Audio Session Defaults#1019
Open
pblazej wants to merge 11 commits into
Open
Conversation
When both playout and recording stop, reset the category to .ambient before deactivating. Without this, the iOS volume rocker stays on the ringer/call register because the last-active category was .playAndRecord, causing in-call volume to appear to "stick" when the user adjusts the rocker after the call ends. .ambient mixes with other audio so any resumed media (e.g. Music) isn't re-interrupted if the session momentarily reactivates afterwards. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two independent motivations:
- .mixWithOthers is a known cause of echo in real-time communication
scenarios where other apps share the audio device — once the option
is set, other apps' audio mixes into the capture path and degrades
the acoustic echo cancellation reference.
- Our WebRTC ADM has a retry loop in the engine restart path
working around -66637 (kAudioUnitErr_Initialized) on
interruption-end recovery when this option is active. Removing
the option from .playAndRecord eliminates the workaround's
triggering condition. Related customer symptom: #1011.
.allowAirPlay is redundant under both .voiceChat and .videoChat per
Apple QA1803 — .voiceChat disallows AirPlay outright, .videoChat
auto-allows the mirrored variant.
.mixWithOthers is intentionally kept on the .playback config used for
listener-only sessions, where mixing with media playback is correct.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The audio session category was re-evaluated on every SessionRequirement
change. Muting via setMicrophone(enabled: false) would downgrade from
.playAndRecord back to .playback, then unmuting would upgrade again —
causing N category switches per session, each one flipping the iOS
volume rocker between the media and ringer registers.
Selection now flows through `selectConfiguration(state:)` and lands on
.playAndRecord when any of the following holds, staying there until
both playout and recording stop:
- recording is enabled (first setMicrophone(enabled: true) or external
acquire(requirement: .recording))
- recording was previously enabled in the same session (sticky bit
`hasEverRecorded`, cleared only on the empty edge)
- the participant wants to publish: the app declared publishing intent
via isRecordingAlwaysPreparedMode AND the local participant has
permission to publish a microphone (`canPublishMicrophone`, driven
by ParticipantPermissions)
Pure audience participants (no permission, never recorded, no current
recording requirement) keep .playback for the entire session. The mic
publish permission defaults to true here (optimistic) and is corrected
once permissions arrive from the server.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the local participant's permissions change (server-issued at JoinResponse time, or via a mid-session permission update), forward the derived "can publish microphone" predicate to the audio session so it can pick `.playback` for audience-only participants without waiting for a publish attempt to fail. Extracts a `canPublish(source:)` helper from the existing `checkPermissions(toPublish:)` so the two callers share one predicate definition. The helper combines the top-level `canPublish` grant with the per-source restriction in `canPublishSources` (empty = no restriction). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Comment out the `&& AudioManager.shared.isRecordingAlwaysPreparedMode` conjunct in selectConfiguration so the pre-emptive .playAndRecord upgrade is purely server (permission) driven. Any participant with mic-publish permission will get .playAndRecord up-front instead of waiting on the app-side intent signal. Trade-off: closer to the "always playAndRecord for permitted users" pattern. Participants who could publish but never do still get the mic permission prompt and ringer volume register on connect — keeping the intent signal commented makes restoration trivial once we validate the desired behaviour. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reflects the behavior shipped in this branch: - .ambient reset on the empty edge before deactivating - Listener vs publisher category selection - Sticky .playAndRecord once recording engages - Audience-only participants (canPublish=false) stay on .playback Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Aids diagnosing why a particular AVAudioSession category was picked without attaching a debugger. Logs the three input signals (current recording state, sticky bit, mic-publish permission) plus the speaker preference and the resulting category. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
wantsToPublish now fires when either signal indicates publishing intent: server-issued mic permission OR the app-declared isRecordingAlwaysPreparedMode flag. Either alone is sufficient to pre-empt to .playAndRecord. In practice the alwaysPrepared path already triggers engine recording via the ADM (which sets isRecordingEnabled, then the sticky bit), so the additional disjunct rarely fires standalone — but it makes the publisher-hint role of the flag explicit in the predicate rather than buried in a comment. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Updates the bullets to match the predicate after the isRecordingAlwaysPreparedMode hint was OR'd with canPublishMicrophone. Audience-only is now defined by absence of both signals, and the Publisher bullet enumerates all three entry points. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
pblazej
commented
May 27, 2026
| /// server-issued mic permission (`canPublishMicrophone`) or the | ||
| /// app-declared `isRecordingAlwaysPreparedMode`. | ||
| private func selectConfiguration(state: State) -> AudioSessionConfiguration { | ||
| let wantsToPublish = state.canPublishMicrophone || AudioManager.shared.isRecordingAlwaysPreparedMode |
Contributor
Author
There was a problem hiding this comment.
This is a heuristic ofc, open to suggestions.
Contributor
Author
|
📱 TF build 2.15.0.b20260527 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Reworks the iOS audio session defaults and category selection. No public API additions; observable behavior changes are listed below with rationale.
.playAndRecordoptions trimmed:.mixWithOthersand.allowAirPlaydropped (removes the-66637interruption-recovery race and an option that's redundant per Apple QA1803)..mixWithOtherskept on the listener.playbackconfig..mixWithOthershas a turbulent history — added in Adjust AVAudioSession CategoryOptions (include .mixWithOthers) #228, removed in Remove.mixWithOtherswhen category.playAndRecord#260, then re-entered the defaults during the AVAudioEngine ADM rewrite (AVAudioEngine version AudioDeviceModule #536); this is a clean revert.-66637engine-init race with a retry loop when this option is on).selectConfiguration(state:)with sticky upgrade — picks.playAndRecordwhen any of: recording is active, recording has happened in this session (stickyhasRecorded), the participant has mic-publish permission (canPublishMicrophone), or the app setisRecordingAlwaysPreparedMode. Otherwise.playback. No more category churn on mute toggles..ambientreset on the empty edge (post-Room.disconnect()) so the iOS volume rocker returns to the media register — fixes the long-standing "in-call volume sticks" symptom.LocalParticipant.set(permissions:)forwards the mic-publish predicate to the audio session via a new sharedcanPublish(source:)helper extracted from the existingcheckPermissions(toPublish:);Docs/audio.mdgains an "Audio session category selection" section.