided to you are VTTCue and VTTRegion
which you can attach to your global that is passed into the various functions.


Use node-vtt. Node-vtt runs vtt.js on a PhantomJS page
from Node so it has access to a full DOM and CSS layout engine which means you can run any part
of the library you want. See the node-vtt repo for more

Developing vtt.js

A few things to note:

  • When bumping the version remember to use the grunt release task as this will
    bump package.json + bower.json and build the dist files for vtt.js in one
  • The Grunt Run Task tool is handy for running the library without having
    to run the whole test suite or set of tests.


Tests are written and run using Mocha on node.js.

To run all the tests, do the following:

$ npm test

If you want to run individual tests, you can install the Mocha command-line
tool globally, and then run tests per-directory:

$ npm install -g mocha
$ cd tests/some/sub/dir
$ mocha --reporter spec --timeout 200000

See the usage docs for further usage info.

Writing Tests

Tests are done by comparing live parsed output to a last-known-good JSON file. The JSON files
can be easily generated using vtt.js, so you don't need to write these by hand
(see details below about Grunt Run Task).


There's a prebuilt API in place for testing different parts of vtt.js. Simply
require the TestRunner
module in the lib directory and start writing tests using mocha. See an example of a test file
with its first test's WebVTT file here
and its corresponding parsing JSON file
and processing JSON file.
You can also check out the tests
directory for more examples on how to write tests.

jsonEqual(vttFile, jsonRefFile, message, onDone)

First parses the WebVTT file as UTF8 and compares it to the reference JSON file
and then parses the WebVTT file as a string and compares it to the reference JSON

jsonEqualStreaming(vttFile, jsonRefFile, message, onDone)

Simulates parsing the file while streaming by splitting the WebVTT file into
chunks. Will simulate parsing like this n times for a single WebVTT file where
n is the length in unicode characters of the file, so use this only on small
files or else you will get a timeout failure on your test.

jsonEqualParsing(vttFile, jsonRefFile, message, onDone)

Runs jsonEqual and jsonEqualStreaming in one go.

jsonEqualProcModel(vttFile, jsonRefFile, message, onDone)

Runs the processing model over the VTTCues and VTTRegions that are returned
from parsing the WebVTT file.

jsonEqualAll(vttFile, jsonRefFile, message, onDone)

Runs jsonEqualParsing and jsonEqualProcModel. Note that jsonRefFile should
contain JSON that is generated from parsing. The processing model test will compare
its results to a JSON file located at [vttFile]-proc.json. Therefore, if you
have a WebVTT file named basic.vtt the JSON reference file for the processing
model tests will be basic-proc.json.

jsonEqualAllNoStream(vttFile, jsonRefFile, message, onDone)

Runs jsonEqual and jsonEqualProcModel use this if you want to do parsing
and processing tests, but do not want to simulate streaming because you
have too big of a WebVTT file.

Grunt Run Task

You can automatically generate a JSON file for a given .vtt file using the
run Grunt task.

To get parsed JSON output from some WebVTT file do:

$ grunt run:my-vtt-file.vtt
$ grunt run:my-vtt-file.vtt > my-json-file.json

To get processed output from the WebVTT file do:

$ grunt run:my-vtt-file.vtt:p
$ grunt run:my-vtt-file.vtt:p > my-json-file.json

By passing the c flag you can automatically copy the output into a JSON file
with the same name as the WebVTT file:

$ grunt run:my-vtt-file.vtt:c
$ grunt run:my-vtt-file.vtt:pc

The parsed JSON output now lives in my-vtt-file.json and the processing JSON
output lives in my-vtt-file-proc.json.

You can also run it over a directory copying the output of parsing or
processing each WebVTT file to a corresponding JSON file like so:

$ grunt run:my-vtt-file-directory
$ grunt run:my-vtt-file-directory:p

This is useful when you've modified how vtt.js works and each JSON file needs
a slight change.

The run task utilizes a script called cue2json, but
does a few other things for you before each run like building a development
build for cue2json to use. It's also a bit easier to type in the CL options
for the task. If you want to know more about cue2json you can run it directly
like so:

$ ./bin/cue2json.js 
$ Generate JSON test files from a reference VTT file.
$ Usage: node ./bin/cue2json.js [options]
$ Options:
$   -v, --vtt      Path to VTT file.                                                                                     
$   -d, --dir      Path to test directory. Will recursively find all JSON files with matching VTT files and rewrite them.
$   -c, --copy     Copies the VTT file to a JSON file with the same name.                                                
$   -p, --process  Generate a JSON file of the output returned from the processing model. 


  • cue2json utilizes the last development build done. This is why the Grunt run task is
    good as you don't have to remember to build it yourself. If you don't build it yourself then you could
    potentially get incorrect results from it.
  • Since cue2json uses the actual parser to generate these JSON files there is the possibility that
    the generated JSON will contain bugs. Therefore, always check the generated JSON files to check that the
    parser actually parsed according to spec.