For the entire publication on the W3C site the liability and trademark rules apply.
Abstract
This document defines a set of JavaScript APIs that allow local media, including audio and video, to be requested from a platform.
Status of This Document
This section describes the status of this document at the time of its publication. Other documents may supersede this document. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at https://www.w3.org/TR/.
This document is not complete. It is subject to major changes and, while early experimentations are encouraged, it is therefore not intended for implementation. The API is based on preliminary work done in the WHATWG.
Publication as an Editor's Draft does not imply endorsement by the W3C Membership. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than work in progress.
This document defines APIs for requesting access to local multimedia devices, such as microphones or video cameras.
This document also defines the MediaStream API, which provides the means to control where multimedia stream data is consumed, and provides some control over the devices that produce the media. It also exposes information about devices able to capture and render media.
2. Conformance
As well as sections marked as non-normative, all authoring guidelines, diagrams, examples, and notes in this specification are non-normative. Everything else in this specification is normative.
The key words MAY, MUST, MUST NOT, NOT REQUIRED, and SHOULD are to be interpreted as described in [RFC2119].
This specification defines conformance criteria that apply to a single product: the User Agent that implements the interfaces that it contains.
Conformance requirements phrased as algorithms or specific steps may be implemented in any manner, so long as the end result is equivalent. (In particular, the algorithms defined in this specification are intended to be easy to follow, and not intended to be performant.)
Implementations that use ECMAScript [ECMA-262] to implement the APIs defined in this specification must implement them in a manner consistent with the ECMAScript Bindings defined in the Web IDL specification [
WEBIDL-1], as this specification uses that specification and terminology.
3. Terminology
HTML Terms:
The EventHandler interface represents a callback used for event handlers as defined in [
HTML5].
A source is the "thing" providing the source of a media stream track. The source is the broadcaster of the media itself. A source can be a physical webcam, microphone, local video or audio file from the user's hard drive, network resource, or static image. Note that this document describes the use of microphone and camera type sources only, the use of other source types is described in other documents.
An application that has no prior authorization regarding sources is only given the number of available sources, their type and any relationship to other devices. Additional information about sources can become available when applications are authorized to use a source (see
9.2.1Access control model).
Sources do not have constraints — tracks have constraints. When a source is connected to a track, it must produce media that conforms to the constraints present on that track. Multiple tracks can be attached to the same source. User Agent processing, such as downsampling, MAY be used to ensure that all tracks have appropriate media.
Sources have constrainable properties which have
capabilities and settings. The constrainable properties are "owned" by the source and are common to any (multiple) tracks that happen to be using the same source (e.g., if two different track objects bound to the same source ask for the same capability or setting information, they will get back the same answer).
Setting (Source Setting)
A setting refers to the immediate, current value of the source's constrainable properties. Settings are always read-only.
A source's settings can change dynamically over time due to environmental conditions, sink configurations, or constraint changes. A source's settings must always conform to the current set of mandatory constraints on all attached tracks. A source that cannot conform to mandatory constraints causes affected tracks to become
overconstrained and therefore muted. A User Agent attempts to ensure that sources adhere to optional constraints as closely as possible, see 11.Constrainable Pattern.
Although settings are a property of the source, they are only exposed to the application through the tracks attached to the source. This is exposed via the ConstrainablePattern interface.
Capabilities
For each constrainable property, there is a capability that describes whether it is supported by the source and if so, the range of supported values. As with settings, capabilities are exposed to the application via the ConstrainablePattern interface.
The values of the supported capabilities must be normalized to the ranges and enumerated types defined in this specification.
A getCapabilities() call on a track returns the same underlying per-source capabilities for all tracks connected to the source.
Source capabilities are effectively constant. Applications should be able to depend on a specific source having the same capabilities for any browsing session.
This API is intentionally simplified. Capabilities are not capable of describing interactions between different values. For instance, it is not possible to accurately describe the capabilities of a camera that can produce a high resolution video stream at a low frame rate and lower resolutions at a higher frame rate. Capabilities describe the complete range of each value. Interactions between constraints are exposed by attempting to apply constraints.
Constraints
Constraints provide a general control surface that allows applications to both select an appropriate source for a track and, once selected, to influence how a source operates.
Constraints limit the range of operating modes that a source can use when providing media for a track. Without provided track constraints, implementations are free to select a source's settings from the full ranges of its supported capabilities. Implementations may also adjust source settings at any time within the bounds imposed by all applied constraints.
getUserMedia() uses constraints to help select an appropriate source for a track and configure it. Additionally, the
ConstrainablePattern interface on tracks includes an API for dynamically changing the track's constraints at any later time.
A track will not be connected to a source using
getUserMedia() if its initial constraints cannot be satisfied. However, the ability to meet the constraints on a track can change over time, and constraints can be changed. If circumstances change such that constraints cannot be met, the ConstrainablePattern interface defines an appropriate error to inform the application. 5.The model: sources, sinks, constraints, and settings explains how constraints interact in more detail.
In general, User Agents will have more flexibility to optimize the media streaming experience the fewer constraints are applied, so application authors are strongly encouraged to use mandatory constraints sparingly.
For each constrainable property, a constraint exists whose name corresponds with the relevant source setting name and capability name.
The terms permission, retrieve the permission
state, request permission and
create a permission storage entry are defined in [permissions].
Note
"request permission" isn't defined yet. Filed as permissions issue 62.
4. MediaStream API
4.1 Introduction
The two main components in the MediaStream API are the
MediaStreamTrack and MediaStream
interfaces. The MediaStreamTrack object represents media of a single type that originates from one media source in the User Agent, e.g. video produced by a web camera. A
MediaStream is used to group several
MediaStreamTrack objects into one unit that can be recorded or rendered in a media element.
Each MediaStream can contain zero or more
MediaStreamTrack objects. All tracks in a
MediaStream are intended to be synchronized when rendered. This is not a hard requirement, since it might not be possible to synchronize tracks from sources that have different clocks. Different
MediaStream objects do not need to be synchronized.
Note
While the intent is to synchronize tracks, it could be better in some circumstances to permit tracks to lose synchronization. In particular, when tracks are remotely sourced and real-time [WEBRTC10], it can be better to allow loss of synchronization than to accumulate delays or risk glitches and other artifacts. Implementations are expected to understand the implications of choices regarding synchronization of playback and the effect that these have on user perception.
A single MediaStreamTrack can represent multi-channel content, such as stereo or 5.1 audio or stereoscopic video, where the channels have a well defined relationship to each other. Information about channels might be exposed through other APIs, such as [
WEBAUDIO], but this specification provides no direct access to channels.
A MediaStream object has an input and an output that represent the combined input and output of all the object's tracks. The output of the MediaStream controls how the object is rendered, e.g., what is saved if the object is recorded to a file or what is displayed if the object is used in a video element. A single MediaStream object can be attached to multiple different outputs at the same time.
A new MediaStream object can be created from existing media streams or tracks using the
MediaStream() constructor. The constructor argument can either be an existing MediaStream object, in which case all the tracks of the given stream are added to the new
MediaStream object, or an array of
MediaStreamTrack objects. The latter form makes it possible to compose a stream from different source streams.
Both MediaStream and
MediaStreamTrack objects can be cloned. A cloned
MediaStream contains clones of all member tracks from the original stream. A cloned MediaStreamTrack has a
set of constraints that is independent of the instance it is cloned from, which allows media from the same source to have different constraints applied for different
consumers. The MediaStream object is also used in contexts outside getUserMedia, such as [WEBRTC10].
4.2 MediaStream
The MediaStream() constructor composes a new stream out of existing tracks. It takes an optional argument of type MediaStream or an array of
MediaStreamTrack objects. When the constructor is invoked, the User Agent must run the following steps:
Let stream be a newly constructed
MediaStream object.
Initialize stream'sid attribute to a newly generated value.
If the constructor's argument is present, construct a set of tracks, tracks based on the type of argument:
The tracks of a MediaStream are stored in a
track set. The track set MUST contain the
MediaStreamTrack objects that correspond to the tracks of the stream. The relative order of the tracks in the set is User Agent defined and the API will never put any requirements on the order. The proper way to find a specific MediaStreamTrack
object in the set is to look it up by its id.
An object that reads data from the output of a
MediaStream is referred to as a
MediaStream consumer. The list of
MediaStream consumers currently include media elements (such as <video> and
<audio>) [HTML5], Web Real-Time Communications (WebRTC; RTCPeerConnection) [WEBRTC10], media recording (MediaRecorder) [mediastream-recording], image capture (ImageCapture) [image-capture], and web audio (MediaStreamAudioSourceNode) [WEBAUDIO].
Note
MediaStream consumers must be able to handle tracks being added and removed. This behavior is specified per consumer.
A MediaStream object is said to be active when it has at least one
MediaStreamTrack that has not ended. A MediaStream that does not have any tracks or only has tracks that are ended is inactive.
The User Agent may update a MediaStream's track set in response to, for example, an external event. This specification does not specify any such cases, but other specifications using the MediaStream API may. One such example is the WebRTC 1.0 [WEBRTC10] specification where the track set of a MediaStream, received from another peer, can be updated as a result of changes to the media session.
When the User Agent initiates adding a track to a
MediaStream, with the exception of initializing a newly created MediaStream with tracks, the User Agent
MUST queue a task that runs the following steps:
Let track be the MediaStreamTrack
in question and stream the MediaStream
object to which track is to be added.
If track is already in stream'strack set, then abort these steps.
When a MediaStream object is created, the User Agent MUST generate an identifier string, and MUST initialize the object's id attribute to that string. A good practice is to use a UUID [rfc4122], which is 36 characters long in its canonical form. To avoid fingerprinting, implementations SHOULD use the forms in section 4.4 or 4.5 of RFC 4122 when generating UUIDs.
The id attribute MUST return the value to which it was initialized when the object was created.
active of type boolean, readonly
The active attribute MUST return
true if this MediaStream is
active and false otherwise.
The event type of this event handler is removetrack.
Methods
getAudioTracks
Returns a sequence of MediaStreamTrack
objects representing the audio tracks in this stream.
The getAudioTracks method MUST return a sequence that represents a snapshot of all the MediaStreamTrack objects in this stream's
track set whose kind is equal to "
audio". The conversion from the track set to the sequence is User Agent defined and the order does not have to be stable between calls.
No parameters.
Return type:sequence<MediaStreamTrack>
getVideoTracks
Returns a sequence of MediaStreamTrack
objects representing the video tracks in this stream.
The getVideoTracks method MUST return a sequence that represents a snapshot of all the MediaStreamTrack objects in this stream's
track set whose kind is equal to "
video". The conversion from the track set to the sequence is User Agent defined and the order does not have to be stable between calls.
No parameters.
Return type:sequence<MediaStreamTrack>
getTracks
Returns a sequence of MediaStreamTrack
objects representing all the tracks in this stream.
The getTracks method
MUST return a sequence that represents a snapshot of all the
MediaStreamTrack objects in this stream's
track set, regardless of kind. The conversion from the track set to the sequence is User Agent defined and the order does not have to be stable between calls.
No parameters.
Return type:sequence<MediaStreamTrack>
getTrackById
The getTrackById method MUST return either a MediaStreamTrack
object from this stream's track set whose id is equal to trackId, or null, if no such track exists.
A MediaStreamTrack object represents a media source in the User Agent. An example source is a device connected to the User Agent. Other specifications may define sources for
MediaStreamTrack that override the behavior specified here. Several MediaStreamTrack objects can represent the same media source, e.g., when the user chooses the same camera in the UI shown by two consecutive calls to getUserMedia()
.
The data from a MediaStreamTrack object does not necessarily have a canonical binary form; for example, it could just be "the video currently coming from the user's video camera". This allows User Agents to manipulate media in whatever fashion is most suitable on the user's platform.
A script can indicate that a MediaStreamTrack
object no longer needs its source with the stop() method. When all tracks using a source have been stopped or ended by some other means, the source is stopped. If the source is a device exposed by getUserMedia(), then when the source is stopped, the UA MUST run the following steps:
Let deviceId be the device's deviceId.
Set [[devicesLiveMap]][deviceId] to false.
If the result of
retrieving the permission state of the permission associated with the device's kind and deviceId, is not “granted”, then set [[devicesAccessibleMap]]
[deviceId] to false.
An implementation may use a per-source reference count to keep track of source usage, but the specifics are out of scope for this specification.
To clone a track the User Agent MUST run the following steps:
Initialize trackClone's id attribute to a newly generated value.
Initialize trackClone's kind, label, readyState, and
enabled attributes by copying the corresponding values from
track.
Let trackClone's underlying source be the source of
track.
Set trackClone's constraints to the active constrains of track.
Return trackClone.
4.3.1 Life-cycle and Media Flow
Life-cycle
A MediaStreamTrack has two states in its life-cycle: live and ended. A newly created
MediaStreamTrack can be in either state depending on how it was created. For example, cloning an ended track results in a new ended track. The current state is reflected by the object's
readyState attribute.
In the live state, the track is active and media is available for use by consumers (but may be replaced by zero-information-content if the MediaStreamTrack is
muted or disabled, see below).
A muted or disabled MediaStreamTrack renders either silence (audio), black frames (video), or a zero-information-content equivalent. For example, a video element sourced by a muted or disabled MediaStreamTrack
(contained within a MediaStream ), is playing but the rendered content is the muted output.
If the source is a device exposed by getUserMedia(), then when a track becomes either muted or disabled, and this brings all tracks connected to the device to be either muted, disabled, or stopped, then the UA MAY, using the device's deviceId, deviceId, set [[devicesLiveMap]]
[deviceId] to
false, provided the UA sets it back to true as soon as any unstopped track connected to this device becomes un-muted or enabled again.
The muted/unmuted state of a track reflects whether the source provides any media at this moment. The enabled/disabled state is under application control and determines whether the track outputs media (to its consumers). Hence, media from the source only flows when a
MediaStreamTrack object is both unmuted and enabled.
A MediaStreamTrack is muted when the source is temporarily unable to provide the track with data. A track can be muted by a user. Often this action is outside the control of the application. This could be as a result of the user hitting a hardware switch or toggling a control in the operating system / browser chrome. A track can also be muted by the User Agent.
Applications are able to enable or disable a MediaStreamTrack to prevent it from rendering media from the source. A muted track will however, regardless of the enabled state, render silence and blackness. A disabled track is logically equivalent to a muted track, from a consumer point of view.
For a newly created MediaStreamTrack object, the following applies. The track is always enabled unless stated otherwise (for example when cloned) and the muted state reflects the state of the source at the time the track is created.
A MediaStreamTrack object is said to
end when the source of the track is disconnected or exhausted.
When a MediaStreamTrack object ends for any reason (e.g., because the user rescinds the permission for the page to use the local camera, or because the application invoked the
stop() method on the MediaStreamTrack object, or because the User Agent has instructed the track to end for any reason) it is said to be
ended.
When a MediaStreamTracktrack ends for any reason other than the stop() method being invoked, the User Agent MUST queue a task that runs the following steps:
If the track'sreadyState attribute has the value ended already, then abort these steps.
Muted is out of control for the application, but can be observed by the application by reading the muted attribute and listening to the associated events mute and unmute. There can be several reasons for a MediaStreamTrack to be muted: the user pushing a physical mute button on the microphone, the user toggling a control in the operating system, the user clicking a mute button in the browser chrome, the User Agent (on behalf of the user) mutes, etc.
To update a track's muted state to newState, the User Agent MUST queue a task to run the following steps:
Let track be the MediaStreamTrack in question.
Set track's muted attribute to
newState.
If newState is true let
eventName be mute, otherwise
unmute.
Fire a simple event named eventName on
track.
Enabled/disabled on the other hand is available to the application to control (and observe) via the
enabled attribute.
The result for the consumer is the same in the sense that whenever
MediaStreamTrack is muted or disabled (or both) the consumer gets zero-information-content, which means silence for audio and black frames for video. In other words, media from the source only flows when a MediaStreamTrack object is both unmuted and enabled. For example, a video element sourced by a muted or disabled MediaStreamTrack (contained in a
MediaStream ), is playing but rendering blackness.
For a newly created MediaStreamTrack object, the following applies: the track is always enabled unless stated otherwise (for example when cloned) and the muted state reflects the state of the source at the time the track is created.
Whether Constraints were provided at track initialization time or need to be established later at runtime, the APIs defined in the ConstrainablePattern Interface allow the retrieval and manipulation of the constraints currently established on a track.
If the overconstrained event is thrown, the track MUST be muted until either new satisfiable constraints are applied or the existing constraints become satisfiable.
The kind attribute
MUST return the string "audio" if this object represents an audio track or "video" if this object represents a video track.
id of type DOMString, readonly
Unless a MediaStreamTrack object is created as a part of a special purpose algorithm that specifies how the track id must be initialized, the User Agent MUST generate an identifier string and initialize the object's
id attribute to that string. See MediaStream's
id attribute for guidelines on how to generate such an identifier.
An example of an algorithm that specifies how the track id must be initialized is the algorithm to represent an incoming network component with a MediaStreamTrack
object. [WEBRTC10]
id attribute MUST return the value to which it was initialized when the object was created.
label of type DOMString, readonly
User Agents MAY label audio and video sources (e.g., "Internal microphone" or "External USB Webcam"). The label attribute
MUST return the label of the object's corresponding source, if any. If the corresponding source has or had no label, the attribute MUST instead return the empty string.
enabled of type boolean
The enabled attribute controls the enabled state for the object.
On getting, the attribute MUST return the value to which it was last set. On setting, it MUST be set to the new value.
Note
Thus, after a
MediaStreamTrack has ended, its enabled attribute still changes value when set; it just doesn't do anything with that new value.
muted of type boolean, readonly
The muted attribute
MUST return true if the track is muted, and false otherwise.
The track is active (the track's underlying media source is making a best-effort attempt to provide data in real time).
The output of a track in the live state can be switched on and off with the enabled attribute.
ended
The track has ended (the track's underlying media source is no longer providing data, and will never provide more data for this track). Once a track enters this state, it never exits it.
For example, a video track in a
MediaStream ends when the user unplugs the USB web camera that acts as the track's media source.
4.3.4 MediaTrackSupportedConstraints
MediaTrackSupportedConstraints represents the list of constraints recognized by a User Agent for controlling the
Capabilities of a MediaStreamTrack object. This dictionary is used as a function return value, and never as an operation argument.
Future specifications can extend the MediaTrackSupportedConstraints dictionary by defining a partial dictionary with dictionary members of type boolean.
A camera can report multiple facing modes. For example, in a high-end telepresence solution with several cameras facing the user, a camera to the left of the user can report both "left" and "user". See facingMode for additional details.
If the source cannot do echo cancellation a single
false is reported. If echo cancellation cannot be turned off, a single true is reported. If the script can control the feature, the source reports a list with both true and false as possible values. See echoCancellation for additional details.
The names of the initial set of constrainable properties for MediaStreamTrack are defined below.
The following constrainable properties are defined to apply to both video and audio MediaStreamTrack objects:
Property Name
Values
Notes
deviceId
DOMString
The origin-unique identifier for the source of the
MediaStreamTrack. The same identifier MUST be valid between browsing sessions of this origin, but MUST also be different for other origins. Some sort of GUID is recommended for the identifier. Note that the setting of this property is uniquely determined by the source that is attached to the
MediaStreamTrack. In particular,
getCapabilities() will return only a single value for deviceId. This property can therefore be used for initial media selection with getUserMedia(). However, it is not useful for subsequent media control with applyConstraints(), since any attempt to set a different value will result in an unsatisfiable ConstraintSet.
groupId
DOMString
The browsing session-unique group identifier for the source of the MediaStreamTrack. Two devices have the same group identifier if they belong to the same physical device; for example, the audio input and output devices representing the speaker and microphone of the same headset would have the same groupId. Note that the setting of this property is uniquely determined by the source that is attached to the
MediaStreamTrack. In particular,
getCapabilities() will return only a single value for groupId. Since this property is not stable between browsing sessions its usefulness for initial media selection with
getUserMedia() is limited. It is not useful for subsequent media control with applyConstraints(), since any attempt to set a different value will result in an unsatisfiable ConstraintSet.
The following constrainable properties are defined to apply only to video MediaStreamTrack objects:
The width or width range, in pixels. As a capability, the range should span the video source's pre-set width values with min being the smallest width and max being the largest width.
The height or height range, in pixels. As a capability, the range should span the video source's pre-set height values with min being the smallest height and max being the largest height.
The exact frame rate (frames per second) or frame rate range. If this frame rate cannot be determined (e.g. the source does not natively provide a frame rate, or the frame rate cannot be determined from the source stream), then this value MUST refer to the User Agent's vsync display rate.
The exact aspect ratio (width in pixels divided by height in pixels, represented as a double rounded to the tenth decimal place) or aspect ratio range.
This string (or each string, when a list) should be one of the members of VideoFacingModeEnum. The members describe the directions that the camera can face, as seen from the user's perspective. Note that getConstraints may not return exactly the same string for strings not in this enum. This preserves the possibility of using a future version of WebIDL enum for this property.
The volume or volume range, as a multiplier of the linear audio sample values. A volume of 0.0 is silence, while a volume of 1.0 is the maximum supported volume. A volume of 0.5 will result in an approximately 6 dBSPL change in the sound pressure level from the maximum volume. Note that any ConstraintSet that specifies values outside of this range of 0 to 1 can never be satisfied.
When one or more audio streams is being played in the processes of various microphones, it is often desirable to attempt to remove the sound being played from the input signals recorded by the microphones. This is referred to as echo cancellation. There are cases where it is not needed and it is desirable to turn it off so that no audio artifacts are introduced. This allows applications to control this behavior.
The latency or latency range, in seconds. The latency is the time between start of processing (for instance, when sound occurs in the real world) to the data being available to the next step in the process. Low latency is critical for some applications; high latency may be acceptable for other applications because it helps with power constraints. The number is expected to be the target latency of the configuration; the actual latency may show some variation from that.
The addtrack and removetrack events notify the script that the track set of a
MediaStream has been updated by the User Agent.
Firing a track event named
e with a MediaStreamTracktrack means that an event with the name e, which does not bubble (except where otherwise stated) and is not cancelable (except where otherwise stated), and which uses the
MediaStreamTrackEvent interface with the
track attribute set to track, MUST be created and dispatched at the given target.
5. The model: sources, sinks, constraints, and settings
This section is non-normative.
Browsers provide a media pipeline from sources to sinks. In a browser, sinks are the <img>, <video>, and <audio> tags. Traditional sources include streamed content, files, and web resources. The media produced by these sources typically does not change over time - these sources can be considered to be static.
The sinks that display these sources to the user (the actual tags themselves) have a variety of controls for manipulating the source content. For example, an <img> tag scales down a huge source image of 1600x1200 pixels to fit in a rectangle defined with
width="400" and height="300".
The getUserMedia API adds dynamic sources such as microphones and cameras - the characteristics of these sources can change in response to application needs. These sources can be considered to be dynamic in nature. A <video> element that displays media from a dynamic source can either perform scaling or it can feed back information along the media pipeline and have the source produce content more suitable for display.
Note
Note: This sort of feedback loop is obviously just enabling an "optimization", but it's a non-trivial gain. This optimization can save battery, allow for less network congestion, etc...
Note that MediaStream sinks (such as
<video>, <audio>, and even
RTCPeerConnection) will continue to have mechanisms to further transform the source stream beyond that which the Settings,
Capabilities, and Constraints described in this specification offer. (The sink transformation options, including those of
RTCPeerConnection, are outside the scope of this specification.)
The act of changing or applying a track constraint may affect the
settings of all tracks sharing that source and consequently all down-level sinks that are using that source. Many sinks may be able to take these changes in stride, such as the
<video> element or RTCPeerConnection. Others like the Recorder API may fail as a result of a source setting change.
The RTCPeerConnection is an interesting object because it acts simultaneously as both a sink and a source for over-the-network streams. As a sink, it has source transformational capabilities (e.g., lowering bit-rates, scaling-up / down resolutions, and adjusting frame-rates), and as a source it could have its own settings changed by a track source.
To illustrate how changes to a given source impact various sinks, consider the following example. This example only uses width and height, but the same principles apply to all of the Settings exposed in this specification. In the first figure a home client has obtained a video source from its local video camera. The source's width and height settings are 800 pixels and 600 pixels, respectively. Three
MediaStream objects on the home client contain tracks that use this same deviceId. The three media streams are connected to three different sinks: a <video> element (A), another <video> element (B), and a peer connection (C). The peer connection is streaming the source video to a remote client. On the remote client there are two media streams with tracks that use the peer connection as a source. These two media streams are connected to two <video> element sinks (Y and Z).
Note that at this moment, all of the sinks on the home client must apply a transformation to the original source's provided dimension settings. B is scaling the video down, A is scaling the video up (resulting in loss of quality), and C is also scaling the video up slightly for sending over the network. On the remote client, sink Y is scaling the video way down, while sink Z is not applying any scaling.
In response to applyConstraints() being called, one of the tracks wants a higher resolution (1920 by 1200 pixels) from the home client's video source.
Note that the source change immediately affects all of the tracks and sinks on the home client, but does not impact any of the sinks (or sources) on the remote client. With the increase in the home client source video's dimensions, sink A no longer has to perform any scaling, while sink B must scale down even further than before. Sink C (the peer connection) must now scale down the video in order to keep the transmission constant to the remote client.
While not shown, an equally valid settings change request could be made on the remote client's side. In addition to impacting sink Y and Z in the same manner as A, B and C were impacted earlier, it could lead to re-negotiation with the peer connection on the home client in order to alter the transformation that it is applying to the home client's video source. Such a change is NOT REQUIRED to change anything related to sink A or B or the home client's video source.
Note that this specification does not define a mechanism by which a change to the remote client's video source could automatically trigger a change to the home client's video source. Implementations may choose to make such source-to-sink optimizations as long as they only do so within the constraints established by the application, as the next example demonstrates.
It is fairly obvious that changes to a given source will impact sink consumers. However, in some situations changes to a given sink may also cause implementations to adjust a source's settings. This is illustrated in the following figures. In the first figure below, the home client's video source is sending a video stream sized at 1920 by 1200 pixels. The video source is also unconstrained, such that the exact source dimensions are flexible as far as the application is concerned. Two
MediaStream objects contain tracks with the same
deviceId, and those MediaStream s are connected to two different <video> element sinks A and B. Sink A has been sized to width="1920" and
height="1200" and is displaying the source's video content without any transformations. Sink B has been sized smaller and, as a result, is scaling the video down to fit its rectangle of 320 pixels across by 200 pixels down.
When the application changes sink A to a smaller dimension (from 1920 to 1024 pixels wide and from 1200 to 768 pixels tall), the browser's media pipeline may recognize that none of its sinks require the higher source resolution, and needless work is being done both on the part of the source and sink A. In such a case and without any other constraints forcing the source to continue producing the higher resolution video, the media pipeline MAY change the source resolution:
In the above figure, the home client's video source resolution was changed to the greater of that from sink A and B in order to optimize playback. While not shown above, the same behavior could apply to peer connections and other sinks.
It is possible that constraints can be applied to a track which a source is unable to satisfy, either because the source itself cannot satisfy the constraint or because the source is already satisfying a conflicting constraint. When this happens, the promise returned from
applyConstraints() will be rejected, without applying any of the new constraints. Since no change in constraints occurs in this case, there is also no required change to the source itself as a result of this condition. Here is an example of this behavior.
In this example, two media streams each have a video track that share the same source. The first track initially has no constraints applied. It is connected to sink N. Sink N has a resolution of 800 by 600 pixels and is scaling down the source's resolution of 1024 by 768 to fit. The other track has a mandatory constraint forcing off the source's fill light; it is connected to sink P. Sink P has a width and height equal to that of the source.
Now, the first track adds a mandatory constraint that the fill light should be forced on. At this point, both mandatory constraints cannot be satisfied by the source (the fill light cannot be simultaneously on and off at the same time). Since this state was caused by the first track's attempt to apply a conflicting constraint, the constraint application fails and there is no change in the source's settings nor to the constraints on either track.
Let's look at a slightly different situation starting from the same point. In this case, instead of the first track attempting to apply a conflicting constraint, the user physically locks the camera into a mode where the fill light is on. At this point the source can no longer satisfy the second track's mandatory constraint that the fill light be off. The second track is transitioned into the muted state and receives an
overconstrained event. At the same time, the source notes that its remaining active sink only requires a resolution of 800 by 600 and so it adjusts its resolution down to match (this is an optional optimization that the User Agent is allowed to make given the situation).
At this point, it is the responsibility of the application to address the problem that led to the overconstrained situation, perhaps by removing the fill light mandatory constraint on the second track or by closing the second track altogether and informing the user.
6. MediaStreams in Media Elements
Note
We are now referencing the [HTML51] specification for MediaStream playback in a HTMLMediaElement. This section, that defines MediaStream-specific exceptions and additions, is still work in progress and feedback is most welcome.
A MediaStream may be assigned to media elements. A
MediaStream is not preloadable or seekable and represents a simple, potentially infinite, linear media timeline. The timeline starts at 0 and increments linearly in real time as long as the
MediaStream is playing. The timeline does not increment when the playout of the MediaStream is paused.
User Agents that support this specification MUST support the
srcObject attribute of the HTMLMediaElement interface defined in [HTML51], which includes support for playing
MediaStream objects.
The [HTML51] document outlines how the HTMLMediaElement works with a media provider object. The following applies when the
media provider object is a MediaStream:
Whenever an [HTML51]
AudioTrack or a
VideoTrack is created, the id and
label attributes must be initialized to the corresponding attributes of the MediaStreamTrack, the kind attribute must be initialized to "main" and the language attribute to the empty string
The User Agent MUST always play the current data from the
MediaStream and MUST NOT buffer.
When the MediaStream state moves from the active to the inactive state, the User Agent
MUST raise an
ended event on the HTMLMediaElement and set its
ended attribute to true. Note that once
ended equals true the HTMLMediaElement will not play media even if new MediaStreamTrack's are added to the MediaStream (causing it to return to the active state) unless autoplay is true or the web application restarts the element, e.g., by calling
play()
Any calls to the
fastSeek method on a HTMLMediaElement must be ignored
The nature of the MediaStream places certain restrictions on the behavior and attribute values of the associated
HTMLMediaElement and on the operations that can be performed on it, as shown below:
A MediaStream is not seekable. Therefore, this attribute MUST always have the value 1.0 and any attempt to alter it
MUST be ignored. Note that this also means that the
ratechange event will not fire.
A MediaStream is not seekable. Therefore, this attribute MUST always have the value 1.0 and any attempt to alter it
MUST be ignored. Note that this also means that the
ratechange event will not fire.
Setting the loop attribute has no effect since a
MediaStream has no defined end and therefore cannot be looped.
When applicable, behavior outlined above for
HTMLMediaElement carry over to
MediaController's.
7. Error Handling
This section and its subsections extend the list of Error subclasses defined in [ES6] following the pattern for NativeError in section 19.5.6 of that specification. Assume the following:
that use of syntax such as [[something]] and %something% is as used in [ES6].
that the rules for ECMAScript standard built-in objects ([ES6], section 17) are in effect in this section.
that the new intrinsic objects %OverconstrainedError% and %OverconstrainedErrorPrototype% are available as if they had been included in ([ES6], Table 7) and all referencing sections, e.g. ([
ES6], section 8.2.2), thus behave appropriately.
7.1 ECMAScript 6 Terminology
The following terms used in this section are defined in [ES6].
The OverconstrainedError Constructor is the %OverconstrainedError% intrinsic object. When OverconstrainedError is called as a function rather than as a constructor, it creates and initializes a new OverconstrainedError object. A call of the object as a function is equivalent to calling it as a constructor with the same arguments. Thus the function call OverconstrainedError(...) is equivalent to the object creation expression new
OverconstrainedError(...) with the same arguments.
The OverconstrainedError constructor is designed to be subclassable. It may be used as the value of an extends clause of a class definition. Subclass constructors that intend to inherit the specified OverconstrainedError behaviour must include a super call to the
OverconstrainedError constructor to create and initialize the subclass instance with an [[ErrorData]] internal slot.
When the OverconstrainedError function is called with arguments constraint and message the following steps are taken:
If NewTarget is undefined, let newTarget be the active function object, else let newTarget be NewTarget.
Let O be OrdinaryCreateFromConstructor(newTarget,
"%OverconstrainedErrorPrototype%", «[[ErrorData]]» ).
ReturnIfAbrupt(O).
If constraint is not undefined, then
Let constraintDesc be the PropertyDescriptor{[[Value]]: constraint, [[Writable]]: false, [[Enumerable]]: false, [[Configurable]]: false}.
Let cStatus be DefinePropertyOrThrow(O, "
constraint", constraintDesc).
Assert: cStatus is not an abrupt completion.
If message is not undefined, then
Let msg be ToString(message).
Let msgDesc be the PropertyDescriptor{[[Value]]:
msg, [[Writable]]: true, [[Enumerable]]:
false, [[Configurable]]: true}.
Let mStatus be DefinePropertyOrThrow(O, "
message", msgDesc).
Assert: mStatus is not an abrupt completion.
Return O.
7.2.2 Properties of the OverconstrainedError Constructor
The value of the [[Prototype]] internal slot of the OverconstrainedError constructor is the intrinsic object %Error%.
Besides the length property (whose value is 1), the OverconstrainedError constructor has the following properties:
7.2.2.1 OverconstrainedError.prototype
The initial value of OverconstrainedError.prototype is the OverconstrainedError
prototype object. This property has the attributes { [[Writable]]: false, [[Enumerable]]: false, [[Configurable]]: false }.
7.2.3 Properties of the OverconstrainedError Prototype Object
The OverconstrainedError prototype object is an ordinary object. It is not an Error instance and does not have an [[ErrorData]] internal slot.
The value of the [[Prototype]] internal slot of the OverconstrainedError prototype object is the intrinsic object %ErrorPrototype%.
The initial value of the constructor property of the prototype for the OverconstrainedError constructor is the intrinsic object %OverconstrainedError%.
7.2.3.2 OverconstrainedError.prototype.constraint
The initial value of the constraint property of the prototype for the OverconstrainedError constructor is the empty String.
7.2.3.3 OverconstrainedError.prototype.message
The initial value of the message property of the prototype for the OverconstrainedError constructor is the empty String.
7.2.3.4 OverconstrainedError.prototype.name
The initial value of the name property of the prototype for the OverconstrainedError constructor is
"OverconstrainedError".
7.2.4 Properties of OverconstrainedError Instances
OverconstrainedError instances are ordinary objects that inherit properties from the OverconstrainedError prototype object and have an [[ErrorData]] internal slot whose value is undefined. The only specified use of [[ErrorData]] is by Object.prototype.toString ([
ES6], section 19.1.3.6) to identify instances of Error or its various subclasses.
Note
The following will need to be updated when we finish out the error definitions.
The following interface is defined for cases when an OverconstrainedError is raised as an event:
This error event fires for each affected track (when multiple tracks share the same source) after the User Agent has evaluated the current constraints against a given source and is not able to configure the source within the limitations established by the intersection of imposed constraints.
Due to being over-constrained, the User Agent must mute each affected track.
The affected track(s) will remain muted until the application adjusts the constraints to accommodate the source's current effective capabilities.
ended
Event
The MediaStreamTrack object's source will no longer provide any data, either because the user revoked the permissions, or because the source device has been ejected, or because the remote peer permanently stopped sending data.
The following events fire on MediaDevices
objects:
Event name
Interface
Fired when...
devicechange
Event
The set of media devices, available to the User Agent, has changed. The current list devices can be retrieved with the
enumerateDevices() method.
9. Enumerating Local Media Devices
This section describes an API that the script can use to query the User Agent about connected media input and output devices (for example a web camera or a headset).
Create three internal slots: [[devicesLiveMap]], [[devicesAccessibleMap]], and [[kindsAccessibleMap]], each initialized to a different empty object.
Create one internal slot: [[storedDeviceList]], initialized to null.
For each kind of device, kind, that
getUserMedia() exposes, set [[kindsAccessibleMap]]
[kind] either to true if the result of retrieving
the permission state of the permission associated with
kind (e.g. "camera", "microphone"), is "granted", or to
false otherwise.
For each individual device that getUserMedia() exposes, using the device's deviceId, deviceId, set [[devicesLiveMap]]
[deviceId] to false, and set [[devicesAccessibleMap]][deviceId] either to
true if the result of
retrieving the permission
state of the permission associated with the device’s kind and deviceId, is “granted”, or to false otherwise.
If the transition is to “granted” from another value, then set [[devicesAccessibleMap]]
[deviceId] to true, if it isn’t already true.
If the transition is from “granted” to another value, and the device is currently stopped, then set [[devicesAccessibleMap]][deviceId] to
false.
When new media input and/or output devices are made available, or any available input and/or output device becomes unavailable, if either the permission state of the "device-info" permission is "granted", or any of the local devices are attached to an active MediaStream in the current browsing context, or the active document in the current browsing context is fully active and has focus, then the User Agent MUST run the following steps:
The User Agent MAY fire just a single event when several devices are added or removed at the same time, e.g. a camera with a microphone. These events are potentially triggered simultaneously across browsing contexts on different origins; user agents MAY add fuzzing on the timing of events to avoid cross-origin activity correlation.
The event type of this event handler is devicechange.
Methods
enumerateDevices
Collects information about the User Agent's available media input and output devices.
This method returns a promise. The promise will be fulfilled with a sequence of MediaDeviceInfo
dictionaries representing the User Agent's available media input and output devices if enumeration is successful.
Camera and microphone sources should be enumerable. Specifications that add additional types of source will provide recommendations about whether the source type should be enumerable.
When the enumerateDevices() method is called, the User Agent must run the following steps:
Let p be a new promise.
Run the following steps in parallel:
If [[storedDeviceList]] is not null, then let resultList be a copy of [[storedDeviceList]], and jump to the step labeled
Complete Enumeration.
Let resultList be an empty list.
If this method has been called previously within this browsing session, let oldList be the list of
MediaDeviceInfo objects that was produced at that call (resultList); otherwise, let oldList be an empty list.
Probe the User Agent for available media devices, and run the following sub steps for each discovered device,
device:
If device is represented by a
MediaDeviceInfo object in
oldList, append that object to
resultList, abort these steps and continue with the next device (if any).
Let deviceInfo be a new
MediaDeviceInfo object to represent device.
If a stored deviceId exists for
device, initialize deviceInfo's
deviceId to that value. Otherwise, let deviceInfo's
deviceId member be a newly generated unique identifier.
If device belongs to the same physical device as a device already represented in
oldList or resultList, initialize deviceInfo's
groupId member to the
groupId value of the existing MediaDeviceInfo object. Otherwise, let deviceInfo's
groupId member be a newly generated unique identifier.
Append deviceInfo to
resultList.
Set [[storedDeviceList]] to resultList.
Complete Enumeration: run the following sub steps to resolve p:
If any of the local devices are attached to a
live MediaStreamTrack in the current browsing context, set list-permission to "granted", otherwise set list-permission to the result of
retrieving the
permission state of the "device-info" permission.
If list-permission is not "granted", let
filteredList be a copy of
resultList, and all its elements, where the
label
member is the empty string.
If filteredList is a non-empty list, then resolve p with filteredList. Otherwise, resolve p with
resultList.
Return p.
Since this method returns persistent information across browsing sessions and origins via the number and grouping of media capture devices, it adds to the fingerprinting surface exposed by the user agent.
Once authorization has been granted to one of the capture devices, it provides additional persistent cross-origin information via the human readable labels associated with available capture devices, which furhter adds to the fingerprinting surface.
No parameters.
Return type:Promise<sequence<MediaDeviceInfo>>
9.2.1 Access control model
The algorithm described above means that the access to media device information depends on whether or not permission has been granted to the page's origin.
If no such access has been granted, the
MediaDeviceInfo dictionary will contain the deviceId, kind, and groupId.
If access has been granted for a media device, the
MediaDeviceInfo dictionary will contain the deviceId, kind, label, and groupId.
All enumerable devices have an identifier that MUST be unique to the page's origin. This identifier MUST be un-guessable by applications of other origins to prevent the identifier from being used to correlate the same user across different origins.
If any local devices have been attached to a live MediaStreamTrack in a page from this origin, or
stored permission to access local devices has been granted to this origin, then this identifier MUST be persisted, except as detailed below. Unique and stable identifiers let the application save, identify the availability of, and directly request specific sources, across multiple visits.
However, as long as no local device has been attached to a live MediaStreamTrack in a page from this origin, and no
stored permission to access local devices has been granted to this origin, then the user agent
MAY clear this identifier once the last browsing session from this origin has been closed. If the user agent chooses not to clear the identifier in this condition, then it MUST provide for the user to visibly inspect and delete the identifier, like a cookie.
Since deviceId may persist across browsing sessions and to reduce its potential as a fingerprinting mechanism, deviceId is to be treated as other persistent storage mechanisms such as cookies [
COOKIES], in that user agents MUST NOT persist device identifiers for sites that are blocked from using cookies, and user agents MUST reset per-origin device identifiers when other persistent storage are cleared.
A label describing this device (for example "External USB Webcam"). If the device has no associated label, then this attribute MUST return the empty string.
groupId of type DOMString, readonly
Returns the group identifier of the represented device. Two devices have the same group identifier if they belong to the same physical device; for example a monitor with a built-in camera and microphone.
Serializer
Instances of this interface are serialized as a map with entries for each of the serializable attributes.
Returns a MediaTrackCapabilities object describing the primary audio or video track of a device's
MediaStream (according to its
kind value), in the absence of any user-supplied constraints. These capabilities MUST be identical to those that would have been obtained by calling
getCapabilities() on the first
MediaStreamTrack of this type in a
MediaStream returned by
getUserMedia({deviceId: id}) where id is the value of the deviceId attribute of this
MediaDeviceInfo.
If no access has been granted to any local devices and this
InputDeviceInfo has been filtered with respect to unique identifying information (see above description of
enumerateDevices() result), then this method returns
null.
No parameters.
Return type:MediaTrackCapabilities
10. Obtaining local multimedia content
This section extends NavigatorUserMedia and
MediaDevices with APIs to request permission to access media input devices available to the User Agent.
When on an insecure origin [mixed-content], User Agents are encouraged to warn about usage of navigator.mediaDevices.getUserMedia,
navigator.getUserMedia, and any prefixed variants in their developer tools, error logs, etc. It is explicitly permitted for User Agents to remove these APIs entirely when on an insecure origin, as long as they remove all of them at once (e.g., they should not leave just the prefixed version available on insecure origins).
All Documents have a user media enabled flag. This specification does not detail how or when this flag is set.
10.1 NavigatorUserMedia Interface Extensions
Note
The definition of getUserMedia() in this section reflects two major changes from the method definition that has existed here for many months.
First, the official definition for the getUserMedia() method, and the one which developers are encouraged to use, is now at MediaDevices. This decision reflected consensus as long as the original API remained available here under the Navigator object for backwards compatibility reasons, since the working group acknowledges that early users of these APIs have been encouraged to define getUserMedia as "var getUserMedia = navigator.getUserMedia || navigator.webkitGetUserMedia || navigator.mozGetUserMedia;" in order for their code to be functional both before and after official implementations of getUserMedia() in popular browsers. To ensure functional equivalence, the getUserMedia() method here is defined in terms of the method under MediaDevices.
Second, the decision to change all other callback-based methods in the specification to be based on Promises instead required that the navigator.getUserMedia() definition reflect this in its use of navigator.mediaDevices.getUserMedia(). Because navigator.getUserMedia() is now the only callback-based method remaining in the specification, there is ongoing discussion as to a) whether it still belongs in the specification, and b) if it does, whether its syntax should remain callback-based or change in some way to use Promises. Input on these questions is encouraged, particularly from developers actively using today's implementations of this functionality.
Note that the other methods that changed from a callback-based syntax to a Promises-based syntax were not considered to have been implemented widely enough in any form to have to consider legacy usage.
The successCallback will be invoked with a suitable
MediaStream object as its argument if the user accepts valid tracks as described in getUserMedia() on
MediaDevices.
The errorCallback will be invoked if there is a failure in finding valid tracks or if the user denies permission, as described in getUserMedia() on
MediaDevices.
When the getUserMedia() method is called, the User Agent MUST run the following steps:
Let constraints be the method's first argument.
Let successCallback be the callback indicated by the method's second argument.
Let errorCallback be the callback indicated by the method's third argument.
Run the steps specified by the getUserMedia() algorithm with constraints as the argument, and let
p be the resulting promise.
Upon fulfillment of p with value
stream, run the following step:
Invoke successCallback with
stream as the argument.
Upon rejection of p with reason r, run the following step:
Invoke errorCallback with r as the argument.
Parameter
Type
Nullable
Optional
Description
constraints
MediaStreamConstraints
✘
✘
successCallback
NavigatorUserMediaSuccessCallback
✘
✘
errorCallback
NavigatorUserMediaErrorCallback
✘
✘
Return type:void
10.2 MediaDevices Interface Extensions
Note
The definition of getUserMedia() in this section reflects two major changes from the method definition that has existed under NavigatorUserMedia for many months.
First, the official definition for the getUserMedia() method, and the one which developers are encouraged to use, is now the one defined here under MediaDevices. This decision reflected consensus as long as the original API remained available at
NavigatorUserMedia.getUserMedia under the Navigator object for backwards compatibility reasons, since the working group acknowledges that early users of these APIs have been encouraged to define getUserMedia as "var getUserMedia = navigator.getUserMedia || navigator.webkitGetUserMedia || navigator.mozGetUserMedia;" in order for their code to be functional both before and after official implementations of getUserMedia() in popular browsers. To ensure functional equivalence, the getUserMedia() method under NavigatorUserMedia is defined in terms of the method here.
Second, the method defined here is Promises-based, while the one defined under NavigatorUserMedia is currently still callback-based. Developers expecting to find getUserMedia() defined under NavigatorUserMedia are strongly encouraged to read the detailed Note given there.
The getSupportedConstraints method is provided to allow the application to determine which constraints the User Agent recognizes.
Returns a dictionary whose members are the constrainable properties known to the User Agent. A supported constrainable property MUST be represented and any constrainable properties not supported by the User Agent MUST NOT be present in the returned dictionary. The values returned represent what the browser implements and will not change during a browsing session.
No parameters.
Return type:MediaTrackSupportedConstraints
getUserMedia
Prompts the user for permission to use their Web cam or other video or audio input.
This method returns a promise. The promise will be fulfilled with a suitable MediaStream object if the user accepts valid tracks as described below.
The promise will be rejected if there is a failure in finding valid tracks or if the user denies permission, as described below.
When the getUserMedia() method is called, the User Agent MUST run the following steps:
Let constraints be the method's first argument.
Let requestedMediaTypes be the set of media types in constraints with either a dictionary value or a value of "true".
If requestedMediaTypes is the empty set, return a promise rejected with a TypeError. The word "optional" occurs in the WebIDL due to WebIDL rules, but the argument MUST be supplied in order for the call to succeed.
If the current [HTML51] Document's user media
enabled flag is unset, return a promise rejected with a
DOMException object whose
name attribute has the value
SecurityError.
Let p be a new promise.
Run the following steps in parallel:
Let finalSet be an (initially) empty set.
For each media type T in
requestedMediaTypes,
For each possible source for that media type, construct an unconstrained MediaStreamTrack with that source as its source.
Call this set of tracks the
candidateSet.
If candidateSet is the empty set, reject p with a new
DOMException object whose
name attribute has the value
NotFoundError and abort these steps.
If the value of the T entry of
constraints is "true", set CS to the empty constraint set (no constraint). Otherwise, continue with CS set to the value of the T entry of constraints.
Run the SelectSettings algorithm on each track in CandidateSet with CS as the constraint set. If the algorithm returns undefined, remove the track from candidateSet. This eliminates devices unable to satisfy the constraints, by verifying that at least one settings dictionary exists that satisfies the constraints.
If candidateSet is the empty set, let
constraint be any required constraint whose fitness distance was infinity for all settings dictionaries examined while executing the
SelectSettings algorithm, let
message be either undefined or an informative human-readable message, and reject
p with a new
OverconstrainedError created by calling
OverconstrainedError(constraint,
message), then abort these steps.
This error gives information about what the underlying device is not capable of producing, before the user has given any authorization to any device, and can thus be used as a fingerprinting surface.
Retrieve the permission state for all candidate devices in candidateSet that are not attached to a live MediaStreamTrack in the current browsing context. Remove from
candidateSet any device for which the permission state is "denied".
If candidateSet is now empty, indicating that all devices of this type are in state "denied", jump to the step labeled
PermissionFailure below.
Add all tracks from candidateSet to
finalSet.
Optionally, e.g., based on a previously-established user preference, for security reasons, or due to platform limitations, jump to the step labeled Permission
Failure below.
For the origin identified by
originIdentifier, request permission for use of the devices, while considering all devices attached to a live MediaStreamTrack in the current
browsing context to have permission status "granted", resulting in a set of provided media.
The provided media MUST include precisely one track of each media type in requestedMediaTypes from the
finalSet. The decision of which devices to choose from the finalSet is completely up to the User Agent and may be determined by asking the user. Once selected, the source of a
MediaStreamTrackMUST NOT change.
The User Agent MAY use the value of the computed "fitness distance" from the SelectSettings algorithm, or any other internally-available information about the devices, as an input to the selection algorithm.
User Agents are encouraged to default to using the user's primary or system default camera and/or microphone (when possible) to generate the media stream. User Agents
MAY allow users to use any media source, including pre-recorded media files.
If the result of the request is "granted", then for each device that is sourcing the provided media, using the device's deviceId, deviceId, set [[devicesLiveMap]]
[deviceId] to
true, if it isn’t already true, and set the [[devicesAccessibleMap]][deviceId] to true, if it isn’t already
true.
If the result is "denied", jump to the step labeled
Permission Failure below. If the user never responds, this algorithm stalls on this step.
If the user grants permission but a hardware error such as an OS/program/webpage lock prevents access, reject p with a new
DOMException object whose
name attribute has the value
NotReadableError and abort these steps.
If the result is "granted" but device access fails for any reason other than those listed above, reject
p with a new DOMException
object whose name attribute has the value AbortError and abort these steps.
Let stream be the
MediaStream object for which the user granted permission.
Run the ApplyConstraints() algorithm on all tracks in stream with the appropriate constraints.
Resolve p with stream and abort these steps.
Permission Failure: Reject p with a new DOMException object whose
name attribute has the value
NotAllowedError.
Return p.
Parameter
Type
Nullable
Optional
Description
constraints
MediaStreamConstraints
✘
✘
Return type:Promise<MediaStream>
Note
In the algorithm above, constraints are checked twice - once at device selection, and once after access approval. Time may have passed between those checks, so it is conceivable that the selected device is no longer suitable. In this case, a NotReadableError will result.
video of type (boolean or MediaTrackConstraints),
defaulting to false
If true, it requests that the returned
MediaStream contain a video track. If a Constraints structure is provided, it further specifies the nature and settings of the video Track. If false, the
MediaStreamMUST NOT contain a video Track.
audio of type (boolean or MediaTrackConstraints),
defaulting to false
If true, it requests that the returned
MediaStream contain an audio track. If a
Constraints structure is provided, it further specifies the nature and settings of the audio Track. If
false, the MediaStreamMUST NOT contain an audio Track.
The User Agent is encouraged to reserve resources when it has determined that a given call to
getUserMedia() will be successful. It is preferable to reserve the resource prior to resolving the returned promise. Subsequent calls to getUserMedia() (in this page or any other) should treat the resource that was previously allocated, as well as resources held by other applications, as busy. Resources marked as busy should not be provided as sources to the current web page, unless specified by the user. Optionally, the User Agent may choose to provide a stream sourced from a busy source but only to a page whose origin matches the owner of the original stream that is keeping the source busy.
This document recommends that in the permission grant dialog or device selection interface (if one is present), the user be allowed to select any available hardware as a source for the stream requested by the page (provided the resource is able to fulfill any specified mandatory constraints). Although not specifically recommended as best practice, note that some User Agents may support the ability to substitute a video or audio source with local files and other media. A file picker may be used to provide this functionality to the user.
This document also recommends that the user be shown all resources that are currently busy as a result of prior calls to getUserMedia() (in this page or any other page that is still alive) and be allowed to terminate that stream and utilize the resource for the current page instead. If possible in the current operating environment, it is also suggested that resources currently held by other applications be presented and treated in the same manner. If the user chooses this option, the track corresponding to the resource that was provided to the page whose stream was affected must be removed.
Best Practice 2: Stored
Permissions
When permission is requested for a device, the User Agent may choose to create a
permission storage entry for later use by the same origin, so that the user does not need to grant permission again at a later time. [
RTCWEB-SECURITY-ARCH] Section 5.2 requires that such storing MUST only be done when the page is secure (served over HTTPS and having no mixed content). It is a User Agent choice whether it offers functionality to store permission to each device separately, all devices of a given class, or all devices; the choice needs to be apparent to the user, and permission must have been granted for the entire set whose permission is being stored, e.g., to store permission to use all cameras the user must have given permission to use all cameras and not just one.
As described, this specification does not dictate whether or not granting permission results in a stored permission. When permission is not stored, permission will last only until such time as all MediaStreamTracks sourced from that device have been stopped.
Best Practice 3: Handling multiple
devices
A MediaStream may contain more than one video and audio track. This makes it possible to include video from two or more webcams in a single stream object, for example. However, the current API does not allow a page to express a need for multiple video streams from independent sources.
It is recommended for multiple calls to
getUserMedia() from the same page to be allowed as a way for pages to request multiple discrete video and/or audio streams.
Note also that if multiple getUserMedia() calls are done by a page, the order in which they request resources, and the order in which they complete, is not constrained by this specification.
A single call to getUserMedia() will always return a stream with either zero or one audio tracks, and either zero or one video tracks. If a script calls getUserMedia() multiple times before reaching a stable state, this document advises the UI designer that the permission dialogs should be merged, so that the user can give permission for the use of multiple cameras and/or media sources in one dialog interaction. The constraints on each getUserMedia call can be used to decide which stream gets which media sources.
11. Constrainable Pattern
The Constrainable pattern allows applications to inspect and adjust the properties of objects implementing it (the constrainable
object). It is broken out as a separate set of definitions so that it can be referred to by other specifications. The core concept is the Capability, which consists of a constrainable property of an object and the set of its possible values, which may be specified either as a range or as an enumeration. For example, a camera might be capable of framerates (a property) between 20 and 50 frames per second (a range) and may be able to be positioned (a property) facing towards the user, away from the user, or to the left or right of the user (an enumerated set). The application can examine a constrainable property's supported Capabilities via the
getCapabilities() accessor.
The application can select the (range of) values it wants for an object's Capabilities by means of basic and/or advanced ConstraintSets and the applyConstraints() method. A ConstraintSet consists of the names of one or more properties of the object plus the desired value (or a range of desired values) for each property. Each of those property/value pairs can be considered to be an individual constraint. For example, the application may set a ConstraintSet containing two constraints, the first stating that the framerate of a camera be between 30 and 40 frames per second (a range) and the second that the camera should be facing the user (a specific value). How the individual constraints interact depends on whether and how they are given in the basic Constraint structure, which is a ConstraintSet with an additional 'advanced' property, or whether they are in a ConstraintSet in the advanced list. The behavior is as follows: all 'min', 'max', and 'exact' constraints in the basic Constraint structure are together treated as the 'required' set, and if it is not possible to satisfy simultaneously all of those individual constraints for the indicated property names, the User Agent MUST reject the returned promise. Otherwise, it must apply the required constraints. Next, it will consider any ConstraintSets given in the 'advanced' list, in the order in which they are specified, and will try to satisfy/apply each complete ConstraintSet (i.e., all constraints in the ConstraintSet together), but will skip a ConstraintSet if and only if it cannot satisfy/apply it in its entirety. Next, the User Agent MUST attempt to apply, individually, any 'ideal' constraints or a constraint given as a bare value for the property. Of these properties, it MUST satisfy the largest number that it can, in any order. Finally, the User Agent MUST resolve the returned promise.
Note
Any constraint provided via this API will only be considered if the given constrainable property is supported by the browser. JavaScript application code is expected to first check, via
getSupportedConstraints(), that all the named properties that are used are supported by the browser. The reason for this is that WebIDL drops any unsupported names from the dictionary holding the constraints, so the browser does not see them and the unsupported names end up being silently ignored. This will cause confusing programming errors as the JavaScript code will be setting constraints but the browser will be ignoring them. Browsers that support (recognize) the name of a required constraint but cannot satisfy it will generate an error, while browsers that do not support the constrainable property will not generate an error.
The following examples may help to understand how constraints work. The first shows a basic Constraint structure. Three constraints are given, each of which the User Agent will attempt to satisfy individually. Depending upon the resolutions available for this camera, it is possible that not all three constraints can be satisfied at the same time. If so, the User Agent will satisfy two if it can, or only one if not even two constraints can be satisfied together. Note that if not all three can be satisfied simultaneously, it is possible that there is more than one combination of two constraints that could be satisfied. If so, the User Agent will choose.
Example 1
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["aspectRatio"]) {
// Treat like an error.
}
var constraints =
{
width: 1280,
height: 720,
aspectRatio: 1.5
};
This next example adds a small bit of complexity. The ideal values are still given for width and height, but this time with minimum requirements on each as well as a minimum frameRate that must be satisfied. If it cannot satisfy the frameRate, width or height minimum it will reject the promise. Otherwise, it will try to satisfy the width, height, and aspectRatio target values as well and then resolve the promise. Note that the frameRate minimum might be within the capabilities of the camera and satisfiable in ideal lighting conditions, but not in low light, and could therefore result in firing of the onoverconstrained event handler under poor lighting conditions.
Example 2
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["aspectRatio"] || !supports["frameRate"]) {
// Treat like an error.
}
var constraints =
{
frameRate: {min: 20},
width: {min: 640, ideal: 1280},
height: {min: 480, ideal: 720},
aspectRatio: 1.5
};
This example illustrates the full control possible with the Constraints structure by adding the 'advanced' property. In this case, the User Agent behaves the same way with respect to the required constraints, but before attempting to satisfy the ideal values it will process the 'advanced' list. In this example the 'advanced' list contains two ConstraintSets. The first specifies width and height constraints, and the second specifies an aspectRatio constraint. Note that in the advanced list, these bare values are treated as 'exact' values. This example represents the following: "I need my video to be at least 640 pixels wide and at least 480 pixels high. My preference is for precisely 1920x1280, but if you can't give me that, give me an aspectRatio of 4x3 if at all possible. If even that is not possible, give me a resolution as close to 1280x720 as possible."
Example 3
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["width"] || !supports["height"]) {
// Treat like an error.
}
var constraints =
{
width: {min: 640, ideal: 1280},
height: {min: 480, ideal: 720},
advanced: [{width: 1920, height: 1280},
{aspectRatio: 1.3333333333}]
};
The ordering of advanced ConstraintSets is significant. In the preceding example it is impossible to satisfy both the 1920x1280 ConstraintSet and the 4x3 aspect ratio ConstraintSet at the same time. Since the 1920x1280 occurs first in the list, the User Agent will attempt to satisfy it first. Application authors can therefore implement a backoff strategy by specifying multiple optional ConstraintSets for the same property. For example, an application might specify three optional ConstraintSets, the first asking for a frame rate greater than 500, the second asking for a frame rate greater than 400, and the third asking for one greater than 300. If the User Agent is capable of setting a frame rate greater than 500, it will (and the subsequent two ConstraintSets will be trivially satisfied). However, if the User Agent cannot set the frame rate above 500, it will skip that ConstraintSet and attempt to set the frame rate above 400. If that fails, it will then try to set it above 300. If the User Agent cannot satisfy any of the three ConstraintSets, it will set the frame rate to any value it can get. If the developers wanted to insist on 300 as a lower bound, they could provide that as a 'min' value in the basic ConstraintSet. In that case, the User Agent would fail altogether if it couldn't get a value over 300, but would choose a value over 500 if possible, then try for a value over 400.
Note that, unlike basic constraints, the constraints within a ConstraintSet in the advanced list must be satisfied together or skipped together. Thus, {width: 1920, height: 1280} is a request for that specific resolution, not a request for that width or that height. One can think of the basic constraints as requesting an or (non-exclusive) of the individual constraints, while each advanced ConstraintSet is requesting an and of the individual constraints in the ConstraintSet. An application may inspect the full set of Constraints currently in effect via the
getConstraints() accessor.
The specific value that the User Agent chooses for a constrainable property is referred to as a Setting. For example, if the application applies a ConstraintSet specifying that the frameRate must be at least 30 frames per second, and no greater than 40, the Setting can be any intermediate value, e.g., 32, 35, or 37 frames per second. The application can query the current settings of the object's constrainable properties via the getSettings() accessor.
11.1 Interface Definition
Although this specification formally defines ConstrainablePattern as a WebIDL interface, it is actually a template or pattern for other interfaces and cannot be inherited directly since the return values of the methods need to be extended, something WebIDL cannot do. Thus, each interface that wishes to make use of the functionality defined here will have to provide its own copy of the WebIDL for the functions and interfaces given here. However it can refer to the semantics defined here, which will not change. See MediaStreamTrack Interface
Definition for an example of this.
When the User Agent is no longer able to satisfy the
requiredConstraints from the currently valid Constraints, the User Agent MUST queue a task that fires an
OverconstrainedErrorEvent, initialized as described in the following paragraph, at the constrainable object. The event firing task MAY also be used to update the constrainable object as a result of the overconstrained situation.
The OverconstrainedErrorEvent references an
OverconstrainedError whose constraint attribute is set to one of the requiredConstraints that can no longer be satisfied. The message attribute of the
OverconstrainedErrorSHOULD contain a string that is useful for debugging. The conditions under which this error might occur are platform and application-specific. For example, the user might physically manipulate a camera in a way that makes it impossible to provide a resolution or frameRate that satisfies the constraints. The User Agent
MAY take other actions as a result of the overconstrained situation.
The getCapabilities() method returns the dictionary of the names of the constrainable properties that the object supports.
Note
It is possible that the underlying hardware may not exactly map to the range defined for the constrainable property. Where this is possible, the entry SHOULD define how to translate and scale the hardware's setting onto the values defined for the property. For example, suppose that a hypothetical fluxCapacitance property ranges from -10 (min) to 10 (max), but there are common hardware devices that support only values of "off" "medium" and "full". The constrainable property definition might specify that for such hardware, the User Agent should map the range value of -10 to "off", 10 to "full", and 0 to "medium". It might also indicate that given a ConstraintSet imposing a strict value of 3, the User Agent should attempt to set the value of "medium" on the hardware, and that
getSettings() should return a fluxCapacitance of 0, since that is the value defined as corresponding to "medium".
No parameters.
Return type:Capabilities
getConstraints
The getConstraints method returns the Constraints that were the argument to the most recent successful call of
applyConstraints(), maintaining the order in which they were specified. Note that some of the optional ConstraintSets returned may not be currently satisfied. To check which ConstraintSets are currently in effect, the application should use getSettings.
No parameters.
Return type:Constraints
getSettings
The getSettings() method returns the current settings of all the constrainable properties of the object, whether they are platform defaults or have been set by
applyConstraints(). Note that the actual setting of a property MUST be a single value.
No parameters.
Return type:Settings
applyConstraints
The applyConstraints() algorithm for applying constraints is stated below. Here are some preliminary definitions that are used in the statement of the algorithm:
We use the term settings dictionary for the set of values that might be applied as settings to the object.
For string valued constraints, we define "==" below to be true if one of the values in the sequence is exactly the same as the value being compared against.
We define the fitness distance between a
settings dictionary and a constraint set CS as the sum, for each constraint provided for a constraint name in
CS, of the following values:
If the constraint is not supported by the browser, the fitness distance is 0.
If the constraint is required ('min', 'max', or 'exact'), and the settings dictionary's value for the constraint does not satisfy the constraint, the fitness distance is positive infinity.
If no ideal value is specified, the fitness distance is 0.
For all positive numeric non-required constraints (such as height, width, frameRate, aspectRatio, sampleRate and sampleSize), the fitness distance is the result of the formula
For all string and enum non-required constraints (deviceId, groupId, facingMode, echoCancellation), the fitness distance is the result of the formula
(actual == ideal) ? 0 : 1
More definitions:
We refer to each element of a ConstraintSet (other than the special term 'advanced') as a 'constraint' since it is intended to constrain the acceptable settings for the given property from the full list or range given in the corresponding Capability of the ConstrainablePattern object to a value that is within the range or list of values it specifies.
We refer to the "effective Capability" C of an object O as the possibly proper subset of the possible values of C (as returned by getCapabilities) taking into consideration environmental limitations and/or restrictions placed by other constraints. For example given a ConstraintSet that constrains the aspectRatio, height, and width properties, the values assigned to any two of the properties limit the effective Capability of the third. The set of effective Capabilities may be platform dependent. For example, on a resource-limited device it may not be possible to set properties P1 and P2 both to 'high', while on another less limited device, this may be possible.
A settings dictionary, which is a set of values for the constrainable properties of an object O, satisfies ConstraintSet CS if the fitness distance between the set and CS is less than infinity.
A set of ConstraintSets CS1...CSn (n >= 1) can be satisfied by an object O if it is possible to find a settings dictionary of O that satisfies CS1...CSn simultaneously.
To apply a set of ConstraintSets CS1...CSn to object O is to choose such a sequence of values that satisfy CS1...CSn and assign them as the settings for the properties of O.
We define the SelectSettings algorithm as follows:
Each constraint specifies one or more values (or a range of values) for its property. A property MAY appear more than once in the list of 'advanced' ConstraintSets. If an empty object or list has been given as the value for a constraint, it MUST be interpreted as if the constraint were not specified (in other words, an empty constraint == no constraint).
Note that unknown properties are discarded by WebIDL, which means that unknown/unsupported required constraints will silently disappear. To avoid this being a surprise, application authors are expected to first use the
getSupportedConstraints() method as shown in the Examples below.
Let object be the
ConstrainablePattern object on which this algorithm is applied. Let copy be an unconstrained copy of object (i.e., copy should behave as if it were object with all ConstraintSets removed.)
If candidates is empty, return
undefined as the result of the
SelectSettings() algorithm.
Iterate over the 'advanced' ConstraintSets in
newConstraints in the order in which they were specified. For each ConstraintSet:
compute the fitness distance between it and each settings dictionary in candidates, treating bare values of properties as exact.
If the fitness distance is finite for one or more settings dictionaries in candidates, keep those settings dictionaries in candidates, discarding others.
If the fitness distance is infinite for all settings dictionaries in candidates, ignore this ConstraintSet.
Select one settings dictionary from candidates, and return it as the result of the
SelectSettings() algorithm. The UA SHOULD use the one with the smallest fitness distance, as calculated in step 3.
When applyConstraints is called, the User Agent
MUST run the following steps:
Let p be a new promise.
Let newContraints be the argument to this function.
Run the following steps in parallel:
Let successfulSettings be the result of running the SelectSettings algorithm with
newConstraints as the constraint set.
If successfulSettings is
undefined, let failedConstraint be any required constraint whose fitness distance was infinity for all settings dictionaries examined while executing the SelectSettings algorithm, let message be either
undefined or an informative human-readable message, reject p with a new
OverconstrainedError created by calling
OverconstrainedError(failedConstraint,
message), and abort these steps. The existing constraints remain in effect in this case.
In a single operation, remove the existing constraints from object, apply newConstraints, and apply successfulSettings as the current settings.
Finally, resolve p. From this point on until applyConstraints() is called successfully again, getConstraints() must return the newConstraints that were passed as an argument to this call.
Return p.
Note
Any implementation that has the same result as the algorithm above is an allowed implementation. For instance, the implementation may choose to keep track of the maximum and minimum values for a setting that are OK under the constraints considered, rather than keeping track of all possible values for the setting.
Note
When picking a settings dictionary, the UA can use any information available to it. Examples of such information may be whether the selection is done as part of device selection in getUserMedia, whether the energy usage of the camera varies between the settings dictionaries, or whether using a settings dictionary will cause the device driver to apply resampling.
The User Agent MAY choose new settings for the constrainable properties of the object at any time. When it does so it MUST attempt to satisfy the current Constraints, in the manner described in the algorithm above.
Parameter
Type
Nullable
Optional
Description
constraints
Constraints
✘
✔
A new constraint structure to apply to this object.
Here is another example, specifically for a video track where I must have a particular camera and have separate preferences for the width and height:
Example 5
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["deviceId"]) {
// Treat like an error.
}
var constraints = {
deviceId: {exact: "20983-20o198-109283-098-09812"},
advanced: [{
width: {
min: 800,
max: 1200
}
}, {
height: {
min: 600
}
}]
};
And here's one for an audio track:
Example 6
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["deviceId"] || !supports["volume"]) {
// Treat like an error.
}
var constraints = {
advanced: [{
deviceId: "64815-wi3c89-1839dk-x82-392aa"
}, {
volume: 0.5
}]
};
Here's an example of use of ideal:
Example 7
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["aspectRatio"] || !supports["facingMode"]) {
// Treat like an error.
}
var gotten = navigator.mediaDevices.getUserMedia({
video: {
width: {min: 320, ideal: 1280, max: 1920},
height: {min: 240, ideal: 720, max: 1080},
frameRate: 30, // Shorthand for ideal.// facingMode: "environment" would be optional.
facingMode: {exact: "environment"}
}});
Here's an example of "I want 720p, but I can accept up to 1080p and down to VGA.":
Example 8
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["width"] || !supports["height"]) {
// Treat like an error.
}
var gotten = navigator.mediaDevices.getUserMedia({video: {
width: {min: 640, ideal: 1280, max: 1920},
height: {min: 480, ideal: 720, max: 1080},
}});
Here's an example of "I want a front-facing camera and it must be VGA.":
Example 9
var supports = navigator.mediaDevices.getSupportedConstraints();
if(!supports["width"] || !supports["height"] || !supports["facingMode"]) {
// Treat like an error.
}
var gotten = navigator.mediaDevices.getUserMedia({video: {
facingMode: {exact: "user"},
width: {exact: 640},
height: {exact: 480}
}});
11.2 Types for Constrainable Properties
The syntax for the specification of the set of legal values depends on the type of the values. In addition to the standard atomic types (boolean, long, double, DOMString), legal values include lists of any of the atomic types, plus min-max ranges, as defined below.
List values MUST be interpreted as disjunctions. For example, if a property 'facingMode' for a camera is defined as having legal values ["left", "right", "user", "environment"], this means that 'facingMode' can have the values "left", "right", "environment", and "user". Similarly
Constraints restricting 'facingMode' to ["user", "left", "right"] would mean that the User Agent should select a camera (or point the camera, if that is possible) so that "facingMode" is either "user", "left", or "right". This Constraint would thus request that the camera not be facing away from the user, but would allow the User Agent to allow the user to choose other directions.
Capabilities is a dictionary containing one or more key-value pairs, where each key MUST be a constrainable property, and each value MUST be a subset of the set of values allowed for that property. The exact syntax of the value expression depends on the type of the property. The Capabilities dictionary specifies which constrainable properties that can be applied, as constraints, to the constrainable
object. Note that the Capabilities of a constrainable objectMAY be a subset of the properties defined in the Web platform, with a subset of the set values for those properties. Note that Capabilities are returned from the User Agent to the application, and cannot be specified by the application. However, the application can control the Settings that the User Agent chooses for constrainable properties by means of Constraints.
An example of a Capabilities dictionary is shown below. In this case, the constrainable object is a video source with a very limited set of Capabilities.
The next example below points out that capabilities for range values provide ranges for individual constrainable properties, not combinations. This is particularly relevant for video width and height, since the ranges for width and height are reported separately. In the example, if the constrainable object can only provide 640x480 and 800x600 resolutions the relevant capabilities returned would be:
Note in the example above that the aspectRatio would make clear that arbitrary combination of widths and heights are not possible, although it would still suggest that more than two resolutions were available.
A specification using the Constrainable Pattern should not subclass the below dictionary, but instead provide its own definition. See
MediaTrackCapabilities for an example.
Settings is a dictionary containing one or more key-value pairs. It MUST contain each key returned in
getCapabilities(). There MUST be a single value for each key and the value MUST be a member of the set defined for that property by
getCapabilities(). The Settings dictionary contains the actual values that the User Agent has chosen for the object's constrainable properties. The exact syntax of the value depends on the type of the property.
A conforming User Agent MUST support all the constrainable properties defined in this specification.
An example of a Settings dictionary is shown below. This example is not very realistic in that a browser would actually be required to support more constrainable properties than just these.
Example 12
{
frameRate: 30.0,
facingMode: "user"
}
A specification using the Constrainable Pattern should not subclass the below dictionary, but instead provide its own definition. See
MediaTrackSettings for an example.
Due to the limitations of WebIDL, interfaces implementing the Constrainable Pattern cannot simply subclass Constraints and ConstraintSet as they are defined here. Instead they must provide their own definitions that follow this pattern. See MediaTrackConstraints for an example of this.
dictionary ConstraintSet {
};
Each member of a ConstraintSet corresponds to a constrainable property and specifies a subset of the property's legal Capability values. Applying a ConstraintSet instructs the User Agent to restrict the settings of the corresponding constrainable properties to the specified values or ranges of values. A given property MAY occur both in the basic Constraints set and in the advanced ConstraintSets list, and MAY occur at most once in each ConstraintSet in the advanced list.
This is the list of ConstraintSets that the User Agent MUST attempt to satisfy, in order, skipping only those that cannot be satisfied. The order of these ConstraintSets is significant. In particular, when they are passed as an argument to
applyConstraints, the User Agent MUST try to satisfy them in the order that is specified. Thus if optional ConstraintSets C1 and C2 can be satisfied individually, but not together, then whichever of C1 and C2 is first in this list will be satisfied, and the other will not. The User Agent MUST attempt to satisfy all optional ConstraintSets in the list, even if some cannot be satisfied. Thus, in the preceding example, if optional constraint C3 is specified after C1 and C2, the User Agent will attempt to satisfy C3 even though C2 cannot be satisfied. Note that a given property name may occur only once in each ConstraintSet but may occur in more than one ConstraintSet.
12. Examples
This sample code exposes a button. When clicked, the button is disabled and the user is prompted to offer a stream. The user can cause the button to be re-enabled by providing a stream (e.g., giving the page access to the local camera) and then disabling the stream (e.g., revoking that access).
This example allows people to take photos of themselves from the local video camera. Note that the Image Capture specification [image-capture] provides a simpler way to accomplish this.
Define any<kind>Accessible (e.g.
anyAudioAccessible, anyVideoAccessible) as the logical OR of the [[kindsAccessibleMap]][kind] value and all the [[devicesAccessibleMap]][deviceId] values for devices of that kind.
Define any<kind>Live (e.g.
anyAudioLive, anyVideoLive) to be the logical OR of the [[kindsLiveMap]][kind] value and all the [[devicesLiveMap]]
[deviceId] values for devices of that kind.
Define anyAccessible to be the logical OR of all
any<kind>Accessible values.
Define anyLive to be the logical OR of all
any<kind>Live values.
Then the following are requirements on the User Agent:
The User Agent MUST indicate to the user when the value of anyAccessible changes.
The User Agent MUST indicate to the user when the value of anyLive changes.
If the User Agent provides indication to the user per kind, then for each any<kind>Accessible value and
any<kind>Live value, it MUST at minimum indicate when the value changes.
If the User Agent provides indication to the user per
device, then for each [[devicesAccessibleMap]]
[deviceId] value and [[devicesLiveMap]]
[deviceId] value, it MUST at minimum indicate when the value changes.
Any false-to-true transition indicated MUST remain observable for a sufficient time that a reasonably-observant user could become aware of it.
Any of the above transition indications MAY be combined as long as the combined indication cannot transition to false if any of its component indications are still true.
and the following are encouraged behaviors for the User Agent:
The User Agent is encouraged to provide ongoing indication of the current state of anyAccessible.
The User Agent is encouraged to provide ongoing indication of the current state of anyLive and to make any generic hardware device indicator light match.
If the User Agent provides indication to the user per kind, then for each any<kind>Accessible value and any
<kind>Live value, it is encouraged to provide ongoing indication of the current state of the value. It is also encouraged to make any device-type-specific hardware indicator light match the corresponding any<kind>Live value.
If the User Agent provides indication to the user per
device, then for each [[devicesAccessibleMap]]
[deviceId] value and [[devicesLiveMap]]
[deviceId] value, it is encouraged to provide ongoing indication of the current state of the value. It is also encouraged to make any device-specific hardware indicator light match the corresponding [[devicesLiveMap]][deviceId] value.
Any of the above ongoing indications MAY be used instead of the corresponding required transition indication provided the false-to-true transition requirement is met.
14. Privacy and Security Considerations
This section is non-normative; it specifies no new behavior, but instead summarizes information already present in other parts of the specification.
This document extends the Web platform with the ability to manage input devices for media - in this iteration, microphones, and cameras. It also allows the manipulation of audio output devices (speakers and headphones). Capturing audio and video exposes personally-identifiable information to applications, and this specification requires obtaining explicit user consent before sharing it.
Without authorization (to the "drive-by web"), it offers the ability to tell how many devices there are of each class, and how they are grouped together (e.g. a microphone and camera belonging to a single Web cam). The identifiers for the devices are designed to not be useful for a fingerprint that can track the user between origins, but the number and grouping of devices adds to the fingerprint surface. It recommends to treat the per-origin persistent identifier deviceId as other persistent storage (e.g. cookies) are treated.
When authorization is given, this document describes how to get access to, and use, media data from the devices mentioned. This data may be sensitive; advice is given that indicators should be supplied to indicate that devices are in use, but both the nature of authorization and the indicators of in-use devices are platform decisions.
Authorization may be given on a case-by-case basis, or be persistent. In the case of a case-by-case authorization, it is important that the user be able to say "no" in a way that prevents the UI from blocking user interaction until permission is given - either by offering a way to say a "persistent NO" or by not using a modal permissions dialog.
When authorization to any media device is given, application developers gain access to the labels of all available media capture devices. In most cases, the labels are persistent across browsing sessions and across origins that have also been granted authorization, and thus potentially provide a way to track a given device across time and origins.
For origins to which permission has been granted, the
devicechange event will be emitted across browsing contexts and origins each time a new media device is added or removed; user agents can mitigate the risk of correlation of browsing activity across origins by fuzzing the timing of these events.
Once a developer gains access to a media stream from a capture device, the developer also gains access to detailed information about the device, including its range of operating capabilities (e.g. available resolutions for a camera). These operating capabilities are for the most part persistent across browsing sessions and origins, and thus provide a way to track a given device across time and origins.
Once access to a video stream from a capture device is obtained, that stream can most likely be used to fingerprint uniquely the said device (e.g. via dead pixel detection). Similarly, once access to an audio stream is obtained, that stream can most likely be used to fingerprint user location down to the level of a room or even simultaneous occupation of a room by disparate users (e.g. via analysis of ambient audio or of unique audio purposely played out of the device speaker). User-level mitigation for both audio and video consists of covering up the camera and/or microphone or revoking permission via browser chrome controls.
It is possible to use constraints so that the failure of a getUserMedia call will return information about devices on the system without prompting the user, which increases the surface available for fingerprinting. The User Agent should consider limiting the rate at which failed getUserMedia calls are allowed in order to limit this additional surface.
In the case of persistent authorization via a stored permission, it is important that it is easy to find the list of granted permissions and revoke permissions that the user wishes to revoke.
Once permission has been granted, the User Agent should make two things readily apparent to the user:
That the page has access to the devices for which permission is given
Whether or not any of the devices are presently recording ("on air") indicator
Note
Developers of sites with stored permissions should be careful that these permissions not be abused. These permissions can be revoked using the [permissions] API.
In particular, they should not make it possible to automatically send audio or video streams from authorized media devices to an end point that a third party can select.
Indeed, if a site offered URLs such as
https://webrtc.example.org/?call=user that would automatically set up calls and transmit audio/video to
user, it would be open for instance to the following abuse:
Users who have granted stored permissions to
https://webrtc.example.org/ could be tricked to send their audio/video streams to an attacker EvilSpy by following a link or being redirected to
https://webrtc.example.org/?user=EvilSpy.
Although [RTCWEB-SECURITY-ARCH] Section 5.2 indicates that implementations may refuse all access permissions for HTTP origins, it recommends that implementations allow one-time camera/microphone access. While allowing one-time access for HTTP origins is convenient, this makes it possible for an attacker to obtain access to the camera/microphone of an unsuspecting user.
15. Extensibility
This section is non-normative.
Although new versions of this specification may be produced in the future, it is also expected that other standards will need to define new capabilities that build upon those in this specification. The purpose of this section is to provide guidance to creators of such extensions.
Any WebIDL-defined interfaces, methods, or attributes in the specification may be extended. Two likely extension points are defining a new media type and defining a new constrainable property.
15.1 Defining a new media type (beyond the existing Audio and Video types)
At a minimum, defining a new media type would require
adding a new getXXXXTracks() method for the type to the
MediaStream interface,
if the new type requires user authorization, defining new permissions for it, and defining how they, along with access starting and ending, as well as muting/disabling, affect any new and/or existing "on-air" and "device accessible" indicator states (see
9.2MediaDevices).
Creators of extension specifications are strongly encouraged to notify the Media Capture Task Force of their extension by emailing the list at public-media-capture@w3.org.
Future versions of this specification and others created by the Media Capture Task Force will take into consideration all extensions they are aware of in an attempt to reduce potential usage conflicts.
16. Change Log
This section will be removed before publication.
Changes since November 23, 2016
[#420] Make removal of device(s) fire devicechange event
[#421] Add privacy indicator requirements
Changes since September 13, 2016
[#398] Editorial: Fix links to MediaStream's methods
[#399] Unroll steps in 'add a track' and 'remove a track'
[#401] Define states to be used in indicator requirements
[#410] Rename list-devices permission to device-info to match permission spec
[#412] Specify rules for when devicechange events may fire
[#415] Remove unions in MediaTrackCapabilities
Changes since June 24, 2016
[#376] Allow firing devicechange just once when inserting a camera with mic
[#378] Editorial: Update URLs in copyright to https
[#379] Device change events also occur during active media capture
[#381] Add pre-grant deviceId alternative
[#383] Editorial: Make constraints webidl optional in gUM call
[#392] Add room fingerprinting to security considerations
Changes since May 13, 2016
[#362] Replace "user media in an iFrame" section with use of new 'user media enabled flag'
[#365] Editorial: update content and format of images
[#367] Editorial: convert to WebIDL contiguous mode
Changes since April 23, 2016
[#348, #346] Removes sourceType from the document
[#349, #339] Use 'originIdentifier' in getUserMedia() steps
[#358, #352] Getusermedia: Pull the permissions check into the per-type loop
[#357, #354] Be specific on what live tracks grant implicit permission
Changes since April 6, 2016
[#333] Restrict devicechange event to apps with device-info permission
[#338] Make ended event a simple event
[#342] Remove obsolte error names table
[#343] Clarify language around stored permission and revocation
[#345] Issue 316: add links to constrainable properties
Changes since February 22, 2016
[#318] MediaStreamTrack.stop(): Address early return for non-locally sourced tracks
[#319] Add text that incorporates Permission API's definition of permission handling.
[#321] Remove track attributes "remote" and "readonly"
[#323] Remove unused images
[#324, #325] editorial fixes
[#329, #330] Clarify that, with permission, deviceID persists as with other storage such as cookies but that groupID never persists
[#331] Use NotAllowedError for permission failure in getUserMedia algorithm
Changes since December 23, 2015
[#301, #292] getSupportedConstraints(): Remove prose about dictionary member default values
[#306, #288] Explain WebIDL limitation for ConstrainablePattern
[#308] Clarify how the generic applyConstraints algorithm applies to MediaStreamTracks
[#309, #267] Specify permission model for nested browsing contexts
[#313, #268] Extend iframe with a new allowusermedia attribute
[#315, #307, #296] Clarify how Capabilites should work
Changes since December 8, 2015
[#286] Fix facingMode in MediaTrackCapabilities and example
[#289] Describe the echoCancellation capability
[#290] Use single double instead of same max and min for the aspectRadio capability in example
[#291] Remove MediaStream active/inactive events
[#284] Clarify that constraint in overconstrainederror is defined in selectSettings
[#297] Add new extension text
Changes since September 25, 2015
[#259] Specify that groupId is to be cleared when cookies are
[#261, #255] Warn against UUID fingerprinting
[#260] Update privacy section with lessons from TAG questionnaire
[#262] Mark sources of fingerprinting
[#258] Reorganize definition of constrainable properties without a registry
[#266] Add detailed steps for removeTrack() and fix a security considerations issue
[#272] Fix applyConstraints steps to avoid unused var and bail on reject
[#273] Add exception for stored permissions in text about stopped sources and revoked permissions
[#278, #250] PING: Refer to security-arch on the MUST inside Best Practice for HTTPS
[#280] Let removeTrack() reference the 'remove a track' steps
[#282] Clarify equality for string constraints
[#279, #193] Used frame rate for generic concept
Changes since August 26, 2015
[#226] Typedef MediaStreamError to be an object.
[#253] Clarify in section 5 that tracks don't request changes, but they can want different settings based on application use of applyConstraints.
[#239] Reject gUM promise with a TypeError if no media types are given.
[#240] Specify that groupid is origin-unique.
[#241] Reference WebIDL-1.
[#245] Make applyConstraints argument optional.
[#247] Clean up language in SelectSettings.
Changes since June 29, 2015
[#204] Use WebIDL [SameObject] extended attribute
[#209] Once you implement a MediaTrackSupportedConstraints member, you support it
[#208] Add WebIDL definitions to Constrainable Pattern section to pass WebIDL validator
[#201, #207] Specify order of add/removetrack and active/inactive events
[#194, #162] New Overconstrainederror
[#213] Fix Not Found Error in gUM algorithm
[#181] Remove text about firing event handlers (from event handler attribute descriptions)
Changes since May 23, 2015
[LC-3016, #171] Referring to html5.1 for describing srcObject in media elements
[LC-3011, #196, #200] Replace 'invoke MediaDevices.getUserMedia()' with a reference to the algorithm
[LC-3017, #188] Changed SourceAvailableError to a NotReadableError DOMException
[LC-3017, #187] Changed PermissionDeniedError to a SecurityError DOMException
[LC-3017, #185] Changed NotSupportedError from a MediaStreamError to a DOMException
[LC-3017, #184] Changed AbortError from a MediaStreamError to a DOMException
[#180, #195] Replaced asynchronously with in parallel
[LC-3017, #186] Change NotFoundError from a MediaStreamError to a DOMException
[#174, #182] Tidy up getUserMedia() algorithm
[LC-3022, #177] Add latency constraint
[#178, #179] Add serializer to MediaDeviceInfo.
[LC-3025, #175] Add note about addtrack and removetrack not being used by this spec
[LC-3018, #163, #172] Use [Exposed]
Changes since April 10, 2015
[PR #156] Fix broken fragment
[Issue #164] Removed detach source from descriptions
[PR #166] Added steps describing the procedure to update a track's muted state.
[PR #167] Refer to internal algorithms for track adding and cloning instead of API layer methods
[PR #168] Remove setting the active state in MediaStream constructor
[PR #169] Use required dictionary track dictionary member on MediaStreamTrackEventInit
[Issue #180] Switch terminology from "asynchronously" to "in parallel" to follow the HTML standard
Changes since March 24, 2015
[PR #151] Remove outdated issues in the doc
[PR #152] using "browsing session" instead of "session"
[PR #153] short descriptions of arguments to callbacks in callback-based gUM
Changes since February 2, 2015
Added getUserMedia() implementation suggestion as discussed in issue #67
Issue 139: Clarified in SelectSettings algorithm that an empty constraint is to be treated as no constraint.
Issue 128: Clarified in SelectSettings that bare values mean exact in the advanced array and ideal otherwise.
[PR#37/Bug 26654] Webidl in Constrainable
[Issue #141] MediaDeviceInfo.label and .groupId should not be nullable
[PR #150] Update text for how constraint dictionaries get extended
Changes since October 27, 2014
Bug 26953: Added more detail to the definition of volume.
Clarified in section 4.1 that synchronization is only an intention because some tracks cannot be synchronized.
Introduced and made consistent use of the term 'constrainable property' everywhere we refer to a property which can have Capabilities, Constraints, and Settings.
Changed constraint definition text using concepts and some direct text from PR 61.
Bug 113 (old 25771): Explanation of constraints in GUM call. Rewrote algorithm with separate SelectSettings step, used both in GUM and in applyConstraints.
Changes since September 24, 2014
Bug 25809: Added note warning about abuse of call-me URLs.
Bug 26918: Added note on clearing deviceId when clearing cookies.
Bug 25777: Added example of capabilities when only two video sizes are available.
Bug 26654: Added ConstrainBoolean.
Bug 26810: All callback-based methods have been converted to use Promises, except for the version of getUserMedia() defined under NavigatorUserMedia.
Changes since September 9, 2014
Bug 22214: How long do permissions persist?
Define algorithm for processing non-required constraints.
Bug 24933: deviceId is not registered as constraints, so apps can't choose device based on the device enumeration
Bug 25609: MediaStreamErrorEvent is incomplete
Changes since August 17, 2014
Bug 25988: Need a list of MediaStreamError "name" values
Bug 26623: Use commonest spelling of "cancellation"
Bug 25767: Missing Ref to Image Capture spec
Bug 22271: Terminology section should not have conformance requirements
Changes since July 4, 2014
Bug 22251: Added new NotFoundError, AbortError, SourceUnavailable errors to gUM call.
Bug 25786: User Agent allowance of files to be substituted for any input device is now permitted but not listed as best practice, i.e., no longer specifically recommended.
Changes since June 19, 2014
Bug 22354: Added privacy and security section.
Bug 25784: "on air" indication is underspecified - separated "access granted" and "on air" indicators.
Bug 26192: add onoverconstrained to MediaStreamTrack
Bug 25776: add groupID to MediaTrackConstraintSet
Bug 25780: Clarify step 3 of MediaStream.clone
Bug 25804: Change 'remote' attribute definition
Bug 25650: In getUserMedia algorithm if user denies permission spec is wrongly redirecting to Constraint Failure.
Bug 25605: Definition of MediaStreamTrackEvent is not complete
Bug 25651: All the links in spec should redirect to specified contents without failure.
Bug 25725: getUserMedia constraints should be non-nullable
Bug 25763: does the ID really have to be exactly 36 char long?
Bug 24934: invalid definition for the "seekable" attribute when MediaStream is set to srcObject.
Removed MediaStreamTrack new state (sourceType none removed as a consequence) (as discussed in bug 25787).
Bug 25801: Remove getNativeSettings()
Changes since May 7, 2014
Clarified that skipping of optional/advanced ConstraintSets is only permitted if they cannot be satisfied, not merely because the User Agent wishes to.
Bug 25855: Clarification about conformance requirements phrased as algorithms
Bug 25803: Mark section entitled "The model: sources, sinks, constraints, and settings" as non-normative
Bug 24015: Add callback to indicate when available media devices change (introduced Navigator.mediaDevices)
Bug 25860: make sure we have a bug to have a getTracks that gives you all the tracks
Bug 25884: applied constraint syntax consensus as realized in June 9 WG email from Peter Thatcher.
Moved getSupportedConstraints() method to MediaDevices object.
Added stricter requirements on the getSupportedConstraints() return value.
Added issue note in Constrainable Pattern section that ideal is not yet defined.
Added issue note for applyConstraints that how multiple unorderedConstraints are to be satisfied together is not yet defined.
Added informative notes that WebIDL discards unknown required properties and that application authors need to use the
getSupportedConstraints() method.
Cleaned up the MediaStream API intro section (mainly MediaStream behavior that have moved to MediaStreamTrack).
The concept of MediaStreamTrack with a detachable source is now used throughout the spec (removed language saying that a MST could be disassociated from its track).
Moved peerIdentity related text to WebRTC.
Changes since March 21, 2014
New webIDL for Constrainable and Constraints.
Bug 24931: changed MediaError to MediaStreamError.
Bug 23817: Redundant TOC headers 8.1 & 9.1
Bug 25230: readyState attribute must be inherited while cloning a MediaStreamTrack
Bug 25249: Source should be detached when a MediaStreamTrack stops for any reason other than stop
Updated Event Summary section to match the spec regarding MediaStreamTrack.stop() (as discussed in bug 25248)
Made the MediaStream() constructor behave like addTrack() WRT adding ended tracks (as discussed in bug 25250).
Bug 25262: MediaStream Constructor algorithm must also check for MediaStreamTracks "ended" state while initializing "active" state.
Bug 25276: Initialization for VideoTrack.selected attribute is missing while specifying steps for "Loading and Playing a MediaStream in a Media Element"
Changed syntax of constraints to use 'require' and 'advanced' and support non-required, non-advanced constraints.
Bug 25360: MediaStreamTrack should not be considered as ended just because remote peer stopped sending data.
Bug 25275: VideoTrackList.selectedIndex initialization conflicts with HTML5 spec, "if no track is selected".
Removed mentioning of MediaStream received from other peer (as discussed in bug 25361).
Bug 22263: Clarify synchronization of tracks in a MediaStream
Bug 25441: Overconstrained muted state should not link with MediaStreamTrack.readyState
Changes since February 18, 2014
Bug 24928: Remove MediaStream state check from addTrack() algorithm.
Bug 24930: Remove MediaStream state check from the removeTrack() algorithm.
Added native settings to tracks.
Removed videoMediaStreamTrack and audioMediaStreamTrack since they are no longer necessary.
Changes since December 25, 2013
Make optional constraints a list of ConstraintSets. Make ConstraintSet an object.
Remove noaccess, move peerIdentity
Add constraints for sampleRate, sampleSize, and echoCancellation.
Aligned text in remainder of document with Constrainable changes.
Removed statements that constraints are not applied to read-only sources
Changes since November 5, 2013
ACTION-25: Switch mediastream.inactive to mediastream.active.
ACTION-26: Rewrite stop to only detach the track's source.
Bug 22338: Arbitrary changing of tracks.
Bug 23125: Use double rather than float.
Bug 22712: VideoFacingMode enum needs an illustration.
Moved constraints into a separate Constrainable interface.
Created a separate section on error handling.
Changes since October 17, 2013
Bug 23263: Add output device enumeration to GetSources
Introduced the Constrainable interface.
Change consensus note on constraints in IANA section.
Removed createObjectURL.
Bug 22209: Should not use MUST requirements on values provided by the developer.
Changes since August 24, 2013
Bug 22269: Renamed getSourceInfos() to getSources() and made the result async.
Bug 22229: Editorial input
Bug 22243: Clarify readonly track
Bug 22259: Disabled mediastreamtrack and state of media element
Bug 22226: Remove check of same source from MediaStream constructor algorithm
Replaced ended with inactive for MediaStream (resolves bug 21618).
Bug 22264: MediaStream.ended set to true on creation
Bug 22272: Permission revocation via MediaStreamTrack.stop()
Bug 22248: Relationship between MediaStreamTrack and HTML5 VideoTrack/AudioTrack after MediaStream assignment
Bug 22247: Setting loop attribute on a media element reading from a MediaStream
Changes since July 4, 2013
Bug 21967: Added paragraph on MediaStreamTrack enabled state and updated cloning algorithm.
Bug 22210: Make getUserMedia() algorithm use all numbered items.
Bug 22250: Fixed accidentally overridden error.
Bug 22211: Added async error when no valid media type is requested.
Bug 22216: Made NavigatorUserMediaError extend DOMError.
Bug 22249: Throw on attempts to set currentTime on media elements playing MediaStream objects.
Bug 22246: Made media.buffered have length 0.
Bug 22692: Updated media element to use HAVE_NOTHING state before media arrives on the played MediaStream and HAVE_ENOUGH_DATA as soon as media arrives.
May 29, 2013
Bug 22252: fixed usage of MUST in MediaStream() constructor description.
Bug 22215: made MediaStream.ended readonly.
Bug 21967: clarified MediaStreamTrack.enabled state initial value.
Added aspectRatio constraint, capability, and state.
Updated usage of MediaStreams in media elements.
May 15, 2013
Added explanatory section for constraints, capabilities, and states.
Added VideoFacingModeEnum (including left and right options).
Added getSourceInfos() and SourceInfo dictionary.
Added isolated streams.
April 29, 2013
Removed remaining photo APIs and references (since we have a separate Image Capture Spec).
March 20, 2013
Added readonly and remote attributes to MediaStreamTrack
Removed getConstraint(), setConstraint(), appendConstraint(), and prependConstraint().
Added source states. Added states() method on tracks. Moved sourceType and sourceId to be states.
Added source capabilities. Added capabilities() method on tracks.
Added clarifying text about MediaStreamTrack lifecycle and mediaflow.
Made MediaStreamTrack cloning explicit.
Removed takePhoto() and friends from VideoStreamTrack (we have a separate Image Capture Spec).
Made getUserMedia() error callback mandatory.
December 12, 2012
Changed error code to be string instead of number.
Added core of settings proposal allowing for constraint changes after stream/track creation.
November 15 2012
Introduced new representation of tracks in a stream (removed MediaStreamTrackList).
Updated MediaStreamTrack.readyState to use an enum type (instad of unsigned short constants).
Renamed MediaStream.label to MediaStream.id (the definition needs some more work).
October 1 2012
Limited the track kind values to "audio" and "video" only (could previously be user defined as well).
Made MediaStream extend EventTarget.
Simplified the MediaStream constructor.
June 23 2012
Rename title to "Media Capture and Streams".
Update document to comply with HTML5.
Update image describing a MediaStream.
Add known issues and various other editorial changes.
Add definitions of MediaStreams and related objects.
Dec 21 2011
Changed to make wanted media opt in (rather than opt out). Minor edits.
Nov 29 2011
Changed examples to use MediaStreamOptions objects rather than strings. Minor edits.
Nov 15 2011
Removed MediaStream stuff. Refers to webrtc 1.0 spec for that part instead.
Nov 9 2011
Created first version by copying the webrtc spec and ripping out stuff. Put it on github.
A. Acknowledgements
The editors wish to thank the Working Group chairs and Team Contact, Harald Alvestrand, Stefan Håkansson, Erik Lagerway and Dominique Hazaël-Massieux, for their support. Substantial text in this specification was provided by many people including
Jim Barnett,
Harald Alvestrand,
Travis Leithead,
Josh Soref,
Martin Thomson,
Jan-Ivar Bruaroey,
Peter Thatcher,
Dominique Hazaël-Massieux,
and Stefan Håkansson. Dan Burnett would like to acknowledge the significant support received from Voxeo and Aspect during the development of this specification.
