Audio Session

1. Introduction

People increasingly consume media (audio/video) through the Web, which has become a primary channel for accessing this type of content. However, media on the Web often lacks seamless integration with underlying platforms. The Audio Session API addresses this gap by enhancing media handling across platforms that support audio session management or similar audio focus features. This API improves how web-based audio interacts with other apps, allowing for better audio mixing or exclusive playback, depending on the context, to provide a more consistent and integrated media experience across devices.

Additionally, some platforms automatically manage a site’s audio session based on media playback and the APIs used to play audio. However, this behavior might not always align with user expectations. This API allows developers to override the default behavior and gain more control over an audio session.

2. Concepts

A web page can do audio processing in various ways, combining different APIs like HTMLMediaElement or AudioContext. This audio processing has a start and a stop, which aggregates all the different audio APIs being used. An audio session represents this aggregated audio processing. It allows web pages to express the general nature of the audio processing done by the web page.

An audio session can be of a particular type, and be in a particular state. An audio session manages the audio for a set of individual sources (microphone recording) and sinks (audio rendering), named audio session elements.

An audio session's element has a number of properties:

• A default type, which is used to compute the audio session type, in case of "auto".

• An audible flag, which is either true if the element is playing/recording audio, or false otherwise.

An audio session element is an audible element if its audible flag is true.

Additionaly, an audio session element has associated steps for dealing with various state changes. By default, each of these steps is empty list of steps:

• Element update steps, which are run whenever the audio session state changes.

• Element suspend steps, which are run when the audio session state moves from active to either interrupted or inactive.

• Element resume steps, which are run when audio session state moves from interrupted to active.

This specification is defining these steps, the default type and the audible flag for some audio session's elements in section § 6 Audio source and sink integration. Specifications defining other elements need to define these steps and properties.

A top-level browsing context has a selected audio session. In case of a change to any audio session, the user agent will update which audio session becomes the selected audio session. A top-level browsing context is said to have audio focus if its selected audio session is not null and its state is active.

3. The AudioSession interface

AudioSession is the main interface for this API. It is accessed through the Navigator interface (see § 4 Extensions to the Navigator interface).

[Exposed=Window]
interface AudioSession : EventTarget {
  attribute AudioSessionType type;
  readonly attribute AudioSessionState state;
  attribute EventHandler onstatechange;
};

To create an AudioSession object in realm, run the following steps:

1. Let audioSession be a new AudioSession object in realm, initialized with the following internal slots:

   1. [[type]] to store the audio session type, initialized to auto.

   2. [[state]] to store the audio session state, initialized to inactive.

   3. [[elements]] to store the audio session elements, initialized to an empty list.

   4. [[interruptedElements]] to store the audio session elements that where interrupted while being audible, initialized to an empty list.

   5. [[appliedType]] to store the type applied to the audio session, initialized to auto.

   6. [[isTypeBeingApplied]] flag to store whether the type is being applied to the audio session, initialized to false.

2. Return audioSession.

Each AudioSession object is uniquely tied to its underlying audio session.

The AudioSession state attribute reflects its audio session state. On getting, it MUST return the AudioSession [[state]] value.

The AudioSession type attribute reflects its audio session type, except for auto.

On getting, it MUST return the AudioSession [[type]] value.

On setting, it MUST run the following steps with newValue being the new value being set on audioSession:

1. If audioSession.[[type]] is equal to newValue, abort these steps.

2. Set audioSession.[[type]] to newValue.

3. Update the type of audioSession.

3.1. Audio session types

By convention, there are several different audio session types for different purposes. In the API, these are represented by the AudioSessionType enum:

playback

Playback audio, which is used for video or music playback, podcasts, etc. They should not mix with other playback audio. (Maybe) they should pause all other audio indefinitely.

transient

Transient audio, such as a notification ping. They usually should play on top of playback audio (and maybe also "duck" persistent audio).

transient-solo

Transient solo audio, such as driving directions. They should pause/mute all other audio and play exclusively. When a transient-solo audio ended, it should resume the paused/muted audio.

ambient

Ambient audio, which is mixable with other types of audio. This is useful in some special cases such as when the user wants to mix audios from multiple pages.

play-and-record

Play and record audio, which is used for recording audio. This is useful in cases microphone is being used or in video conferencing applications.

auto

Auto lets the user agent choose the best audio session type according the use of audio by the web page. This is the default type of AudioSession.

enum AudioSessionType {
  "auto",
  "playback",
  "transient",
  "transient-solo",
  "ambient",
  "play-and-record"
};

An AudioSessionType is an exclusive type if it is playback, play-and-record or transient-solo.

3.2. Audio session states

An audio session can be in one of the following state , which are represented in the API by the AudioSessionState enum:

active

the audio session is playing sound or recording microphone.

interrupted

the audio session is not playing sound nor recording microphone, but can resume when it will get uninterrupted.

inactive

the audio session is not playing sound nor recording microphone.

enum AudioSessionState {
  "inactive",
  "active",
  "interrupted"
};

The audio session's state may change, which will automatically be reflected on its AudioSession object via the steps to notify the state’s change.

4. Extensions to the Navigator interface

Each Window has an associated AudioSession, which is an AudioSession object. It represents the default audio session that is used by the user agent to automatically set up the audio session parameters. The user agent will request or abandon audio focus when audio session elements start or finish playing. Upon creation of the Window object, its associated AudioSession MUST be set to a newly created AudioSession object with the Window object’s relevant realm.

The associated AudioSession list of elements is updated dynamically as audio sources and sinks of the Window object are created or removed.

[Exposed=Window]
partial interface Navigator {
  // The default audio session that the user agent will use when media elements start/stop playing.
  readonly attribute AudioSession audioSession;
};

5. Audio session algorithms

5.1. Update AudioSession’s type

To update the type of audioSession, the user agent MUST run the following steps:

1. If audioSession.[[isTypeBeingApplied]] is true, abort these steps.

2. Set audioSession.[[isTypeBeingApplied]] to true.

3. Queue a task to run the following steps:

   1. Set audioSession.[[isTypeBeingApplied]] to false.

   2. If audioSession.[[type]] is the same as audioSession.[[appliedType]], abort these steps.

   3. Set audioSession.[[appliedType]] to audioSession.[[type]].

   4. Update all AudioSession states of audioSession’s top-level browsing context with audioSession.

   5. For each element of audioSession.[[elements]], update element.

   6. Let newType be the result of computing the type of audioSession.

   7. In parallel, set the type of audioSession’s audio session to newType.

5.2. Update AudioSession’s state

When an audio session element is starting or stopping, the user agent will run steps that set the state of an audio session, via the inactivate and try activating algorithms. Setting an audio session's state to active has consequences, especially if the audio session's type is an exclusive type:

• It can inactivate AudioSession objects of the top-level browsing context, as defined in the algorithms below.

• It can pause the audio of another tab or another application.

Conversely, an audio session state can be modified outside of audio session element changes. When the user agent observes such a modification, the user agent MUST queue a task to notify the state’s change with audioSession, the AudioSession object tied to the modified audio session and with newState being the new audio session state.

To notify the state’s change with audioSession and newState, the user agent MUST run the following steps:

1. Let isMutatingState be true if audioSession.[[state]] is not newState and false otherwise.

2. Set audioSession.[[state]] to newState.

3. If newState is inactive, set audioSession.[[interruptedElements]] to an empty list.

4. For each element of audioSession.[[elements]], update element.

5. If isMutatingState is false, abort these steps.

6. Update all AudioSession states of audioSession’s top-level browsing context with audioSession.

7. Fire an event named statechange at audioSession.

To inactivate an AudioSession named audioSession, the user agent MUST run the following steps:

1. If audioSession.[[state]] is inactive, abort these steps.

2. Run the following steps in parallel:

   1. Set the state of audioSession’s audio session to inactive.

   2. Assert: audioSession’s audio session's state is inactive.

   3. Queue a task to notify the state’s change with audioSession and with its audio session's state.

To try activating an AudioSession named audioSession, the user agent MUST run the following steps:

1. If audioSession.[[state]] is active, abort these steps.

2. Run the following steps in parallel:

   1. Set the state of audioSession’s audio session to active. Setting the state to active can fail, in which case the audio session's state will either be inactive or interrupted.

   2. Queue a task to notify the state’s change with audioSession and with its audio session's state.

5.3. Update the selected audio session

To update the selected audio session of a top-level browsing context named context, the user agent MUST run the following steps:

1. Let activeAudioSessions be the list of all the audio sessions tied to AudioSession objects of context and its children in a breadth-first order, that match both the following constraints:

   1. Its state is active.

   2. The result of computing the type of the AudioSession object is an exclusive type.

2. If activeAudioSessions is empty, abort these steps.

3. If there is only one audio session in activeAudioSessions, set the selected audio session to this audio session and abort these steps.

4. Assert: for any AudioSession object tied to an audio session in activeAudioSessions’s named audioSession, audioSession.[[type]] is auto.

   It is expected that only one audio session with an explicit exclusive type can be active at any point in time. If there are multiple active audio sessions in activeAudioSessions, their [[type]] can only be auto.

5. The user agent MAY apply specific heuristics to reorder activeAudioSessions.

6. Set the selected audio session to the first audio session in activeAudioSessions.

5.4. Other algorithms

To update all AudioSession states of a top-level browsing context named context with updatedAudioSession, run the following steps:

1. Update the selected audio session of context.

2. Let updatedType be the result of computing the type of updatedAudioSession.

3. If updatedType is not an exclusive type or updatedAudioSession.[[state]] is not active, abort these steps.

4. Let audioSessions be the list of all the AudioSession objects of context and its children in a breadth-first order.

5. For each audioSession of audioSessions except for updatedAudioSession, run the following steps:

   1. If audioSession.[[state]] is not active, abort these steps.

   2. Let type be the result of computing the type of audioSession.

   3. If type is not an exclusive type, abort these steps.

   4. If type and updatedType are both auto, abort these steps.

   5. Inactivate audioSession.

To compute the audio session type of audioSession, the user agent MUST run the following steps:

1. If audioSession.[[type]] is not auto, return audioSession.[[type]].

2. If any element of audioSession.[[elements]] has a default type of play-and-record and its state is active, return play-and-record.

3. If any element of audioSession.[[elements]] has a default type of playback and its state is active, return playback.

4. If any element of audioSession.[[elements]] has a default type of transient-solo and its state is active, return transient-solo.

5. If any element of audioSession.[[elements]] has a default type of transient and its state is active, return transient.

6. Return ambient.

6. Audio source and sink integration

This section describes audio session element's steps and properties for AudioContext, HTMLMediaElement and microphone MediaStreamTrack.

An element state is:

• interrupted if it is in its AudioSession's [[interruptedElements]].

• active if it is an audible element.

• inactive otherwise.

To update an element named element, the user agent MUST run the following steps:

1. Let audioSession be element’s AudioSession.

2. Run element’s update steps.

3. If element is an audible element and audioSession.[[state]] is interrupted, run the following steps:

   1. Add element to audioSession.[[interruptedElements]].

   2. Run element’s suspend steps.

4. If element is in audioSession.[[interruptedElements]], and audioSession.[[state]] is active, run the following steps:

   1. Remove element from audioSession.[[interruptedElements]].

   2. Run element’s resume steps.

When the audible flag of one of audioSession’s elements is changing, the user agent MUST run the following steps:

1. If the audible flag is changing to true, try activating audioSession.

2. Otherwise, if any element of audioSession.[[elements]] has a state of interrupted, abort these steps.

3. Otherwise, inactivate audioSession.

6.1. AudioContext

An AudioContext is an element with the following properties:

• Its default type is ambient.

• Its audible flag is true if its state is running and is sending non zero samples to its destination.

• Its suspend steps are:

  1. Let audioContext be the AudioContext object.

  2. Queue a control message to suspend audioContext.

• Its resume steps are:

  1. Let audioContext be the AudioContext object.

  2. Queue a control message to unsuspend audioContext.

When an AudioContext is created, the user agent MUST run the following steps:

1. Let audioContext be the newly created AudioContext.

2. Let audioSession be the AudioSession's object of the Window object in which is created audioContext.

3. Add audioContext to audioSession.[[elements]].

6.2. HTMLMediaElement

A HTMLMediaElement is an element with the following properties:

• Its default type is playback.

• Its audible flag is true if it is playing, its volume is not 0, it is not muted and it has audio tracks.

• Its suspend steps are:

  1. Let mediaElement be the HTMLMediaElement object.

  2. Queue a task to run the internal pause steps of mediaElement.

• Its resume steps are:

  1. Let mediaElement be the HTMLMediaElement object.

  2. Queue a task to run the internal play steps of mediaElement.

When an HTMLMediaElement's node document is changing, the user agent MUST run the following steps:

1. Let mediaElement be the HTMLMediaElement whose node document is changing.

2. Let previousWindow be the Window object associated to mediaElement’s previous node document, if any or null otherwise.

3. If previousWindow is not null, remove mediaElement from previousWindow’s associated AudioSession.[[elements]].

4. Let newWindow be the Window object associated to mediaElement’s new node document, if any or null otherwise.

5. If newWindow is not null, add mediaElement to newWindow’s associated AudioSession.[[elements]].

6.3. Microphone MediaStreamtrack

A microphone capture MediaStreamTrack is an element with the following properties:

• Its default type is play-and-record.

• Its audible flag is true if it is neither ended nor muted.

• Its element update steps are:

  1. Let track be the MediaStreamTrack object.

  2. Let audioSession be track’s AudioSession.

  3. If audioSession.[[type]] is not play-and-record or auto, end track.

• Its suspend steps are:

  1. Let track be the MediaStreamTrack object.

  2. Queue a task to set the muted state of track to true.

• Its resume steps are:

  1. Let track be the MediaStreamTrack object.

  2. Queue a task to set the muted state of track to false.

When a microphone capture MediaStreamTrack is created, the user agent MUST run the following steps:

1. Let track be the newly created MediaStreamTrack.

2. Let audioSession be the AudioSession's object of the Window object in which is created track.

3. Add track to audioSession.[[elements]].

FIXME: We should be hooking to the audio track’s sources stored in the Window’s mediaDevices’s mediaStreamTrackSources, instead of MediaStreamTrack. This should handle the case of transferred’s microphone tracks.

7. Privacy considerations

8. Security considerations

9. Examples

9.1. A site sets its audio session type proactively to "play-and-record"

navigator.audioSession.type = 'play-and-record';
// From now on, volume might be set based on 'play-and-record'.
...
// Start playing remote media
remoteVideo.srcObject = remoteMediaStream;
remoteVideo.play();
// Start capturing
navigator.mediaDevices
  .getUserMedia({ audio: true, video: true })
  .then((stream) => {
    localVideo.srcObject = stream;
  });

9.2. A site reacts upon interruption

navigator.audioSession.type = "play-and-record";
// From now on, volume might be set based on 'play-and-record'.
...
// Start playing remote media
remoteVideo.srcObject = remoteMediaStream;
remoteVideo.play();
// Start capturing
navigator.mediaDevices
  .getUserMedia({ audio: true, video: true })
  .then((stream) => {
    localVideo.srcObject = stream;
  });
navigator.audioSession.onstatechange = async () => {
  if (navigator.audioSession.state === "interrupted") {
    localVideo.pause();
    remoteVideo.pause();
    // Make it clear to the user that the call is interrupted.
    showInterruptedBanner();
    for (const track of localVideo.srcObject.getTracks()) {
      track.enabled = false;
    }
  } else {
    // Let user decide when to restart the call.
    const shouldRestart = await showOptionalRestartBanner();
    if (!shouldRestart) {
      return;
    }
    for (const track of localVideo.srcObject.getTracks()) {
      track.enabled = true;
    }
    localVideo.play();
    remoteVideo.play();
  }
};

10. Acknowledgements

The Working Group acknowledges the following people for their invaluable contributions to this specification:

• Becca Hughes

• Mounir Lamouri

• Zhiqiang Zhang