PR #37 inlined VadWorkerPolicy into the desktop capture path but missed
the iOS files. Instead of restoring the deleted types, inline the same
direct if-let-Some pattern into ios_voice_unit.rs (the only remaining
iOS backend) and permanently remove VadWorkerPolicy and
callback_vad_worker_policy from vad/mod.rs.
iOS uses Option<AppleCoreMlVadWorker> with &mut self (no Mutex), so the
inline is simpler than the desktop's try_lock pattern.
Delete ios_raw_unit.rs (538 lines) and all SonoraExperimental references
from the core audio crate. The experimental RemoteIO path that bypassed
Apple VPIO in favor of software WebRTC APM was never shipped and is no
longer needed. VPIO is the sole production iOS audio backend.
- Delete ios_raw_unit.rs entirely
- Remove Raw variant from IosVoiceBackend enum in engine.rs
- Remove SonoraExperimental from IosVoiceProcessingMode enum
- Simplify validate_for_ios() (single-variant enum, no mode check)
- Remove 2 SonoraExperimental validation tests
- Remove ios_raw_unit module declaration from lib.rs
- Remove include_str!-based debug_wav test for the deleted file
* feat(audio): add Silero ONNX VAD with WebRTC fallback
Introduce SileroOnnxVad and SileroOnnxVadWorker for desktop targets. The worker runs Silero v6 ONNX inference on a dedicated thread, accumulating 10 ms frames into the 512-sample 16 kHz input the model expects. Add VadOutput, VoiceActivityDetector trait, and WebRtcFallbackVad to provide a uniform VAD interface with graceful fallback when the ONNX model is unavailable. Wire the new VadBackend variants through AudioProcessingConfig and the snapshot stats so the bridge can report which detector is active.
* feat(audio): integrate desktop VAD worker into capture engine
Wire SileroOnnxVadWorker into the desktop capture path so voice activity can open the transmit gate before encoding. The capture callback now processes all audio through resample, downmix, and VAD unconditionally; transmit_active still gates Opus encoding.
Add new_desktop_audio_processing_state() to construct the config/stats/worker triple, and apply_desktop_vad_backend() to synchronously load or clear the worker on config changes. Override processing_backend to Noop for desktop so bridge diagnostics report the correct backend rather than the iOS-oriented PlatformVoiceProcessing default.
Includes review-driven cleanups: StreamConfig clone to deref per clippy, and a comment explaining why two try_lock calls on silero_vad_worker are structurally necessary (borrow checker requires the policy probe and the fallback path to not share a lock guard because mark_vad_fallback_active takes &mut self).
* fix(audio): modernize Windows PTT to current windows-rs API
Port the Raw Input plus low-level keyboard hook PTT backend to the newer windows-rs patterns: OptionalHandle, Result-returning CreateWindowExW, and None for CallNextHookEx. Replaces the old HHOOK(0) pointer casts. Add deterministic tests for mouse button 4 and 5 press and release driving the gate.
* build(windows): force MSVC release CRT for audiopus cmake builds
audiopus_sys calls cmake::build(opus_path), so downstream Cargo env cannot use cmake-rs Config::define() to override CMake's MSVC Debug CRT defaults. Point cmake-rs at a small wrapper that injects the policy and cache variables during configure while passing cmake --build, --version, and -E through unchanged. Keeps Opus Debug builds on Rust's release dynamic CRT (/MD) instead of CMake's default debug CRT (/MDd), which otherwise pulls in unresolved __imp__CrtDbgReportW symbols at test link.
Document that the iOS deployment target is intentionally absent from this file. It is enforced by tools/build-ios.sh and the Xcode project; setting it globally here would make native macOS cargo check runs try to link iPhone objects against the macOS SDK.
* build(flutter): update pubspec.lock after plugin additions
Regenerated lockfile reflecting the local_notifications and connectivity_plus plugin additions from the poke-notifications feature.
* fix(audio): address PR #37 review findings
Six fixes from independent PR review:
1. BLOCKER: Replace Windows-only cmake .cmd wrapper with cross-platform
CMake env vars. Setting CMAKE=tools/cmake-msvc-release-crt.cmd
globally broke non-Windows hosts because cmake-rs would try to
execute a .cmd file on macOS/Linux. Instead, set
CMAKE_POLICY_DEFAULT_CMP0091=NEW and CMAKE_MSVC_RUNTIME_LIBRARY=
MultiThreadedDLL as env vars that CMake reads natively. MSVC-
specific vars are safely ignored by GCC/Clang toolchains. Delete
the now-unnecessary wrapper script.
2. IMPORTANT: Join the Silero worker thread in Drop instead of
detaching it. The old code dropped the JoinHandle which detaches
the thread; the new code calls handle.join() after closing the
channel, ensuring the ONNX session is cleaned up before the
worker is replaced during config changes.
3. IMPORTANT: Single-try_lock refactor of the capture VAD callback.
The double try_lock (policy probe + send) is replaced by a single
scoped try_lock that both probes availability and sends the frame.
The guard is dropped before the fallback path, which needs &mut
self for mark_vad_fallback_active. This also eliminates the
VadWorkerPolicy enum and callback_vad_worker_policy function,
whose behavior is now inlined into the callback.
4. IMPORTANT: Remove tracing from the realtime capture callback.
mark_vad_fallback_active and sync_vad_backend emitted info!/warn!
from the audio thread. Replace with silent atomic state
publishing via SharedAudioProcessingStats; the bridge stats
stream already exposes vad_fallback_active for diagnostics.
5. IMPORTANT: Defer ONNX model load outside the worker mutex.
apply_desktop_vad_backend_to_worker now constructs the new worker
before taking the lock, then swaps it in under a short hold.
This prevents the realtime callback from being blocked during
model I/O + thread spawn.
6. MINOR: Remove unused VadBackend import from vad/mod.rs after
deleting the policy code.
* fix(audio): address PR #37 second-pass review findings
5-agent review found 5 blocking issues. All addressed:
1. BLOCKER: CMake env vars don't reach CMake cache. Restored .cmd wrapper
but scoped to Windows MSVC targets only via [target.x86_64-pc-windows-msvc]
and [target.aarch64-pc-windows-msvc] in .cargo/config.toml. Non-Windows
hosts are unaffected.
2. BLOCKER: processing_backend normalized in set_audio_processing_config
on desktop (cfg-gated override to Noop), mirroring startup default.
3. BLOCKER: Model-path reload was already wired via reload_audio_processing_config.
Fixed misleading doc comment in core/lib.rs.
4. BLOCKER: DEC-030 updated to reflect desktop VoiceActivity enablement.
Traceability docs (SRS, SysDes, SAD, SDD, implementation-status) updated.
5. Silero ONNX cfg narrowed to desktop-only (excludes macOS/Android).
Cargo.toml ort dependency target cfg narrowed similarly.
6. Realtime callback debt documented as TODO at CaptureState::ingest.
* fix(audio): exclude ort dep on Android target
ort does not provide first-class Android prebuilts in our pin, mirror the
iOS/macOS exclusion so cargo metadata succeeds for android targets.
* test(audio): fix stale select_ptt_backend import in ptt_privacy
The helper moved out of the ptt_backends submodule onto the crate root;
update the integration test imports so the test compiles again.
* build(windows): scope MSVC release CRT cmake wrapper via Cargo [env]
Cargo's [target.<triple>] table only forwards a fixed allowlist
(linker, runner, rustflags, rustdocflags, ar), so setting CMAKE there
was silently dropped and audiopus_sys kept linking the debug CRT,
producing LNK4098 'MSVCRTD conflicts' and __imp__CrtDbgReportW errors
on x86_64-pc-windows-msvc test builds.
Move the override to Cargo's [env] table using cc/cmake-rs's
target-suffixed CMAKE_<triple> lookup (force=true, relative=true) so it
applies to MSVC targets only and not to host tooling. Add stdout
markers to the wrapper so its invocation is provable in cargo -vv logs.
Verified: cargo test -p chanora_audio --target x86_64-pc-windows-msvc
--lib --no-run now links cleanly; CMakeCache.txt records
CMAKE_MSVC_RUNTIME_LIBRARY=MultiThreadedDLL and CMP0091=NEW.
* fix(flutter): gate VoiceActivity transmit mode by platform support
VoiceActivity relies on the native VAD worker, which is only wired up
on Windows, Linux, and Android. Showing the option on iOS, macOS, or
web let users select a mode that silently never transmitted.
Add voiceActivityTransmitAvailable + transmitModeSegmentsFor() helpers
in voice_settings_controls.dart, hide the VAD row in voice_compact.dart
and drop the VAD segment from the settings dialog when unsupported.
Keep the legacy const transmitModeSegments for the existing widget test
and add two new tests covering the gated helper.
* fix(ios-audio): add voice join session coordinator
* fix(ios-audio): activate session before voice joins
* docs(ios-audio): align activation lifecycle comments
* fix(ios-audio): keep session active when already-in-channel
The 'already in channel' server response (code 0x0302) is treated as a
successful join by _onJoinChannel: the user stays in the channel and
local state is updated to reflect the joined target. But the underlying
voiceJoin call still raises BridgeError_ServerRejected, which the
joinVoiceChannelWithIosAudioSession helper used to interpret as a join
failure and deactivate the iOS audio session. Result: the UI shows the
user as joined while the audio session is dead and capture/playback
remain silent.
Add an isJoinSuccess predicate to the ordering helper. When the
predicate matches, the helper rethrows (so the caller can still run its
success-on-already-joined branch) without deactivating the session.
Wire _onJoinChannel to pass _isAlreadyInChannel as the predicate so the
0x0302 path keeps the session active.
Adds two regression tests covering the success-on-rethrow and the
predicate-false-still-deactivates paths.
* docs(security): regenerate license inventories
Cargo inventory: pick up chanora_resolver bump from 0.1.0 to
0.2.0-beta.1 so it matches the workspace; also adds a trailing newline
so 'cargo about generate' is idempotent in CI license-drift checks.
Flutter inventory: pick up flutter_local_notifications (+ platform
interfaces) and timezone pulled in by the prior notification
permission work.
Adopt a call-scoped VoIP audio session lifecycle so other apps' audio is
not stopped while Chanora is idle and the in-call session does not get
clobbered by media-server resets unrelated to voice.
AppDelegate.swift
- Set .ambient + .mixWithOthers as the idle baseline so the app does not
hold a VoiceChat session when no call is active.
- Switch to .playAndRecord + .voiceChat + .mixWithOthers + .duckOthers
on demand via the new chanora/ios_audio_session MethodChannel, and
revert to .ambient on deactivate.
- Gate the media-services-reset rebuild on voiceSessionActive so a
stray reset during idle no longer reactivates VoiceChat.
ios_audio_session_controller.dart (new)
- Thin Dart wrapper around chanora/ios_audio_session with activate/
deactivate; no-op on non-iOS; swallows PlatformException to keep
audio start/stop resilient to platform-side races.
main.dart
- Activate the iOS audio session on BridgeEvent_AudioStarted, deactivate
on BridgeEvent_AudioStopped, fire-and-forget via unawaited().
ios_voice_unit.rs
- Add the 10 local bindings required by the render-callback closure
preamble (wav_recorder_for_render, render_recorder_active,
render_ref_len, render_ref_accum, cb_count, last_num_frames,
num_frames_changes, callbacks_with_audio, callbacks_with_silence)
so the iOS target compiles cleanly with the new lifecycle wiring.
Tests
- 5 unit tests in test/services/ios_audio_session_controller_test.dart
cover activate/deactivate on iOS, no-op on non-iOS, and graceful
PlatformException handling.
Docs
- SRS SRS-110 expanded to cover the call-scoped lifecycle invariant.
- SysDes mobile-voice row updated to reflect the MethodChannel and
.ambient idle baseline.
- implementation-status-2026-05-28 voiceChat row flipped to done.
Verification
- flutter analyze: No issues found (2.8s)
- flutter test: 195 passed / 2 skipped / 0 failed
- cargo build -p chanora_audio --target aarch64-apple-ios: clean
- cargo build -p chanora_audio (macOS host): clean
Device QA matrix (Spotify-keeps-playing-while-idle, mix-during-call,
revert-on-call-end, media-services-reset-during-idle) remains pending
on physical hardware.
- _showLocalNetworkDeniedSnackBar: wrap Process.run with unawaited()
and .catchError() so a rejected future (e.g. macOS sandbox refuses
fork, or 'open' is missing) cannot bubble into the Flutter zone
as an unhandled exception. The synchronous try/catch was a no-op
because Process.run only throws asynchronously.
- macos_permissions_service_test.dart: mirror the
triggerLocalNetworkPrompt error-handling test with one for
checkLocalNetworkAccess. Probe-path failures (NWConnection probe
cannot establish, or Swift side throws) must fall back to cached
state without crashing the caller.
Tests: 186 passed, 2 skipped. Dart analyze clean.
Oracle re-review on PR #30 flagged that _ChatDetailViewState only
called onDraftChanged when _textCtl.text was non-empty. The empty
case is load-bearing: if the user restored a saved draft, deleted
the text, then switched target (or closed the panel), the parent's
draft map kept the stale entry and resurrected it on the next swap.
Fix: call onDraftChanged unconditionally in both didUpdateWidget
(target change) and dispose (tear-down), so the parent map learns
when a draft is now empty.
Adds a regression test exercising the restore-clear-swap sequence.
- ViewportInfo.updateShouldNotify: compare layoutClass only
(not width/height), avoiding unnecessary rebuilds on every
resize frame within the same layout class.
- ChatPanel: use BorderDirectional(start:) for RTL support.
- ChatPanel: localize 'Close chat' tooltip via AppL10n.chatCloseAction.
- Inline panel snackbar: localize via AppL10n.chatPanelCollapsedHint.
New en/zh ARB entries added.
- _saveCurrentDraft(): removed — it was a self-assignment no-op.
Draft persistence relies on ChatDetailView's didUpdateWidget
(fires onDraftChanged on target switch) and dispose (fires on
panel tear-down), both of which already populate _chatDrafts
correctly without an explicit save call.
- _handleInlineChatViewport layout snackbar: use AppL10n.
- Audio level-meter: switch from callback-count (% 3) to time-based
gating (std::time::Duration::from_millis(33)), robust to cpal
buffer-size or sample-rate changes. Remove level_decimation_counter.
- chat_panel_test.dart: add AppL10n.localizationsDelegates so the
test resolves l10n keys.
Tests: 183 passed, 2 skipped. Dart analyze clean.
cargo test -p chanora_audio --lib: 125 passed.
Oracle re-review nit on PR #27: cite Apple's App Sandbox semantics
explicitly. The macOS sandbox classifies any UDP bind() against a
local port as a 'server' operation (covered by network.server),
even when the socket is only used to sendto() a remote peer. This
is the bind()-then-sendto() pattern tokio's UdpSocket uses
internally for tsclientlib's outbound voice traffic. Correct the
sandbox log line to the actual deny string ('Sandbox: ... deny(1)
network-bind') and reference Apple's entitlement reference wording.
- ios_voice_unit.rs: add producer_shutdown AtomicBool flag (macOS only).
The macOS start path spawns a tokio producer task that holds clones
of Arc<Mutex<AudioHandler>>, Arc<ArrayQueue<f32>>, and the output
gain/muted atomics, then loops on a 20 ms tokio interval. Without
a shutdown signal the task runs forever on engine stop/restart and
leaks all four Arcs every cycle. Drop now stores 'true' on the
flag; the producer checks it at the top of each tick and exits,
releasing its clones within at most one 20 ms tick.
- macos/Runner/Release.entitlements: strengthen the existing
justification comment for com.apple.security.network.server.
Document the specific failure mode (tokio::net::UdpSocket::bind
-> sandbox 'network-outbound deny' -> EPERM) and explain why
network.client alone does not cover bind()-then-sendto. The
entitlement is required, not over-broad.
cargo check (host + aarch64-apple-darwin): clean
cargo test -p chanora_audio --lib: 133 passed
flutter test: 186 passed, 2 skipped
dart analyze: clean
The PR #27 refactor moved the iOS render callback to a direct-fill
path with its own peak_i16 == 0 check, dropping the !muted gate
that PR #28 added to the pre-refactor callback. Without this
gate, every muted callback fires a false-positive
increment_output_underrun() because the downmix helper writes
silence (peak_i16 = 0) by design when muted.
Re-apply PR #28's gate to the refactored iOS path so this PR does
not silently reintroduce the bug PR #28 was opened to fix.
Verified:
- cargo test -p chanora_audio --lib: 133 passed, 0 failed
- cargo build -p chanora_audio --target aarch64-apple-ios: clean
Inventory regenerated after refactor(audio): share AudioHandler between iOS and macOS, bump deps (6214139), which dropped windows-core 0.54.0 (and its windows-result 0.1.2 transitive dep). Apache-2.0 crate count drops from 333 to 327. No license-class change.
apps/chanora_flutter/macos/Runner/MacOSAudioLifecycle.swift (new): native MethodChannel handler for chanora/macos_audio_lifecycle. Observes Core Audio HAL default-input and default-output device property changes via AudioObjectAddPropertyListener; posts handleDefaultDeviceChange events with {role: input|output} payload. Mirrors the iOS chanora/ios_audio_lifecycle event surface minus the AVAudioSession-specific events (no interruption / no media services reset equivalents on macOS — no AVAudioSession).
apps/chanora_flutter/macos/Runner/MainFlutterWindow.swift: register MacOSAudioLifecycle next to MacOSPermissionsHandler in awakeFromNib. Closes the iOS/macOS asymmetry noted in SysRS-051.
apps/chanora_flutter/lib/services/audio_lifecycle_service.dart: add wireMacosAudioLifecycle() parallel to wireIosAudioLifecycle() / wireAndroidAudioLifecycle(). The current implementation captures and logs the events; the FRB function that triggers a VPIO re-bind on the engine is a follow-up. Event-shape mirrors the iOS side so a future caller can switch on platform without changing the dispatch shape.
apps/chanora_flutter/test/services/audio_lifecycle_service_test.dart: smoke test for wireMacosAudioLifecycle (4 tests pass, including the new one).
docs/architecture/sdd.md: three new rows in the Audio Detailed Design table. 'VoiceActivity gate (capture-side)' documents voice_activity::VoiceActivityStateMachine — the 10 ms-cadence gate for TransmitMode::VoiceActivity with open-after (40 ms) / hangover (500 ms) / min-tx (200 ms) / weak-hold (30-100 frames) timers, plus live configure() re-clamping behaviour. 'macOS render cadence (producer + ring)' documents the 20 ms tokio producer task + crossbeam ArrayQueue ring with 100 ms prebuffer, the design chosen to decouple ingress quantums (20 ms Opus frames) from egress quantums (whatever VPIO asks for). 'iOS render cadence (direct-fill)' documents the direct-fill callback path with preallocated 4096x2 f32 scratch buffer and try_lock semantics (not blocking lock).
Closes the VoiceActivity / macOS-producer-ring / iOS-direct-fill doc gaps flagged in the PR #27 'Deferred' list.
apps/chanora_flutter/macos/Runner/Release.entitlements: add com.apple.security.network.server = true. The macOS App Sandbox treats every UDP bind() — including the ephemeral 0.0.0.0:0 that tsclientlib uses for outbound TS3 traffic — as a server operation. Without this entitlement UdpSocket::bind fails with EPERM and the TS3 connect never starts. Debug builds already had this entitlement (needed for flutter run hot-reload); release builds were missing it.
apps/chanora_flutter/macos/Runner/DebugProfile.entitlements: expand the existing network.server comment to document the dual rationale (flutter hot-reload + outbound UDP bind), so the entitlement's purpose is clear without spelunking through tsclientlib.
crates/chanora_audio/src/engine.rs: drop the macOS-specific event-queue producer/consumer path; macOS now uses the iOS-style direct AudioHandler::fill_buffer in the VPIO render callback. The shared AudioHandler is an Arc<Mutex<...>>; the realtime callback uses try_lock so it never blocks on the tokio decode task (see ios_voice_unit.rs render callback).
crates/chanora_audio/src/mobile_voice_backend.rs: update VoiceAudioParams cfg gates — handler is now the iOS/macOS/desktop shape (Arc<Mutex<AudioHandler<SessionAudioId>>>), event_producer is Android-only.
crates/chanora_audio/src/lib.rs: widen the audio_event_queue module visibility to test so the macOS-specific path can be exercised by the unit test suite.
Cargo.toml: bump cpal 0.17.3 -> 0.18.0, jni 0.21 -> 0.22.4, windows 0.54 -> 0.62, criterion 0.5 -> 0.8. Cargo.lock follows.
Edison Jwa