dsh-live-voice 0.7.1 → 0.7.2

package/lib/asr/asr.worker.js

@@ -36682,7 +36682,7 @@ async function createQwenRuntime(options) {
 }
 
 // src/modules/recognition/local/assets.ts
-var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.7.1" : "test") + "/";
+var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.7.2" : "test") + "/";
 
 // src/modules/recognition/local/modelCatalog.ts
 var browserAsrModels = [

package/lib/asr/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.7.1",
+  "version": "0.7.2",
   "assets": [
     "asr.worker.js",
     "manifest.json",

package/lib/client.js

@@ -152,7 +152,7 @@ var SessionSpeechMemory = class {
 };
 
 // src/modules/recognition/vad/assets.ts
-var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.7.1" : "test") + "/";
+var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.7.2" : "test") + "/";
 async function sileroVadAvailable(fetchImpl = globalThis.fetch) {
   try {
     const response = await fetchImpl(sileroVadAssetBase() + "availability", {
@@ -3507,7 +3507,7 @@ async function removeBrowserModel(key, storage = caches) {
 }
 
 // src/modules/recognition/local/assets.ts
-var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.7.1" : "test") + "/";
+var browserAsrAssetBase = () => "/api/dsh-live-voice/asr/" + encodeURIComponent(true ? "0.7.2" : "test") + "/";
 
 // src/modules/recognition/local/runtimeIdentity.ts
 var workerProtocol = "asr-worker-v3-ort-1.31.0-dev.20260914-8d85527a0";
@@ -10662,9 +10662,9 @@ var VoiceCoordinator = class _VoiceCoordinator {
       return await this.recognition.suspendCapture() !== false;
     });
   }
-  async _resumeInputCapture() {
+  async _resumeInputCapture(valid = () => true) {
     return this._input(async () => {
-      if (!this.snapshot.listening || typeof this.recognition.resumeCapture !== "function")
+      if (!valid() || !this.snapshot.listening || typeof this.recognition.resumeCapture !== "function")
         return false;
       return await this.recognition.resumeCapture() !== false;
     });
@@ -11458,22 +11458,53 @@ var VoiceCoordinator = class _VoiceCoordinator {
   getSpeechProgress() {
     return this.engines[this.snapshot.speechEngine]?.getPlaybackProgress?.() ?? null;
   }
+  async _restorePausedSpeechInput(epoch, control) {
+    const valid = () => !this.disposed && !this.navigationPaused && epoch === this.speechEpoch && control === this.controlEpoch && this.snapshot.paused && this.snapshot.settings.mode === "speaker";
+    if (!valid()) return;
+    try {
+      if (this._hasSuspendedInputCapture()) {
+        await this._resumeInputCapture(valid);
+      } else if (this.snapshot.conversation && !this.snapshot.listening && !this.snapshot.starting && !this.snapshot.inputRecovery) {
+        await this._startInput(true);
+      }
+    } catch (error) {
+      if (valid()) this.patch({ inputError: message(error), error: message(error) });
+    }
+  }
+  async _gateResumedSpeechInput() {
+    if (this.snapshot.settings.mode !== "speaker") return;
+    const suspended = await this._suspendInputCapture();
+    if (!suspended) {
+      await this.stopListening();
+      if (this.inputReleaseError) throw this.inputReleaseError;
+    }
+  }
   async pauseSpeech(automatic = false) {
     if (this.snapshot.speechRecovery?.retrying) this._cancelRecovery("speech", true);
     if (!automatic) {
       this._cancelInterruptionResume();
       this.interruptionPausedSpeech = false;
     }
-    if (this.disposed || this.snapshot.paused) return;
+    if (this.disposed) return;
+    if (this.snapshot.paused) {
+      const epoch2 = this.speechEpoch;
+      const control2 = ++this.controlEpoch;
+      await this.inputTail;
+      await this._restorePausedSpeechInput(epoch2, control2);
+      return;
+    }
     if (!this.snapshot.speaking && this.snapshot.speechRunActive) {
       this._cancelAssistantSpeechTimer();
+      const control2 = ++this.controlEpoch;
       this.patch({ paused: true });
+      await this._restorePausedSpeechInput(this.speechEpoch, control2);
       return;
     }
     if (!this.snapshot.speaking) return;
     if (this.speechLaunchPending) {
-      ++this.controlEpoch;
+      const control2 = ++this.controlEpoch;
       this.patch({ paused: true });
+      await this._restorePausedSpeechInput(this.speechEpoch, control2);
       return;
     }
     const epoch = this.speechEpoch;
@@ -11484,6 +11515,7 @@ var VoiceCoordinator = class _VoiceCoordinator {
       if (epoch === this.speechEpoch && control === this.controlEpoch && this.snapshot.speaking && result !== false) {
         this.patch({ paused: true });
         this.activeRecordingTimeline?.("assistant-pause");
+        await this._restorePausedSpeechInput(epoch, control);
       }
     } catch (error) {
       if (epoch === this.speechEpoch) this.patch({ error: message(error) });
@@ -11503,12 +11535,11 @@ var VoiceCoordinator = class _VoiceCoordinator {
     if (this.speechLaunchPending) {
       const epoch2 = this.speechEpoch;
       const control2 = ++this.controlEpoch;
-      if (this.snapshot.settings.mode === "speaker") {
-        await this.stopListening();
-        if (this.inputReleaseError) {
-          this.patch({ error: message(this.inputReleaseError) });
-          return;
-        }
+      try {
+        await this._gateResumedSpeechInput();
+      } catch (error) {
+        if (epoch2 === this.speechEpoch) this.patch({ error: message(error) });
+        return;
       }
       if (epoch2 === this.speechEpoch && control2 === this.controlEpoch)
         this.patch({ paused: false });
@@ -11517,19 +11548,19 @@ var VoiceCoordinator = class _VoiceCoordinator {
     const epoch = this.speechEpoch;
     const control = ++this.controlEpoch;
     try {
-      if (this.snapshot.settings.mode === "speaker" && !this._hasSuspendedInputCapture()) {
-        await this.stopListening();
-        if (this.inputReleaseError) throw this.inputReleaseError;
-      }
+      await this._gateResumedSpeechInput();
       if (epoch !== this.speechEpoch || control !== this.controlEpoch) return;
       const result = await this.engines[this.snapshot.speechEngine || this.snapshot.settings.engine].resume();
       if (epoch === this.speechEpoch && control === this.controlEpoch && result !== false) {
         this.patch({ paused: false });
         this.activeRecordingTimeline?.("assistant-resume");
         this._prefetchSpeech();
+      } else {
+        await this._restorePausedSpeechInput(epoch, control);
       }
     } catch (error) {
       if (epoch === this.speechEpoch) this.patch({ error: message(error) });
+      await this._restorePausedSpeechInput(epoch, control);
     }
   }
   async speak(text, messageId = null) {
@@ -11624,6 +11655,7 @@ var VoiceCoordinator = class _VoiceCoordinator {
         if (gating) timeline?.("input-stop-complete");
       }
       if (epoch !== this.speechEpoch || this.disposed) return;
+      if (this.snapshot.paused) await this._restorePausedSpeechInput(epoch, this.controlEpoch);
       const queued = item;
       if (queued?.preparing) await queued.preparing;
       if (epoch !== this.speechEpoch || this.disposed) return;
@@ -14769,7 +14801,7 @@ function createDeviceOptions(devices, kind, systemDefault, fallback) {
 var import_react39 = __toESM(require("react"), 1);
 
 // src/modules/settings/services/releases.ts
-var CURRENT_VERSION = "0.7.1";
+var CURRENT_VERSION = "0.7.2";
 var TESTED_DSH_VERSION = "0.2.2-rc.1";
 var REPOSITORY_URL = "https://github.com/victorwads/dsh-live-voice";
 var RELEASES_URL = `${REPOSITORY_URL}/releases`;
@@ -18081,7 +18113,7 @@ function apply(ctx) {
       entry.chatListeners.add(listener);
       return () => entry.chatListeners.delete(listener);
     };
-    entry.readChat = entry.chat.getSnapshot.bind(entry.chat);
+    entry.readChat = () => entry.chat.getSnapshot();
     const refresh = (reset = false) => {
       if (disposed || entry.closed) return;
       const snapshot = entry.readChat();
@@ -18109,10 +18141,20 @@ function apply(ctx) {
     };
     entry.refreshSpeech = refresh;
     refresh();
-    entry.unsubscribe = entry.chat.subscribe(() => {
+    const publishChat = () => {
       refresh();
       for (const listener of entry.chatListeners) listener();
-    });
+    };
+    entry.unsubscribe = entry.chat.subscribe(publishChat);
+    entry.reconnectChat = () => {
+      if (disposed || entry.closed) return;
+      const chat = ctx.uiConversation.binding(sessionId).target("chat");
+      if (chat === entry.chat) return;
+      entry.unsubscribe?.();
+      entry.chat = chat;
+      entry.unsubscribe = chat.subscribe(publishChat);
+      publishChat();
+    };
     const pendingInteractions = ctx.uiSession.pendingInteractions;
     const refreshPendingQuestion = (baseline = false) => {
       if (disposed || entry.closed || !pendingInteractions) return;
@@ -18289,6 +18331,7 @@ function apply(ctx) {
     import_react75.default.useLayoutEffect(() => {
       if (disposed) return;
       const current = get(sessionId);
+      current.reconnectChat();
       current.refs++;
       if (kind === "buttons") current.buttons++;
       setEntry(current);

package/lib/server.js

@@ -2052,7 +2052,7 @@ var VAD_ASSET_TYPES = {
 };
 var MAX_VAD_ASSET_BYTES = 32 * 1024 * 1024;
 function registerVadAssetRoutes(ctx, options = {}) {
-  const version = options.version ?? (true ? "0.7.1" : "test");
+  const version = options.version ?? (true ? "0.7.2" : "test");
   if (!/^[a-zA-Z0-9][a-zA-Z0-9.+-]{0,127}$/.test(version))
     throw new Error("Invalid VAD asset version");
   const directory = options.directory ?? fileURLToPath(new URL("./vad/", import.meta.url));
@@ -2146,7 +2146,7 @@ import { lstat as lstat2, readFile as readFile5 } from "node:fs/promises";
 import { fileURLToPath as fileURLToPath2 } from "node:url";
 import { join as join8 } from "node:path";
 function registerBrowserAsrAssets(ctx, options = {}) {
-  const version = options.version ?? (true ? "0.7.1" : "test");
+  const version = options.version ?? (true ? "0.7.2" : "test");
   if (!/^[a-zA-Z0-9][a-zA-Z0-9.+-]{0,127}$/.test(version)) throw new Error("invalid-version");
   const base = "/api/dsh-live-voice/asr/" + encodeURIComponent(version) + "/";
   const directory = options.directory ?? fileURLToPath2(new URL("./asr/", import.meta.url));

package/lib/vad/manifest.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": 1,
-  "pluginVersion": "0.7.1",
+  "pluginVersion": "0.7.2",
   "dependencies": {
     "@ricky0123/vad-web": "0.0.31",
     "onnxruntime-web": "1.22.0"
@@ -41,7 +41,7 @@
     },
     "vad.worker.js": {
       "bytes": 79337,
-      "sha256": "36a94fd4b60d3717d25628a033574d1b6c2fb576a8be732c87255ba567c33bd0"
+      "sha256": "6b83547e105d6a61b37f3647687021afcab2162d707071e7bffd9a6ffecade58"
     }
   }
 }

package/lib/vad/vad.worker.js

@@ -1632,7 +1632,7 @@ var vo = Ke;
 Object.defineProperty(I.versions, "web", { value: $t, enumerable: true });
 
 // src/modules/recognition/vad/assets.ts
-var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.7.1" : "test") + "/";
+var sileroVadAssetBase = () => "/api/dsh-live-voice/vad/" + encodeURIComponent(true ? "0.7.2" : "test") + "/";
 
 // src/modules/recognition/vad/StreamingResampler.ts
 var StreamingResampler = class {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-live-voice",
-  "version": "0.7.1",
+  "version": "0.7.2",
   "dshTestedVersion": "0.2.2-rc.1",
   "type": "module",
   "scripts": {

package/src/app/ARCHITECTURE.md

@@ -3,6 +3,7 @@
 `src/app` is the composition boundary between DSH and the voice-domain modules. Client and server entry points remain separate; host-only Node dependencies must never enter the browser bundle.
 
 - `client/apply.tsx` injects DSH client services, creates session controllers and voice ownership, wires engines and settings, and disposes resources. `client/slotDefinitions.ts` and `client/registerSlots.tsx` own the established slot IDs, order, and registration. Preserve those contracts when changing placement. Microphone mode stop invalidates pending input ownership and application continuity without ending current playback; explicit takeover and full End retain their separate handoff/lifecycle semantics.
+- Retained session entries re-resolve the current DSH chat target on committed slot attachment. A replaced target releases its old subscription and republishes through stable readers/listeners without resetting session speech memory; detached-session admission still prevents background playback.
 - Composer detachment parks eligible recording-backed microphone controllers in the existing session registry rather than retiring them. Their capture clock pauses, recognition requests continue, and final results stay on the originating controller. Returning uses the existing ownership barrier before resuming capture and delivering queued results against the latest draft. Explicit stop clears parked intent; plugin disposal retires both mounted and parked entries. This memory is page-lifetime only, not reload recovery.
 - `server/apply.ts` composes host engines and providers. `server/registerRoutes.ts` registers authenticated same-origin routes; do not bypass their validation or expose host access directly to browser code.
 - `client/transcriptionLeaveGuard.ts` owns one activity-scoped native `beforeunload` listener across microphone, meeting sources and Settings test capture. It covers acquisition, listening and unsettled transcription, not playback or voice-mode preference alone. Cancelling the browser dialog does not stop capture; existing `pagehide` cleanup runs only after leaving. Idle and disposed apps remove the listener. Browser wording, user-activation requirements and mobile/forced-close limitations apply; this is not refresh recovery.

package/src/app/client/apply.tsx

@@ -690,7 +690,8 @@ export function apply(ctx) {
       entry.chatListeners.add(listener);
       return () => entry.chatListeners.delete(listener);
     };
-    entry.readChat = entry.chat.getSnapshot.bind(entry.chat);
+    // Stable React store readers must follow a replacement DSH session binding.
+    entry.readChat = () => entry.chat.getSnapshot();
     const refresh = (reset = false) => {
       if (disposed || entry.closed) return;
       const snapshot = entry.readChat();
@@ -729,10 +730,22 @@ export function apply(ctx) {
     };
     entry.refreshSpeech = refresh;
     refresh();
-    entry.unsubscribe = entry.chat.subscribe(() => {
+    const publishChat = () => {
       refresh();
       for (const listener of entry.chatListeners) listener();
-    });
+    };
+    entry.unsubscribe = entry.chat.subscribe(publishChat);
+    entry.reconnectChat = () => {
+      if (disposed || entry.closed) return;
+      const chat = ctx.uiConversation.binding(sessionId).target('chat');
+      if (chat === entry.chat) return;
+      entry.unsubscribe?.();
+      entry.chat = chat;
+      entry.unsubscribe = chat.subscribe(publishChat);
+      // Subscribe first: activation may materialize the target synchronously.
+      // Existing speech memory and detached-session admission rules still apply.
+      publishChat();
+    };
     // DSH 0.1.6 may omit this legacy store. Keep the remaining voice controls
     // available while structured-question integration is unavailable.
     const pendingInteractions = ctx.uiSession.pendingInteractions;
@@ -927,6 +940,7 @@ export function apply(ctx) {
       // Allocation and subscriptions happen only for committed mounts. Suspended
       // or abandoned renders never acquire a session, probe engines, or retain it.
       const current = get(sessionId);
+      current.reconnectChat();
       current.refs++;
       if (kind === 'buttons') current.buttons++;
       setEntry(current);

package/src/modules/core/coordinator.ts

@@ -408,9 +408,13 @@ export class VoiceCoordinator {
       return (await this.recognition.suspendCapture()) !== false;
     });
   }
-  async _resumeInputCapture() {
+  async _resumeInputCapture(valid = () => true) {
     return this._input(async () => {
-      if (!this.snapshot.listening || typeof this.recognition.resumeCapture !== 'function')
+      if (
+        !valid() ||
+        !this.snapshot.listening ||
+        typeof this.recognition.resumeCapture !== 'function'
+      )
         return false;
       return (await this.recognition.resumeCapture()) !== false;
     });
@@ -1370,23 +1374,67 @@ export class VoiceCoordinator {
   getSpeechProgress() {
     return this.engines[this.snapshot.speechEngine]?.getPlaybackProgress?.() ?? null;
   }
+  async _restorePausedSpeechInput(epoch, control) {
+    const valid = () =>
+      !this.disposed &&
+      !this.navigationPaused &&
+      epoch === this.speechEpoch &&
+      control === this.controlEpoch &&
+      this.snapshot.paused &&
+      this.snapshot.settings.mode === 'speaker';
+    if (!valid()) return;
+    try {
+      if (this._hasSuspendedInputCapture()) {
+        await this._resumeInputCapture(valid);
+      } else if (
+        this.snapshot.conversation &&
+        !this.snapshot.listening &&
+        !this.snapshot.starting &&
+        !this.snapshot.inputRecovery
+      ) {
+        await this._startInput(true);
+      }
+    } catch (error) {
+      if (valid()) this.patch({ inputError: message(error), error: message(error) });
+    }
+  }
+  async _gateResumedSpeechInput() {
+    if (this.snapshot.settings.mode !== 'speaker') return;
+    // Preserve recording-backed capture; native engines retain safe stop/restart.
+    const suspended = await this._suspendInputCapture();
+    if (!suspended) {
+      await this.stopListening();
+      if (this.inputReleaseError) throw this.inputReleaseError;
+    }
+  }
   async pauseSpeech(automatic = false) {
     if (this.snapshot.speechRecovery?.retrying) this._cancelRecovery('speech', true);
     if (!automatic) {
       this._cancelInterruptionResume();
       this.interruptionPausedSpeech = false;
     }
-    if (this.disposed || this.snapshot.paused) return;
+    if (this.disposed) return;
+    if (this.snapshot.paused) {
+      // Reaffirming pause invalidates a resume whose capture gate is still pending.
+      const epoch = this.speechEpoch;
+      const control = ++this.controlEpoch;
+      await this.inputTail;
+      await this._restorePausedSpeechInput(epoch, control);
+      return;
+    }
     if (!this.snapshot.speaking && this.snapshot.speechRunActive) {
       this._cancelAssistantSpeechTimer();
+      const control = ++this.controlEpoch;
       this.patch({ paused: true });
+      await this._restorePausedSpeechInput(this.speechEpoch, control);
       return;
     }
     if (!this.snapshot.speaking) return;
     if (this.speechLaunchPending) {
       // Preparation has no player to pause yet; retain an explicit launch gate.
-      ++this.controlEpoch;
+      const control = ++this.controlEpoch;
       this.patch({ paused: true });
+      await this._restorePausedSpeechInput(this.speechEpoch, control);
       return;
     }
     const epoch = this.speechEpoch;
@@ -1403,6 +1451,7 @@ export class VoiceCoordinator {
       ) {
         this.patch({ paused: true });
         this.activeRecordingTimeline?.('assistant-pause');
+        await this._restorePausedSpeechInput(epoch, control);
       }
     } catch (error) {
       if (epoch === this.speechEpoch) this.patch({ error: message(error) });
@@ -1422,12 +1471,11 @@ export class VoiceCoordinator {
     if (this.speechLaunchPending) {
       const epoch = this.speechEpoch;
       const control = ++this.controlEpoch;
-      if (this.snapshot.settings.mode === 'speaker') {
-        await this.stopListening();
-        if (this.inputReleaseError) {
-          this.patch({ error: message(this.inputReleaseError) });
-          return;
-        }
+      try {
+        await this._gateResumedSpeechInput();
+      } catch (error) {
+        if (epoch === this.speechEpoch) this.patch({ error: message(error) });
+        return;
       }
       if (epoch === this.speechEpoch && control === this.controlEpoch)
         this.patch({ paused: false });
@@ -1436,10 +1484,7 @@ export class VoiceCoordinator {
     const epoch = this.speechEpoch;
     const control = ++this.controlEpoch;
     try {
-      if (this.snapshot.settings.mode === 'speaker' && !this._hasSuspendedInputCapture()) {
-        await this.stopListening();
-        if (this.inputReleaseError) throw this.inputReleaseError;
-      }
+      await this._gateResumedSpeechInput();
       if (epoch !== this.speechEpoch || control !== this.controlEpoch) return;
       const result =
         await this.engines[this.snapshot.speechEngine || this.snapshot.settings.engine].resume();
@@ -1447,9 +1492,12 @@ export class VoiceCoordinator {
         this.patch({ paused: false });
         this.activeRecordingTimeline?.('assistant-resume');
         this._prefetchSpeech();
+      } else {
+        await this._restorePausedSpeechInput(epoch, control);
       }
     } catch (error) {
       if (epoch === this.speechEpoch) this.patch({ error: message(error) });
+      await this._restorePausedSpeechInput(epoch, control);
     }
   }
   async speak(text, messageId = null) {
@@ -1548,6 +1596,8 @@ export class VoiceCoordinator {
         if (gating) timeline?.('input-stop-complete');
       }
       if (epoch !== this.speechEpoch || this.disposed) return;
+      // A pause can arrive while the initial capture gate is still settling.
+      if (this.snapshot.paused) await this._restorePausedSpeechInput(epoch, this.controlEpoch);
       const queued = item;
       if (queued?.preparing) await queued.preparing;
       if (epoch !== this.speechEpoch || this.disposed) return;