hey jarvis banger

.claude/settings.local.json

@@ -0,0 +1,8 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(wc -l /Users/arthurbarre/dev/perso/ti-pote/apps/robot-client/src/**/*.ts)",
+      "Bash(ssh tipote@192.168.1.124 \"ps aux | grep node | grep -v grep; echo '---LOGS---'; cat /tmp/tipote.log\")"
+    ]
+  }
+}

apps/backend/src/adapters/inbound/websocket/robot.gateway.ts

@@ -121,8 +121,20 @@ export class RobotGateway implements OnGatewayConnection, OnGatewayDisconnect, I
     @ConnectedSocket() client: AuthenticatedSocket,
     @MessageBody() message: AudioChunkMessage,
   ) {
-    this.conversationPort.processAudioChunk(client.data.deviceId, message.data, message.sampleRate);
+    const chunk = message.data;
+    // Debug: log audio chunk info to diagnose STT issues
+    if (!this._audioLogCount) this._audioLogCount = 0;
+    this._audioLogCount++;
+    if (this._audioLogCount <= 3 || this._audioLogCount % 100 === 0) {
+      this.logger.debug(
+        `Audio chunk #${this._audioLogCount}: type=${typeof chunk}, isBuffer=${Buffer.isBuffer(chunk)}, ` +
+        `constructor=${chunk?.constructor?.name}, length=${chunk?.length ?? chunk?.byteLength ?? 'N/A'}, ` +
+        `sampleRate=${message.sampleRate}`,
+      );
+    }
+    this.conversationPort.processAudioChunk(client.data.deviceId, chunk, message.sampleRate);
   }
+  private _audioLogCount = 0;
 
   @SubscribeMessage('speech_end')
   async handleSpeechEnd(@ConnectedSocket() client: AuthenticatedSocket) {

apps/backend/src/core/services/conversation.service.ts

@@ -97,7 +97,7 @@ export class ConversationService implements IConversationPort {
     this.activeSessions.set(deviceId, session);
 
     const sttStream = await this.sttPort.openStream((result: TranscriptionResult) => {
-      this.logger.debug(
+      this.logger.log(
         `STT [${deviceId}]: "${result.text}" (final: ${result.isFinal}, confidence: ${result.confidence})`,
       );
 
@@ -156,6 +156,9 @@ export class ConversationService implements IConversationPort {
     this.logger.log(`Final transcription for ${deviceId}: "${finalText}"`);
 
     if (!finalText) {
+      this.logger.warn(`No transcription for ${deviceId} — returning to idle`);
+      this.deviceGateway.sendStatus(deviceId, 'idle');
+      this.activeSessions.delete(deviceId);
       return null;
     }
 

apps/robot-client/scripts/wake_word.py

@@ -147,11 +147,15 @@ def run_predict_loop(oww_model, read_chunk, state: State, threshold: float):
                     # Keep draining but don't emit detections.
                     continue
 
-            for _, score in oww_model.prediction_buffer.items():
-                if len(score) > 0 and score[-1] > threshold:
-                    print("DETECTED", flush=True)
-                    oww_model.reset()
-                    break
+            for name, score in oww_model.prediction_buffer.items():
+                if len(score) > 0:
+                    s = score[-1]
+                    if s > 0.05:
+                        print(f"SCORE: {name}={s:.3f}", file=sys.stderr, flush=True)
+                    if s > threshold:
+                        print("DETECTED", flush=True)
+                        oww_model.reset()
+                        break
     except KeyboardInterrupt:
         pass
 

apps/robot-client/src/services/orchestrator.service.ts

@@ -25,7 +25,7 @@ export class OrchestratorService extends EventEmitter {
   /** Timer for Voice Activity Detection (silence timeout) */
   private silenceTimer: ReturnType<typeof setTimeout> | null = null;
   private readonly silenceTimeoutMs = 2000; // 2s of silence = speech end
-  private readonly initialGracePeriodMs = 3000; // 3s grace period before silence detection kicks in
+  private readonly initialGracePeriodMs = 5000; // 5s grace period before silence detection kicks in
 
   /** Track when the last audio chunk was received */
   private lastAudioChunkTime = 0;
@@ -243,6 +243,14 @@ export class OrchestratorService extends EventEmitter {
         }
         this.audioBuffer = [];
       }
+
+      // Reset speech detection: the wake word phrase ("Hey Jarvis") itself
+      // triggers hasDetectedSpeech=true, and the natural pause after saying it
+      // fires the 2s silence timeout before the user can ask their question.
+      // Restart the clock from NOW so the user gets the full grace period.
+      this.hasDetectedSpeech = false;
+      this.lastAudioChunkTime = Date.now();
+      this.conversationStartTime = Date.now();
     }
 
     // The cloud drives the state machine for thinking → speaking → idle
@@ -269,8 +277,10 @@ export class OrchestratorService extends EventEmitter {
       }
     }
 
-    // After playback, return to idle and wait for a new wake word
-    this.returnToIdle();
+    // After playback, keep the conversation going — listen for follow-up
+    // without requiring a new wake word. The conversation ends naturally
+    // when silence exceeds the grace period (no speech detected).
+    this.continueListening();
   }
 
   /**

apps/robot-client/src/services/wifi.service.ts

@@ -37,8 +37,9 @@ export class WifiService {
     }
 
     try {
-      const { stdout } = await execAsync('nmcli -t -f TYPE,STATE device | grep wifi');
-      return stdout.includes('connected') && !stdout.includes('disconnected');
+      const { stdout } = await execAsync('nmcli -t -f TYPE,STATE device');
+      // Check that the main wifi device (not wifi-p2p) is connected
+      return stdout.split('\n').some((line) => line === 'wifi:connected');
     } catch {
       return false;
     }