Ajout sample CPP

Unreal/PS_AI_Agent/Plugins/PS_AI_Agent_ElevenLabs/Source/PS_AI_Agent_ElevenLabs/Private/ElevenLabsConversationalAgentComponent.cpp

@@ -604,9 +604,15 @@ void UElevenLabsConversationalAgentComponent::OnMicrophoneDataCaptured(const TAr
 {
 	if (!IsConnected() || !bIsListening) return;
 
+	// Echo suppression: skip sending mic audio while the agent is speaking.
+	// This prevents the agent from hearing its own voice through the speakers,
+	// which would confuse the server's VAD and STT. Matches the approach used
+	// in the official ElevenLabs C++ SDK (outputPlaying_ flag).
+	if (bAgentSpeaking) return;
+
 	// Convert this callback's samples to int16 bytes and accumulate.
-	// WASAPI fires every ~5ms (158 bytes at 16kHz). ElevenLabs needs ≥100ms
-	// (3200 bytes) per chunk for reliable VAD and STT. We hold bytes here
+	// WASAPI fires every ~5ms (158 bytes at 16kHz). ElevenLabs needs ≥250ms
+	// (8000 bytes) per chunk for reliable VAD and STT. We hold bytes here
 	// until we have enough, then send the whole batch in one WebSocket frame.
 	TArray<uint8> PCMBytes = FloatPCMToInt16Bytes(FloatPCM);
 	MicAccumulationBuffer.Append(PCMBytes);

Unreal/PS_AI_Agent/Plugins/PS_AI_Agent_ElevenLabs/Source/PS_AI_Agent_ElevenLabs/Private/ElevenLabsWebSocketProxy.cpp

@@ -491,6 +491,17 @@ void UElevenLabsWebSocketProxy::HandleAudioResponse(const TSharedPtr<FJsonObject
 		return;
 	}
 
+	// Discard audio belonging to an interrupted generation (event_id approach).
+	// Matches the official ElevenLabs C++ and Python SDKs: only AUDIO is filtered
+	// by event_id — transcripts, agent_response, etc. are always processed.
+	int32 EventId = 0;
+	(*AudioEvent)->TryGetNumberField(TEXT("event_id"), EventId);
+	if (EventId > 0 && EventId <= LastInterruptEventId)
+	{
+		UE_LOG(LogElevenLabsWS, Verbose, TEXT("Discarding audio event_id=%d (interrupted at %d)."), EventId, LastInterruptEventId);
+		return;
+	}
+
 	FString Base64Audio;
 	if (!(*AudioEvent)->TryGetStringField(TEXT("audio_base_64"), Base64Audio))
 	{
@@ -591,7 +602,20 @@ void UElevenLabsWebSocketProxy::HandleAgentChatResponsePart(const TSharedPtr<FJs
 
 void UElevenLabsWebSocketProxy::HandleInterruption(const TSharedPtr<FJsonObject>& Root)
 {
-	UE_LOG(LogElevenLabsWS, Log, TEXT("Agent interrupted (server ack received)."));
+	// Extract the interrupt event_id so we can filter stale audio frames.
+	// { "type": "interruption", "interruption_event": { "event_id": 42 } }
+	const TSharedPtr<FJsonObject>* InterruptEvent = nullptr;
+	if (Root->TryGetObjectField(TEXT("interruption_event"), InterruptEvent) && InterruptEvent)
+	{
+		int32 EventId = 0;
+		(*InterruptEvent)->TryGetNumberField(TEXT("event_id"), EventId);
+		if (EventId > LastInterruptEventId)
+		{
+			LastInterruptEventId = EventId;
+		}
+	}
+
+	UE_LOG(LogElevenLabsWS, Log, TEXT("Agent interrupted (server ack, LastInterruptEventId=%d)."), LastInterruptEventId);
 	OnInterrupted.Broadcast();
 }
 

Unreal/PS_AI_Agent/Plugins/PS_AI_Agent_ElevenLabs/Source/PS_AI_Agent_ElevenLabs/Public/ElevenLabsConversationalAgentComponent.h

@@ -400,5 +400,5 @@ private:
 	// ElevenLabs needs at least ~100ms (3200 bytes) per chunk for reliable VAD/STT.
 	// We accumulate here and only call SendAudioChunk once enough bytes are ready.
 	TArray<uint8> MicAccumulationBuffer;
-	static constexpr int32 MicChunkMinBytes = 3200; // 100ms @ 16kHz 16-bit mono
+	static constexpr int32 MicChunkMinBytes = 8000; // 250ms @ 16kHz 16-bit mono (4000 samples, matches ElevenLabs SDK recommendation)
 };

Unreal/PS_AI_Agent/Plugins/PS_AI_Agent_ElevenLabs/Source/PS_AI_Agent_ElevenLabs/Public/ElevenLabsWebSocketProxy.h

@@ -226,6 +226,13 @@ private:
 	// Used to compute [T+Xs] session-relative timestamps in all log messages.
 	double SessionStartTime = 0.0;
 
+	// ── Interrupt filtering (event_id approach, matching official SDK) ────────
+	// When the server sends an "interruption" event it includes an event_id.
+	// Audio events whose event_id <= LastInterruptEventId belong to the cancelled
+	// generation and must be discarded. Only AUDIO is filtered — transcripts,
+	// agent_response, agent_chat_response_part etc. are always processed.
+	int32 LastInterruptEventId = 0;
+
 public:
 	// Set by UElevenLabsConversationalAgentComponent before calling Connect().
 	// Controls turn_timeout in conversation_initiation_client_data.