feat: macOS architecture, cloud ASR/LLM providers, and 6-step iOS onboarding

- Add macOS menu-bar dictation app with local ASR models (SenseVoice/Qwen3), global Option hotkey, and bottom overlay - Add cloud ASR/LLM providers (Anthropic, Volcengine, Bailian, and more) with provider logos, model listing, and connection checks - Add shared 7-day usage stats UI (UsageStatsCluster / SevenDayUsageChart) - Add iOS onboarding step 6 for polish LLM setup; hide custom-language-model diagnostic toggle behind DEBUG - Unify iOS onboarding tagline with the macOS brand line ("开口即文字。") - Rewrite README (Chinese-first, product-oriented) and refresh GitHub Pages
2026-07-11 19:10:20 +08:00
parent cdf833935a
commit cc8dd1070a
116 changed files with 6659 additions and 2634 deletions
@@ -105,6 +105,12 @@ public final class KeyboardViewController: UIInputViewController {
        super.viewDidAppear(animated)
        disableSystemGestureDelays()
        keyboardHeightConstraint?.constant = targetKeyboardHeight
+        refreshReturnKeyRole()
+    }
+
+    public override func textDidChange(_ textInput: UITextInput?) {
+        super.textDidChange(textInput)
+        refreshReturnKeyRole()
    }

    public override var preferredScreenEdgesDeferringSystemGestures: UIRectEdge {
@@ -195,6 +201,21 @@ public final class KeyboardViewController: UIInputViewController {
        }
    }

+    private func refreshReturnKeyRole() {
+        state.returnKeyRole = returnKeyRole(for: textDocumentProxy.returnKeyType ?? .default)
+    }
+
+    private func returnKeyRole(for returnKeyType: UIReturnKeyType) -> State.ReturnKeyRole {
+        switch returnKeyType {
+        case .send, .go, .search, .join, .route, .google, .yahoo, .continue, .emergencyCall:
+            return .send
+        case .default, .next, .done:
+            return .newline
+        @unknown default:
+            return .newline
+        }
+    }
+
    // MARK: - System keyboard chrome

    private func configureDictationBehavior() {
@@ -48,8 +48,21 @@ final class KeyboardFlowCoordinator {
    /// `adoptHostBusyStateIfNeeded` from re-entering `.processing` after a
    /// stale App Group snapshot still says `reason=processing`.
    private var lastConsumedUtteranceId: UUID?
+    /// Utterance we just asked the host to stop. Until the host publishes
+    /// processing/final state, stale App Group snapshots can still say
+    /// `reason=recording`; do not re-adopt that utterance as locally active.
+    private var lastStoppedUtteranceId: UUID?
    private var currentCommandSeq: Int64 = 0
    private var lastAvailabilityTraceSignature = ""
+    /// When true, `completeFlowStartHandoff` starts recording after the host
+    /// publishes ready — set only for an explicit mic press.
+    private var recordAfterHandoff = false
+    /// When true, `startHostReadyWaitIfNeeded` starts recording once ready
+    /// (mic pressed while session was still warming / mid ready-flap).
+    private var recordWhenHostReady = false
+    /// Ignores single-frame "host dead" samples before allowing a cold-start jump
+    /// from non-press recovery paths.
+    private var coldStartDebouncer = FlowColdStartDebouncer()

    init(
        state: KeyboardState,
@@ -120,6 +133,8 @@ final class KeyboardFlowCoordinator {

        recomputeMicVoiceAvailability()
        startHostReadyWaitIfNeeded()
+        // Proactive host auto-launch is disabled (FlowHandoffPolicy): a single
+        // stale ready snapshot after finalize must never open startflow.

        // Only surface "session ended" when the session contract *genuinely*
        // dropped (expired / cleared). A transient host-ready flap — engine
@@ -221,6 +236,7 @@ final class KeyboardFlowCoordinator {
            // forever miss the real delivery and leaves the mic white forever.
            guard let busyId = snapshot.busyUtteranceId else { return }
            guard busyId != lastConsumedUtteranceId else { return }
+            guard busyId != lastStoppedUtteranceId else { return }
            activeSessionId = sessionId
            currentUtteranceId = busyId
            isPendingFlowStart = false
@@ -268,6 +284,7 @@ final class KeyboardFlowCoordinator {
        state.lastTranscript = ""
        stopFlowWatchdog()
        currentUtteranceId = nil
+        lastStoppedUtteranceId = nil
        traceState(
            "stickyProcessing.cleared",
            extra: hostReady ? "hostReady=1" : "hostReady=0"
@@ -280,16 +297,24 @@ final class KeyboardFlowCoordinator {
        guard !isPendingFlowStart else { return }
        guard FlowSessionBridge.isSessionActive() else {
            stopHostReadyWait()
+            if recordWhenHostReady {
+                // Session gone while waiting — escalate to a real cold start.
+                let shouldRecord = recordWhenHostReady
+                recordWhenHostReady = false
+                beginFlowStart(recordAfterHandoff: shouldRecord)
+            }
            return
        }
        // Host busy ≠ waiting for ready. Do not spin the ready-wait poll.
        if let reason = FlowSessionBridge.readySnapshot()?.reason,
           reason == .recording || reason == .processing {
            stopHostReadyWait()
+            recordWhenHostReady = false
            return
        }
-        guard !FlowSessionBridge.isHostReady() else {
+        if FlowSessionBridge.isHostReady() {
            stopHostReadyWait()
+            finishHostReadyWaitIfNeeded()
            return
        }

@@ -300,16 +325,50 @@ final class KeyboardFlowCoordinator {
                guard let self, !Task.isCancelled else { return }
                FlowSessionBridge.reloadFromDisk()
                self.recomputeMicVoiceAvailability()
-                if self.state.micVoiceAvailability.isReady
-                    || self.state.micVoiceAvailability == .recording
+                if self.state.micVoiceAvailability.isReady {
+                    self.finishHostReadyWaitIfNeeded()
+                    return
+                }
+                if self.state.micVoiceAvailability == .recording
                    || self.state.micVoiceAvailability == .processing {
+                    self.recordWhenHostReady = false
+                    return
+                }
+                // Host died mid-wait — only cold-start after debounced dead samples.
+                let dead = FlowHandoffPolicy.shouldOpenHostColdStart(
+                    sessionActive: FlowSessionBridge.isSessionActive(),
+                    hostReachable: FlowSessionBridge.isHostReachable(),
+                    hostStale: FlowSessionBridge.isHostStale(),
+                    withinReadyGrace: false
+                )
+                if self.coldStartDebouncer.observe(hostTrulyDead: dead) {
+                    let shouldRecord = self.recordWhenHostReady
+                    self.recordWhenHostReady = false
+                    self.coldStartDebouncer.reset()
+                    self.beginFlowStart(recordAfterHandoff: shouldRecord)
                    return
                }
                try? await Task.sleep(nanoseconds: 150_000_000)
            }
+            // Timed out still not ready — if the user asked to record, cold-start.
+            guard let self else { return }
+            if self.recordWhenHostReady {
+                let shouldRecord = self.recordWhenHostReady
+                self.recordWhenHostReady = false
+                self.beginFlowStart(recordAfterHandoff: shouldRecord)
+            }
        }
    }

+    private func finishHostReadyWaitIfNeeded() {
+        coldStartDebouncer.reset()
+        guard recordWhenHostReady else { return }
+        recordWhenHostReady = false
+        guard state.micVoiceAvailability.isReady else { return }
+        startFlowRecording()
+        traceState("hostReadyWait.recordStarted")
+    }
+
    private func stopHostReadyWait() {
        hostReadyWaitTask?.cancel()
        hostReadyWaitTask = nil
@@ -338,9 +397,6 @@ final class KeyboardFlowCoordinator {
        recomputeMicVoiceAvailability()

        switch state.micVoiceAvailability {
-        case .ready:
-            detectAndStoreAppContext()
-            startFlowRecording()
        case .unavailable(.missingAPIKey):
            return
        case .unavailable(.noFullAccess):
@@ -348,18 +404,43 @@ final class KeyboardFlowCoordinator {
            state.phase = .error(.fullAccessRequired, message: msg)
            scheduleAutoClearError()
            recomputeMicVoiceAvailability()
+            return
        case .unavailable(.appGroupUnavailable):
            let msg = ExtL10n.string("keyboard.error.appGroupCommunication")
            state.phase = .error(.appGroupUnavailable, message: msg)
            scheduleAutoClearError()
            recomputeMicVoiceAvailability()
-        case .unavailable(.preparingSession):
+            return
+        default:
+            break
+        }
+
+        let withinReadyGrace = lastHostReadyAt > 0
+            && (Date().timeIntervalSince1970 - lastHostReadyAt) <= Self.hostReadyGrace
+        let action = FlowHandoffPolicy.micPressAction(
+            availability: state.micVoiceAvailability,
+            sessionActive: FlowSessionBridge.isSessionActive(),
+            hostReachable: FlowSessionBridge.isHostReachable(),
+            hostStale: FlowSessionBridge.isHostStale(),
+            withinReadyGrace: withinReadyGrace
+        )
+        switch action {
+        case .startRecording:
            detectAndStoreAppContext()
-            beginFlowStart()
-        case .unavailable(.hostNotReady):
+            startFlowRecording()
+        case .waitForHostReady(let recordWhenReady):
            detectAndStoreAppContext()
-            beginFlowStart()
-        case .recording, .processing:
+            recordWhenHostReady = recordWhenReady
+            coldStartDebouncer.reset()
+            startHostReadyWaitIfNeeded()
+            traceState(
+                "pressBegan.waitForHostReady",
+                extra: recordWhenReady ? "recordWhenReady=1" : "recordWhenReady=0"
+            )
+        case .openHostColdStart:
+            detectAndStoreAppContext()
+            beginFlowStart(recordAfterHandoff: true)
+        case .ignore:
            return
        }
    }
@@ -374,19 +455,23 @@ final class KeyboardFlowCoordinator {
        isFlowRecording = false
        stopUtteranceCountdown()
        ExtensionScreenWakeLock.release()
+        lastStoppedUtteranceId = currentUtteranceId
        writeCommand(.stopRecording)
        debug("pressEnded wrote stop command")
        state.phase = .processing
        state.lastTranscript = ExtL10n.string("keyboard.flow.transcribing")
-        recomputeMicVoiceAvailability()
        startFlowResultWatchdog()
+        recomputeMicVoiceAvailability()
    }

-    func beginFlowStart() {
+    func beginFlowStart(recordAfterHandoff: Bool = false) {
        guard !isPendingFlowStart else {
            traceState("beginFlowStart.ignored", extra: "reason=pendingAlreadyTrue")
            return
        }
+        self.recordAfterHandoff = recordAfterHandoff
+        recordWhenHostReady = false
+        coldStartDebouncer.reset()
        isPendingFlowStart = true
        isFlowRecording = false
        flowStartDeadline = Date().timeIntervalSince1970 + FlowWatchdog.startTimeout
@@ -394,7 +479,10 @@ final class KeyboardFlowCoordinator {
        recomputeMicVoiceAvailability()
        openHostApp("startflow")
        startFlowStartWatchdog()
-        traceState("beginFlowStart.started")
+        traceState(
+            "beginFlowStart.started",
+            extra: recordAfterHandoff ? "recordAfterHandoff=1" : "recordAfterHandoff=0"
+        )
    }

    func handleHostAppOpenResult(path: String, success: Bool) {
@@ -406,6 +494,7 @@ final class KeyboardFlowCoordinator {
        // guide the user to open OSGKeyboard manually.
        if path == "startflow", isPendingFlowStart {
            isPendingFlowStart = false
+            recordAfterHandoff = false
            flowStartDeadline = 0
            stopFlowWatchdog()
            traceState("openHostApp.failed", extra: "path=startflow cancelPending=1")
@@ -425,8 +514,10 @@ final class KeyboardFlowCoordinator {
                ExtensionScreenWakeLock.release()
            }
            currentUtteranceId = nil
+            lastStoppedUtteranceId = nil
            isFlowRecording = false
            isPendingFlowStart = false
+            recordAfterHandoff = false
            stopUtteranceCountdown()
            stopFlowWatchdog()
            state.level = 0
@@ -472,6 +563,7 @@ final class KeyboardFlowCoordinator {
                )
                FlowSessionBridge.clearResult()
                lastConsumedUtteranceId = result.utteranceId
+                lastStoppedUtteranceId = nil
                currentUtteranceId = nil
                textInserter.handleFlowTranscript(
                    TranscriptionDelivery(text: text, polishWarning: result.warning)
@@ -483,6 +575,7 @@ final class KeyboardFlowCoordinator {
                stopFlowWatchdog()
                FlowSessionBridge.clearResult()
                lastConsumedUtteranceId = result.utteranceId
+                lastStoppedUtteranceId = nil
                currentUtteranceId = nil
                let error = FlowTranscriptionError(
                    message: result.text ?? ExtL10n.string("keyboard.flow.resultTimeout"),
@@ -528,6 +621,7 @@ final class KeyboardFlowCoordinator {
            ExtensionScreenWakeLock.release()
            writeCommand(.abort)
            currentUtteranceId = nil
+            lastStoppedUtteranceId = nil
            stopFlowWatchdog()
            state.level = 0
            state.phase = .idle
@@ -547,10 +641,12 @@ final class KeyboardFlowCoordinator {
        isAwaitingFlowResult = false
        isFlowRecording = false
        isPendingFlowStart = false
+        recordAfterHandoff = false
        stopUtteranceCountdown()
        ExtensionScreenWakeLock.release()
        writeCommand(.abort)
        currentUtteranceId = nil
+        lastStoppedUtteranceId = nil
        stopFlowWatchdog()
        state.level = 0
        let message = ExtL10n.string("keyboard.flow.hostDisconnected")
@@ -585,9 +681,29 @@ final class KeyboardFlowCoordinator {

    private func startFlowRecording() {
        recomputeMicVoiceAvailability()
-        guard state.micVoiceAvailability.isReady else {
-            traceState("startFlowRecording.blocked", extra: "availability=\(String(describing: state.micVoiceAvailability))")
-            beginFlowStart()
+        let withinReadyGrace = lastHostReadyAt > 0
+            && (Date().timeIntervalSince1970 - lastHostReadyAt) <= Self.hostReadyGrace
+        if !state.micVoiceAvailability.isReady {
+            let action = FlowHandoffPolicy.micPressAction(
+                availability: state.micVoiceAvailability,
+                sessionActive: FlowSessionBridge.isSessionActive(),
+                hostReachable: FlowSessionBridge.isHostReachable(),
+                hostStale: FlowSessionBridge.isHostStale(),
+                withinReadyGrace: withinReadyGrace
+            )
+            traceState(
+                "startFlowRecording.blocked",
+                extra: "availability=\(String(describing: state.micVoiceAvailability)) action=\(action)"
+            )
+            switch action {
+            case .waitForHostReady(let recordWhenReady):
+                recordWhenHostReady = recordWhenReady
+                startHostReadyWaitIfNeeded()
+            case .openHostColdStart:
+                beginFlowStart(recordAfterHandoff: true)
+            case .startRecording, .ignore:
+                break
+            }
            return
        }
        isPendingFlowStart = false
@@ -596,11 +712,23 @@ final class KeyboardFlowCoordinator {

        guard let sessionId = FlowSessionBridge.readySnapshot()?.sessionId else {
            traceState("startFlowRecording.blocked", extra: "reason=missingSessionIdInReadySnapshot")
-            beginFlowStart()
+            // Snapshot lag with a live session → wait; only cold-start if host is dead.
+            if FlowHandoffPolicy.shouldOpenHostColdStart(
+                sessionActive: FlowSessionBridge.isSessionActive(),
+                hostReachable: FlowSessionBridge.isHostReachable(),
+                hostStale: FlowSessionBridge.isHostStale(),
+                withinReadyGrace: withinReadyGrace
+            ) {
+                beginFlowStart(recordAfterHandoff: true)
+            } else {
+                recordWhenHostReady = true
+                startHostReadyWaitIfNeeded()
+            }
            return
        }
        activeSessionId = sessionId
        currentUtteranceId = UUID()
+        lastStoppedUtteranceId = nil
        writeCommand(.startRecording)
        isFlowRecording = true
        state.lastTranscript = ""
@@ -640,8 +768,12 @@ final class KeyboardFlowCoordinator {

    private func cancelPendingFlowStart() {
        isPendingFlowStart = false
+        recordAfterHandoff = false
+        recordWhenHostReady = false
        flowStartDeadline = 0
+        coldStartDebouncer.reset()
        stopFlowWatchdog()
+        stopHostReadyWait()
        state.phase = .idle
        state.lastTranscript = ""
        recomputeMicVoiceAvailability()
@@ -660,6 +792,7 @@ final class KeyboardFlowCoordinator {
                let now = Date().timeIntervalSince1970
                if self.flowStartDeadline > 0, now > self.flowStartDeadline {
                    self.isPendingFlowStart = false
+                    self.recordAfterHandoff = false
                    self.flowStartDeadline = 0
                    self.traceState("startWatchdog.timeout")
                    self.showManualOpenHint(path: "startflow")
@@ -671,13 +804,20 @@ final class KeyboardFlowCoordinator {
    }

    private func completeFlowStartHandoff() {
+        let shouldRecord = recordAfterHandoff
        isPendingFlowStart = false
+        recordAfterHandoff = false
        flowStartDeadline = 0
        stopFlowWatchdog()
        state.lastTranscript = ""
        refreshSessionState()
-        startFlowRecording()
-        traceState("completeFlowStartHandoff.done")
+        if shouldRecord {
+            startFlowRecording()
+            traceState("completeFlowStartHandoff.done", extra: "record=1")
+        } else {
+            recomputeMicVoiceAvailability()
+            traceState("completeFlowStartHandoff.done", extra: "record=0 warmOnly")
+        }
    }

    private func startFlowLevelWatchdog() {
@@ -735,6 +875,7 @@ final class KeyboardFlowCoordinator {
                    )
                    FlowSessionBridge.clearResult()
                    self.lastConsumedUtteranceId = result.utteranceId
+                    self.lastStoppedUtteranceId = nil
                    self.currentUtteranceId = nil
                    self.debug("resultWatchdog consumed delivery len=\(text.count)")
                    self.textInserter.handleFlowTranscript(
@@ -747,6 +888,7 @@ final class KeyboardFlowCoordinator {
                    self.stopFlowWatchdog()
                    FlowSessionBridge.clearResult()
                    self.lastConsumedUtteranceId = result.utteranceId
+                    self.lastStoppedUtteranceId = nil
                    self.currentUtteranceId = nil
                    let error = FlowTranscriptionError(
                        message: result.text ?? ExtL10n.string("keyboard.flow.resultTimeout"),
@@ -784,6 +926,7 @@ final class KeyboardFlowCoordinator {
                    self.isAwaitingFlowResult = false
                    self.stopFlowWatchdog()
                    self.currentUtteranceId = nil
+                    self.lastStoppedUtteranceId = nil
                    self.debug("resultWatchdog TIMEOUT after \(Int(resultTimeout))s — no result from host")
                    let msg = ExtL10n.string("keyboard.flow.resultTimeout")
                    self.state.phase = .error(.flowResultTimeout, message: msg)
@@ -12,7 +12,7 @@
 //   │              (transcript preview)         │
 //   │              ┊                          │
 //   │              ◯ mic (centred)              │  ← action cluster:
-//   │   [delete]  [  space  ]  [return]         │     mic + bottom row
+//   │   [delete]  [ return ]  [space]           │     mic + bottom row
 //   │              ┊                          │
 //   └───────────────────────────────────────────┘

@@ -191,7 +191,7 @@ public struct KeyboardRootView: View {

    // MARK: - Action cluster

-    /// Mic centred above a bottom row: delete · space · return (or swapped).
+    /// Mic centred above a bottom row: delete · smart return · space (or swapped).
    /// The side cursor-drag pads are SwiftUI layout wrappers around UIKit
    /// pan recognizers, avoiding SwiftUI gesture delivery issues in
    /// keyboard extensions.
@@ -225,13 +225,13 @@ public struct KeyboardRootView: View {

            HStack(spacing: KeyboardLayoutMetrics.bottomActionSpacing) {
                if swapKeys {
-                    bottomReturnButton(disabled: editingBlocked)
                    bottomSpaceButton(disabled: editingBlocked)
+                    bottomReturnButton(disabled: editingBlocked)
                    bottomDeleteButton(disabled: editingBlocked)
                } else {
                    bottomDeleteButton(disabled: editingBlocked)
-                    bottomSpaceButton(disabled: editingBlocked)
                    bottomReturnButton(disabled: editingBlocked)
+                    bottomSpaceButton(disabled: editingBlocked)
                }
            }
            .opacity(dragging ? 0 : 1)
@@ -265,19 +265,20 @@ public struct KeyboardRootView: View {
        RectangularToolbarButton(spaceStyle: true, label: "space", disabled: disabled) {
            state.insertSpace()
        }
-        .frame(height: KeyboardLayoutMetrics.bottomActionRowHeight)
-    }
-
-    private func bottomReturnButton(disabled: Bool) -> some View {
-        RectangularToolbarButton(systemName: "return", label: "newline", disabled: disabled) {
-            state.insertNewline()
-        }
        .frame(
            width: KeyboardLayoutMetrics.bottomActionFixedWidth,
            height: KeyboardLayoutMetrics.bottomActionRowHeight
        )
    }

+    private func bottomReturnButton(disabled: Bool) -> some View {
+        let title = ExtL10n.string(state.returnKeyRole.titleKey)
+        return RectangularToolbarButton(title: title, label: title, disabled: disabled) {
+            state.insertNewline()
+        }
+        .frame(height: KeyboardLayoutMetrics.bottomActionRowHeight)
+    }
+
    /// Option C: block typing keys during the full voice-input pipeline.
    private var voiceInputBlocksEditing: Bool {
        switch state.phase {
@@ -10,6 +10,7 @@ import OSGKeyboardShared

 private enum ToolbarButtonMetrics {
    static let iconSize: CGFloat = 14
+    static let titleSize: CGFloat = 16
    static let cornerRadius: CGFloat = 12
    static let spaceBarCapsuleWidth: CGFloat = 31
    static let pressScale: CGFloat = 0.94
@@ -138,6 +139,7 @@ struct RectangularToolbarButton: View {

    let systemName: String?
    let spaceStyle: Bool
+    let title: String?
    let label: String
    let disabled: Bool
    let action: () -> Void
@@ -145,14 +147,25 @@ struct RectangularToolbarButton: View {
    init(systemName: String, label: String, disabled: Bool = false, action: @escaping () -> Void) {
        self.systemName = systemName
        self.spaceStyle = false
+        self.title = nil
        self.label = label
        self.disabled = disabled
        self.action = action
    }

+    init(title: String, label: String, disabled: Bool = false, action: @escaping () -> Void) {
+        self.systemName = nil
+        self.spaceStyle = false
+        self.label = label
+        self.disabled = disabled
+        self.action = action
+        self.title = title
+    }
+
    init(spaceStyle: Bool, label: String, disabled: Bool = false, action: @escaping () -> Void) {
        self.systemName = nil
        self.spaceStyle = spaceStyle
+        self.title = nil
        self.label = label
        self.disabled = disabled
        self.action = action
@@ -170,6 +183,10 @@ struct RectangularToolbarButton: View {
                Image(systemName: systemName)
                    .font(.system(size: ToolbarButtonMetrics.iconSize, weight: .semibold))
                    .foregroundStyle(palette.textPrimary)
+            } else if let title {
+                Text(title)
+                    .font(.system(size: ToolbarButtonMetrics.titleSize, weight: .semibold))
+                    .foregroundStyle(palette.textPrimary)
            }
        }
        .contentShape(Rectangle())
@@ -25,6 +25,7 @@
 "common.delete" = "Delete";
 "common.space" = "Space";
 "common.newline" = "Return";
+"common.send" = "Send";

 /* Privacy footnote (onboarding welcome page) */
 "privacy.audio.title" = "On-device transcription";
@@ -69,13 +70,13 @@
 "settings.reset.title" = "Reset all settings?";
 "settings.reset.message" = "API key, model, and base URL will be cleared.";
 "settings.reset.confirm" = "Reset all settings";
-"settings.engine.title" = "Recognition method";
-"settings.engine.subtitle" = "Pick the recognition engine. Local engine does transcription only, no API key needed.";
-"settings.engine.local.title" = "On-device recognition";
+"settings.engine.title" = "Speech transcription method";
+"settings.engine.subtitle" = "Local: built-in system transcription. Cloud: ASR transcription with optional LLM polish.";
+"settings.engine.local.title" = "On-device transcription";
 "settings.engine.local.ios26" = "Always on-device, no network.";
-"settings.engine.local.legacy" = "On-device ASR. Transcription only, no polish.";
-"settings.engine.cloud.title" = "Cloud recognition & polish";
-"settings.engine.cloud.subtitle" = "ASR + LLM polish. API key required.";
+"settings.engine.local.legacy" = "Transcribe with the built-in system speech service";
+"settings.engine.cloud.title" = "Cloud transcription";
+"settings.engine.cloud.subtitle" = "Cloud ASR transcription, with optional LLM polish";
 "settings.provider.title" = "Provider";
 "settings.provider.subtitle" = "Pick the LLM that polishes your dictation.";
 "settings.api.title" = "API";
@@ -24,7 +24,8 @@
 "common.cancel" = "取消";
 "common.delete" = "删除";
 "common.space" = "空格";
-"common.newline" = "换行";
+"common.newline" = "回车";
+"common.send" = "发送";

 /* Privacy footnote (onboarding welcome page) */
 "privacy.audio.title" = "支持本地转写";
@@ -69,13 +70,13 @@
 "settings.reset.title" = "重置所有设置?";
 "settings.reset.message" = "API key、model 和 base URL 都会被清空。";
 "settings.reset.confirm" = "重置所有设置";
-"settings.engine.title" = "识别方式";
-"settings.engine.subtitle" = "选择识别方式。本地引擎无需 API Key,仅做语音转录。";
-"settings.engine.local.title" = "本地识别";
+"settings.engine.title" = "语音转写方式";
+"settings.engine.subtitle" = "本地：系统内置转写；云端：ASR 转写，可用 LLM 润色。";
+"settings.engine.local.title" = "本地转写";
 "settings.engine.local.ios26" = "始终端侧,无需联网。";
-"settings.engine.local.legacy" = "端侧 ASR,仅转录,无润色。";
-"settings.engine.cloud.title" = "云端识别与润色";
-"settings.engine.cloud.subtitle" = "ASR 转录 + LLM 润色,需要 API Key。";
+"settings.engine.local.legacy" = "通过系统内置服务语音转写";
+"settings.engine.cloud.title" = "云端转写";
+"settings.engine.cloud.subtitle" = "通过云端 ASR 转写，可用 LLM 润色";
 "settings.provider.title" = "云端引擎";
 "settings.provider.subtitle" = "选择 LLM 提供商。";
 "settings.api.title" = "接口";