feat: add Silero VAD toggle to audio pipeline

Integrates @ricky0123/vad-web's MicVAD as an optional voice activity detector
alongside the noise gate. When enabled, the Silero ONNX model classifies each
audio frame as speech or silence; silence frames mute the worklet's output via
a new VAD gate message. VAD is wired into Publisher.ts alongside the existing
noise gate transformer. Vite is configured to copy the worklet bundle, ONNX
model, and ORT WASM files to /vad/ so they're reachable at runtime.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/state/CallViewModel/localMember/Publisher.ts

@@ -41,11 +41,13 @@ import {
   transientSuppressorEnabled,
   transientThreshold,
   transientRelease,
+  vadEnabled,
 } from "../../../settings/settings.ts";
 import {
   type NoiseGateParams,
   NoiseGateTransformer,
 } from "../../../livekit/NoiseGateTransformer.ts";
+import { SileroVADGate } from "../../../livekit/SileroVADGate.ts";
 import { observeTrackReference$ } from "../../observeTrackReference";
 import { type Connection } from "../remoteMembers/Connection.ts";
 import { ObservableScope } from "../../ObservableScope.ts";
@@ -435,6 +437,7 @@ export class Publisher {
 
     let transformer: NoiseGateTransformer | null = null;
     let audioCtx: AudioContext | null = null;
+    let vadGate: SileroVADGate | null = null;
 
     const currentParams = (): NoiseGateParams => ({
       threshold: noiseGateThreshold.getValue(),
@@ -446,6 +449,32 @@ export class Publisher {
       transientReleaseMs: transientRelease.getValue(),
     });
 
+    const stopVAD = (): void => {
+      if (vadGate) {
+        void vadGate.destroy();
+        vadGate = null;
+      }
+      // Reset gate to open so audio flows if VAD is toggled off mid-call
+      transformer?.setVADOpen(true);
+    };
+
+    const startVAD = (track: LocalAudioTrack, ctx: AudioContext): void => {
+      stopVAD();
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const rawTrack: MediaStreamTrack | undefined = (track as any).mediaStreamTrack;
+      if (!rawTrack) {
+        this.logger.warn("[VAD] no underlying MediaStreamTrack — skipping VAD");
+        return;
+      }
+      const stream = new MediaStream([rawTrack]);
+      vadGate = new SileroVADGate(stream, ctx);
+      vadGate.onSpeechStart = (): void => transformer?.setVADOpen(true);
+      vadGate.onSpeechEnd = (): void => transformer?.setVADOpen(false);
+      vadGate.start().catch((e: unknown) => {
+        this.logger.error("[VAD] failed to start", e);
+      });
+    };
+
     // Attach / detach processor when enabled state or the track changes.
     combineLatest([audioTrack$, noiseGateEnabled.value$])
       .pipe(scope.bind())
@@ -459,18 +488,20 @@ export class Publisher {
           this.logger.info("[NoiseGate] AudioContext state before resume:", audioCtx.state);
           // eslint-disable-next-line @typescript-eslint/no-explicit-any
           (audioTrack as any).setAudioContext(audioCtx);
-          audioCtx.resume().then(() => {
+          audioCtx.resume().then(async () => {
             this.logger.info("[NoiseGate] AudioContext state after resume:", audioCtx?.state);
             return audioTrack
               // eslint-disable-next-line @typescript-eslint/no-explicit-any
               .setProcessor(transformer as any);
           }).then(() => {
             this.logger.info("[NoiseGate] setProcessor resolved");
+            if (vadEnabled.getValue() && audioCtx) startVAD(audioTrack, audioCtx);
           }).catch((e: unknown) => {
             this.logger.error("[NoiseGate] setProcessor failed", e);
           });
         } else if (!enabled && audioTrack.getProcessor()) {
           this.logger.info("[NoiseGate] removing processor");
+          stopVAD();
           void audioTrack.stopProcessor();
           void audioCtx?.close();
           audioCtx = null;
@@ -482,6 +513,18 @@ export class Publisher {
         }
       });
 
+    // Start/stop VAD when its toggle changes.
+    combineLatest([audioTrack$, vadEnabled.value$])
+      .pipe(scope.bind())
+      .subscribe(([audioTrack, enabled]) => {
+        if (!audioTrack || !audioCtx) return;
+        if (enabled) {
+          startVAD(audioTrack, audioCtx);
+        } else {
+          stopVAD();
+        }
+      });
+
     // Push param changes to the live worklet without recreating the processor.
     combineLatest([
       noiseGateThreshold.value$,