Add local VAD and conversation-driven NPC pause

MicrophoneCaptureComponent:
- Local Voice Activity Detection (RMS-based, independent of ElevenLabs)
- Configurable threshold, onset time, silence time
- bIsUserSpeaking flag + OnUserVoiceActivityChanged delegate
- Hysteresis prevents flickering between speech/silence

AIController gaze bridge:
- Resolve MicComponent from player Pawn (not NPC) via reflection
- ConversationPaused BB key blocks movement branches via BT decorator
- NPC stops only when user actually speaks (not just on proximity connect)
- NPC resumes when conversation disconnects
- Spline PauseFollowing/ResumeFollowing on conversation start/end

BT setup required:
- Add Blackboard Condition (ConversationPaused Is Not Set, Aborts=Both)
  on spline and patrol branches

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Unreal/PS_AI_Agent/Plugins/PS_AI_Behavior/Source/PS_AI_Behavior/Private/PS_AI_Behavior_AIController.cpp

@@ -100,6 +100,9 @@ void APS_AI_Behavior_AIController::OnUnPossess()
 	ClearGazeTarget();
 	CachedGazeComponent = nullptr;
 	CachedConvAgentComponent = nullptr;
+	CachedMicComponent = nullptr;
+	bConversationPaused = false;
+	bUserHasSpokenInConversation = false;
 	ProximityGazeTarget = nullptr;
 
 	PersonalityComp = nullptr;
@@ -199,6 +202,12 @@ void APS_AI_Behavior_AIController::SetupBlackboard()
 		PreferCoverEntry.EntryName = PS_AI_Behavior_BB::PreferCover;
 		PreferCoverEntry.KeyType = NewObject<UBlackboardKeyType_Bool>(BlackboardAsset);
 		BlackboardAsset->Keys.Add(PreferCoverEntry);
+
+		// ConversationPaused (bool: NPC paused during active conversation)
+		FBlackboardEntry ConvPausedEntry;
+		ConvPausedEntry.EntryName = PS_AI_Behavior_BB::ConversationPaused;
+		ConvPausedEntry.KeyType = NewObject<UBlackboardKeyType_Bool>(BlackboardAsset);
+		BlackboardAsset->Keys.Add(ConvPausedEntry);
 	}
 
 	UBlackboardComponent* RawBBComp = nullptr;
@@ -536,12 +545,29 @@ void APS_AI_Behavior_AIController::TryBindGazeComponent()
 		}
 	}
 
+	// Cache MicrophoneCaptureComponent for local VAD
+	// The mic lives on the PLAYER's Pawn, not the NPC — find it on the first player
+	static UClass* MicClass = nullptr;
+	if (!MicClass)
+	{
+		MicClass = LoadClass<UActorComponent>(nullptr,
+			TEXT("/Script/PS_AI_ConvAgent.PS_AI_ConvAgent_MicrophoneCaptureComponent"));
+	}
+	if (MicClass)
+	{
+		MicProp_bIsUserSpeaking = CastField<FBoolProperty>(
+			MicClass->FindPropertyByName(TEXT("bIsUserSpeaking")));
+		// Don't cache the component here — resolve dynamically in UpdateGazeTarget
+		// because the player Pawn may not exist yet at OnPossess time
+	}
+
 	UE_LOG(LogPS_AI_Behavior, Log,
-		TEXT("[%s] Gaze bridge bound: TargetActor=%s, BodyTracking=%s, Conversation=%s"),
+		TEXT("[%s] Gaze bridge bound: TargetActor=%s, BodyTracking=%s, Conversation=%s, VAD=%s"),
 		*GetName(),
 		GazeProp_TargetActor ? TEXT("OK") : TEXT("MISS"),
 		GazeProp_bEnableBodyTracking ? TEXT("OK") : TEXT("MISS"),
-		ConvProp_bNetIsConversing ? TEXT("OK") : TEXT("N/A"));
+		ConvProp_bNetIsConversing ? TEXT("OK") : TEXT("N/A"),
+		MicProp_bIsUserSpeaking ? TEXT("OK") : TEXT("N/A"));
 }
 
 void APS_AI_Behavior_AIController::SetGazeTarget(AActor* Target, bool bEnableBody)
@@ -628,13 +654,117 @@ void APS_AI_Behavior_AIController::UpdateGazeTarget(float DeltaSeconds)
 		}
 	}
 
-	// ── Priority 2: Conversation active → do NOT touch gaze ─────────
+	// ── Priority 2: Conversation active → manage movement pause via local VAD
-	// TODO: Add movement pause when local VAD detects user speech (not just connection)
-	if (IsConversationActiveOnPawn())
 	{
-		ProximityGazeTarget = nullptr;
+		const bool bConversing = IsConversationActiveOnPawn();
-		ProximityGazeDuration = 0.0f;
+
-		return;
+
+		// Check local VAD: is the user actually speaking?
+		// The mic is on the PLAYER's Pawn — find it dynamically
+		bool bUserSpeaking = false;
+		if (MicProp_bIsUserSpeaking)
+		{
+			// Try cached mic first
+			UActorComponent* MicComp = CachedMicComponent.Get();
+			if (!MicComp)
+			{
+				// Find mic on any player pawn
+				static UClass* MicClass = LoadClass<UActorComponent>(nullptr,
+					TEXT("/Script/PS_AI_ConvAgent.PS_AI_ConvAgent_MicrophoneCaptureComponent"));
+				if (MicClass)
+				{
+					for (FConstPlayerControllerIterator It = GetWorld()->GetPlayerControllerIterator(); It; ++It)
+					{
+						if (APlayerController* PC = It->Get())
+						{
+							if (APawn* PlayerPawn = PC->GetPawn())
+							{
+								MicComp = PlayerPawn->FindComponentByClass(MicClass);
+								if (MicComp)
+								{
+									CachedMicComponent = MicComp;
+									break;
+								}
+							}
+						}
+					}
+				}
+			}
+
+			if (MicComp)
+			{
+				bUserSpeaking = MicProp_bIsUserSpeaking->GetPropertyValue_InContainer(MicComp);
+			}
+		}
+
+		// Once user has spoken during this conversation, stay paused until conversation ends
+		if (bConversing && bUserSpeaking)
+		{
+			bUserHasSpokenInConversation = true;
+		}
+
+		if (bConversing && bUserHasSpokenInConversation)
+		{
+			// Set BB flag to block movement branches via decorator
+			if (!bConversationPaused)
+			{
+				bConversationPaused = true;
+
+				if (Blackboard)
+				{
+					Blackboard->SetValueAsBool(PS_AI_Behavior_BB::ConversationPaused, true);
+				}
+
+				StopMovement();
+
+				APawn* ConvPawn = GetPawn();
+				if (ConvPawn)
+				{
+					auto* Spline = ConvPawn->FindComponentByClass<UPS_AI_Behavior_SplineFollowerComponent>();
+					if (Spline)
+					{
+						Spline->PauseFollowing();
+					}
+				}
+
+				UE_LOG(LogPS_AI_Behavior, Log, TEXT("[%s] User speaking — conversation pause."), *GetName());
+			}
+
+			ProximityGazeTarget = nullptr;
+			ProximityGazeDuration = 0.0f;
+			return; // Gaze managed by ConvAgent
+		}
+
+		if (bConversing)
+		{
+			// Connected but user hasn't spoken yet — don't touch gaze, don't pause
+			ProximityGazeTarget = nullptr;
+			ProximityGazeDuration = 0.0f;
+			return;
+		}
+
+		// Conversation ended — clear BB flag, resume movement
+		if (bConversationPaused)
+		{
+			bConversationPaused = false;
+			bUserHasSpokenInConversation = false;
+
+			if (Blackboard)
+			{
+				Blackboard->SetValueAsBool(PS_AI_Behavior_BB::ConversationPaused, false);
+			}
+
+			APawn* ConvPawn = GetPawn();
+			if (ConvPawn)
+			{
+				if (auto* Spline = ConvPawn->FindComponentByClass<UPS_AI_Behavior_SplineFollowerComponent>())
+				{
+					Spline->ResumeFollowing();
+				}
+			}
+
+			UE_LOG(LogPS_AI_Behavior, Log, TEXT("[%s] Conversation ended — resumed."), *GetName());
+		}
 	}
 
 	// ── Priority 3/4: Proximity gaze ────────────────────────────────
@@ -750,3 +880,4 @@ void APS_AI_Behavior_AIController::UpdateGazeTarget(float DeltaSeconds)
 	}
 #endif
 }
+

Unreal/PS_AI_Agent/Plugins/PS_AI_Behavior/Source/PS_AI_Behavior/Public/PS_AI_Behavior_AIController.h

@@ -162,6 +162,14 @@ private:
 	TWeakObjectPtr<UActorComponent> CachedConvAgentComponent;
 	FBoolProperty* ConvProp_bNetIsConversing = nullptr;
 
+	// Mic VAD cache (local voice activity detection)
+	TWeakObjectPtr<UActorComponent> CachedMicComponent;
+	FBoolProperty* MicProp_bIsUserSpeaking = nullptr;
+
+	// Conversation pause state
+	bool bConversationPaused = false;
+	bool bUserHasSpokenInConversation = false;
+
 	// Proximity gaze state
 	TWeakObjectPtr<AActor> ProximityGazeTarget;
 	float ProximityGazeDuration = 0.0f;

Unreal/PS_AI_Agent/Plugins/PS_AI_Behavior/Source/PS_AI_Behavior/Public/PS_AI_Behavior_Definitions.h

@@ -188,4 +188,5 @@ namespace PS_AI_Behavior_BB
 	inline const FName LastKnownTargetPosition = TEXT("LastKnownTargetPosition");
 	inline const FName ThreatPawnName          = TEXT("ThreatPawnName"); // Debug: name of the owning Pawn behind ThreatActor
 	inline const FName PreferCover             = TEXT("PreferCover");    // Bool: personality-driven cover preference cycle
+	inline const FName ConversationPaused      = TEXT("ConversationPaused"); // Bool: NPC paused during active conversation
 }

Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Private/PS_AI_ConvAgent_MicrophoneCaptureComponent.cpp

@@ -198,6 +198,54 @@ void UPS_AI_ConvAgent_MicrophoneCaptureComponent::OnAudioGenerate(
 		PeakLevel.store(Peak, std::memory_order_relaxed);
 	}
 
+	// ── Local Voice Activity Detection ──────────────────────────────────
+	// Runs on the audio thread for minimal latency. Uses hysteresis to
+	// prevent flickering. Dispatches state change to game thread.
+	{
+		const float Rms = CurrentRMS.load(std::memory_order_relaxed);
+		const float BufferDuration = static_cast<float>(NumFrames) / FMath::Max(1, InSampleRate);
+		const bool bAboveThreshold = (Rms > VoiceActivityThreshold);
+
+		if (bAboveThreshold)
+		{
+			VoiceSilenceAccumulator.store(0.0f, std::memory_order_relaxed);
+			const float NewOnset = VoiceOnsetAccumulator.load(std::memory_order_relaxed) + BufferDuration;
+			VoiceOnsetAccumulator.store(NewOnset, std::memory_order_relaxed);
+
+			if (!bIsUserSpeaking && NewOnset >= VoiceOnsetTime)
+			{
+				bIsUserSpeaking = true;
+				TWeakObjectPtr<UPS_AI_ConvAgent_MicrophoneCaptureComponent> WeakSelf(this);
+				AsyncTask(ENamedThreads::GameThread, [WeakSelf]()
+				{
+					if (auto* Self = WeakSelf.Get())
+					{
+						Self->OnUserVoiceActivityChanged.Broadcast(true);
+					}
+				});
+			}
+		}
+		else
+		{
+			VoiceOnsetAccumulator.store(0.0f, std::memory_order_relaxed);
+			const float NewSilence = VoiceSilenceAccumulator.load(std::memory_order_relaxed) + BufferDuration;
+			VoiceSilenceAccumulator.store(NewSilence, std::memory_order_relaxed);
+
+			if (bIsUserSpeaking && NewSilence >= VoiceSilenceTime)
+			{
+				bIsUserSpeaking = false;
+				TWeakObjectPtr<UPS_AI_ConvAgent_MicrophoneCaptureComponent> WeakSelf(this);
+				AsyncTask(ENamedThreads::GameThread, [WeakSelf]()
+				{
+					if (auto* Self = WeakSelf.Get())
+					{
+						Self->OnUserVoiceActivityChanged.Broadcast(false);
+					}
+				});
+			}
+		}
+	}
+
 	// Resample + downmix to 16000 Hz mono.
 	TArray<float> Resampled = ResampleTo16000(FloatAudio, NumFrames, InNumChannels, InSampleRate);
 

Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Public/PS_AI_ConvAgent_MicrophoneCaptureComponent.h

@@ -11,6 +11,9 @@
 // Delivers captured float PCM samples (16000 Hz mono, resampled from device rate).
 DECLARE_MULTICAST_DELEGATE_OneParam(FOnPS_AI_ConvAgent_AudioCaptured, const TArray<float>& /*FloatPCM*/);
 
+// Fired when local voice activity changes (user starts/stops speaking).
+DECLARE_DYNAMIC_MULTICAST_DELEGATE_OneParam(FOnPS_AI_ConvAgent_VoiceActivityChanged, bool, bIsSpeaking);
+
 /**
  * Lightweight microphone capture component.
  * Captures from the default audio input device, resamples to 16000 Hz mono,
@@ -46,6 +49,31 @@ public:
 	 *  Set by the agent component for echo suppression (skip mic while agent speaks). */
 	std::atomic<bool>* EchoSuppressFlag = nullptr;
 
+	// ── Local Voice Activity Detection ──────────────────────────────────
+
+	/** RMS threshold above which audio is considered voice. Adjust based on mic sensitivity and environment noise. */
+	UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "PS AI ConvAgent|Voice Activity",
+		meta = (ClampMin = "0.001", ClampMax = "0.2"))
+	float VoiceActivityThreshold = 0.015f;
+
+	/** Time (seconds) RMS must stay above threshold before declaring speech onset. Prevents false triggers from clicks/bumps. */
+	UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "PS AI ConvAgent|Voice Activity",
+		meta = (ClampMin = "0.01", ClampMax = "1.0"))
+	float VoiceOnsetTime = 0.1f;
+
+	/** Time (seconds) RMS must stay below threshold before declaring speech ended. Prevents cutting off between words. */
+	UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "PS AI ConvAgent|Voice Activity",
+		meta = (ClampMin = "0.1", ClampMax = "3.0"))
+	float VoiceSilenceTime = 0.5f;
+
+	/** Whether the user is currently speaking (local VAD, independent of ElevenLabs server). */
+	UPROPERTY(BlueprintReadOnly, Category = "PS AI ConvAgent|Voice Activity")
+	bool bIsUserSpeaking = false;
+
+	/** Fired when voice activity changes. Use to pause NPC movement, trigger animations, etc. */
+	UPROPERTY(BlueprintAssignable, Category = "PS AI ConvAgent|Voice Activity")
+	FOnPS_AI_ConvAgent_VoiceActivityChanged OnUserVoiceActivityChanged;
+
 	// ── Debug ────────────────────────────────────────────────────────────────
 
 	/** Enable debug logging for this component.
@@ -99,6 +127,10 @@ private:
 	std::atomic<float> CurrentRMS{0.0f};
 	std::atomic<float> PeakLevel{0.0f};
 
+	// VAD accumulators (written from audio callback thread)
+	std::atomic<float> VoiceOnsetAccumulator{0.0f};
+	std::atomic<float> VoiceSilenceAccumulator{0.0f};
+
 	// Device name cached on StartCapture for HUD display.
 	FString CachedDeviceName;
 };