Fix multi-player regressions: audio/text drops, thread safety

1. StartConversationWithSelectedAgent: remove early return when WebSocket
   is already connected (persistent mode). Always call ServerJoinConversation
   so the pawn is added to NetConnectedPawns and bNetIsConversing is set.

2. ServerSendMicAudioFromPlayer: bypass speaker arbitration in standalone
   mode (<=1 connected pawn). Send audio directly to avoid silent drops
   caused by pawn not being in NetConnectedPawns array. Add warning logs
   for multi-player drops to aid debugging.

3. OnMicrophoneDataCaptured: restore direct WebSocketProxy->SendAudioChunk
   on the server path. This callback runs on the WASAPI audio thread —
   accessing game-thread state (NetConnectedPawns, LastSpeakTime) was
   causing undefined behavior. Internal mic is always the local player,
   no speaker arbitration needed.

4. StopListening flush: send directly to WebSocket (active speaker already
   established, no arbitration needed for the tail of the current turn).

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

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

@@ -543,14 +543,13 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::StopListening()
 		{
 			if (GetOwnerRole() == ROLE_Authority)
 			{
-				// Route through speaker arbitration so the correct player's
+				// Flush the final chunk directly to WebSocket.
-				// final chunk is attributed properly in multi-player.
+				// This is the tail of the current turn — the active speaker
-				APawn* LocalPawn = nullptr;
+				// is already established, no arbitration needed.
-				if (const APlayerController* PC = GetWorld()->GetFirstPlayerController())
+				if (WebSocketProxy && IsConnected())
 				{
-					LocalPawn = PC->GetPawn();
+					WebSocketProxy->SendAudioChunk(MicAccumulationBuffer);
 				}
-				ServerSendMicAudioFromPlayer(LocalPawn, MicAccumulationBuffer);
 			}
 			else
 			{
@@ -1523,17 +1522,15 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::OnMicrophoneDataCaptured(const TArray
 	{
 		if (GetOwnerRole() == ROLE_Authority)
 		{
-			// Route through speaker arbitration (local mic = local player).
+			// Internal mic = local player on the server. Send directly to WebSocket.
-			APawn* LocalPawn = nullptr;
+			// This callback runs on the WASAPI audio thread — accessing game-thread
-			if (UWorld* World = GetWorld())
+			// state (NetConnectedPawns, LastSpeakTime, etc.) is NOT safe here.
+			// Speaker arbitration is only needed for multi-player external mic.
+			if (WebSocketProxy && WebSocketProxy->IsConnected())
 			{
-				if (APlayerController* PC = World->GetFirstPlayerController())
+				WebSocketProxy->SendAudioChunk(MicAccumulationBuffer);
-				{
-					LocalPawn = PC->GetPawn();
 			}
 		}
-			ServerSendMicAudioFromPlayer(LocalPawn, MicAccumulationBuffer);
-		}
 		else
 		{
 			// Route through relay (clients can't call Server RPCs on NPC actors).
@@ -1872,8 +1869,37 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::ServerSendMicAudio_Implementation(
 void UPS_AI_ConvAgent_ElevenLabsComponent::ServerSendMicAudioFromPlayer(
 	APawn* SpeakerPawn, const TArray<uint8>& PCMBytes)
 {
-	if (!SpeakerPawn || !WebSocketProxy || !WebSocketProxy->IsConnected()) return;
+	if (!WebSocketProxy || !WebSocketProxy->IsConnected()) return;
-	if (!NetConnectedPawns.Contains(SpeakerPawn)) return;
+
+	// Standalone / single-player: bypass speaker arbitration entirely.
+	// There's only one player — no need for Contains check or speaker switching.
+	if (NetConnectedPawns.Num() <= 1)
+	{
+		WebSocketProxy->SendAudioChunk(PCMBytes);
+
+		// Keep speaker state in sync for debug display.
+		if (SpeakerPawn && NetActiveSpeakerPawn != SpeakerPawn)
+		{
+			SetActiveSpeaker(SpeakerPawn);
+		}
+		return;
+	}
+
+	// Multi-player path: full speaker arbitration.
+	if (!SpeakerPawn)
+	{
+		UE_LOG(LogPS_AI_ConvAgent_ElevenLabs, Warning,
+			TEXT("ServerSendMicAudioFromPlayer: null SpeakerPawn — audio dropped."));
+		return;
+	}
+
+	if (!NetConnectedPawns.Contains(SpeakerPawn))
+	{
+		UE_LOG(LogPS_AI_ConvAgent_ElevenLabs, Warning,
+			TEXT("ServerSendMicAudioFromPlayer: pawn %s not in NetConnectedPawns (%d players) — audio dropped."),
+			*SpeakerPawn->GetName(), NetConnectedPawns.Num());
+		return;
+	}
 
 	const double Now = FPlatformTime::Seconds();
 	LastSpeakTime.FindOrAdd(SpeakerPawn) = Now;

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

@@ -498,29 +498,32 @@ void UPS_AI_ConvAgent_InteractionComponent::StartConversationWithSelectedAgent()
 		return;
 	}
 
-	if (Agent->IsConnected() || Agent->bNetIsConversing)
-	{
 	if (bDebug)
 	{
-			UE_LOG(LogPS_AI_ConvAgent_Select, Log, TEXT("StartConversationWithSelectedAgent: agent already connected/conversing."));
+		UE_LOG(LogPS_AI_ConvAgent_Select, Log, TEXT("StartConversationWithSelectedAgent: %s (connected=%s conversing=%s)"),
-		}
+			Agent->GetOwner() ? *Agent->GetOwner()->GetName() : TEXT("(null)"),
-		return;
+			Agent->IsConnected() ? TEXT("true") : TEXT("false"),
+			Agent->bNetIsConversing ? TEXT("true") : TEXT("false"));
 	}
 
-	if (bDebug)
+	// Always call Join (idempotent) — even if the WebSocket is already connected
-	{
+	// (persistent session mode), we need to add the pawn to NetConnectedPawns
-		UE_LOG(LogPS_AI_ConvAgent_Select, Log, TEXT("StartConversationWithSelectedAgent: starting conversation with %s"),
+	// and set bNetIsConversing to true.
-			Agent->GetOwner() ? *Agent->GetOwner()->GetName() : TEXT("(null)"));
-	}
-
-	// Route through relay on clients (can't call Server RPCs on NPC actors).
 	if (GetOwnerRole() == ROLE_Authority || (GetWorld() && GetWorld()->GetNetMode() == NM_Standalone))
 	{
-		Agent->StartConversation();
+		APlayerController* PC = nullptr;
+		if (APawn* Pawn = Cast<APawn>(GetOwner()))
+		{
+			PC = Cast<APlayerController>(Pawn->GetController());
+		}
+		if (PC)
+		{
+			Agent->ServerJoinConversation_Implementation(PC);
+		}
 	}
 	else
 	{
-		ServerRelayStartConversation(Agent->GetOwner());
+		ServerRelayJoinConversation(Agent->GetOwner());
 	}
 
 	// Ensure mic is capturing so we can route audio to the agent.