wake word ok !

apps/robot-client/scripts/test_wakeword.ts

@@ -0,0 +1,150 @@
+#!/usr/bin/env npx tsx
+/**
+ * Test wake word detection using live ESP32 audio.
+ *
+ * Usage:
+ *   npx tsx scripts/test_wakeword.ts [--threshold 0.5] [--record out.raw]
+ *
+ * Connects to the ESP32 via serial, reads AUDIO_UP frames, and pipes
+ * the raw PCM into the wake_word.py subprocess. Prints detections live.
+ *
+ * --record <file>  Also dump raw PCM to a file so you can replay it later:
+ *                  python3 scripts/wake_word.py --model hey_jarvis --input stdin < out.raw
+ */
+import { SerialPort } from 'serialport';
+import { spawn, type ChildProcess } from 'node:child_process';
+import { createWriteStream, type WriteStream } from 'node:fs';
+import { parseArgs } from 'node:util';
+import { FrameDecoder, MsgType, encodeFrame } from '../src/hardware/protocol.js';
+
+const { values } = parseArgs({
+  options: {
+    threshold: { type: 'string', default: '0.5' },
+    record: { type: 'string' },
+    model: { type: 'string', default: 'hey_jarvis' },
+    python: { type: 'string', default: process.env.WAKEWORD_PYTHON_PATH || 'python3' },
+    port: { type: 'string', default: '/dev/serial0' },
+    baud: { type: 'string', default: '921600' },
+  },
+});
+
+const threshold = values.threshold!;
+const model = values.model!;
+const pythonPath = values.python!;
+const serialPath = values.port!;
+const baudRate = parseInt(values.baud!, 10);
+
+let recordStream: WriteStream | null = null;
+if (values.record) {
+  recordStream = createWriteStream(values.record);
+  console.log(`📁 Recording raw PCM to ${values.record}`);
+}
+
+// ── Spawn Python wake word process ──
+const pyArgs = [
+  './scripts/wake_word.py',
+  '--model', model,
+  '--threshold', threshold,
+  '--sample-rate', '16000',
+  '--input', 'stdin',
+  '--control-fd', '3',
+];
+
+console.log(`🐍 Spawning: ${pythonPath} ${pyArgs.join(' ')}`);
+console.log(`🎤 Threshold: ${threshold} | Model: ${model}`);
+console.log(`🔌 Serial: ${serialPath} @ ${baudRate}\n`);
+
+const py: ChildProcess = spawn(pythonPath, pyArgs, {
+  stdio: ['pipe', 'pipe', 'pipe', 'pipe'],
+});
+
+py.stdout?.on('data', (data: Buffer) => {
+  const lines = data.toString().trim().split('\n');
+  for (const line of lines) {
+    if (line.trim() === 'DETECTED') {
+      console.log(`\n🟢 DETECTED at ${new Date().toLocaleTimeString()}\n`);
+    }
+  }
+});
+
+py.stderr?.on('data', (data: Buffer) => {
+  const lines = data.toString().trim().split('\n');
+  for (const line of lines) {
+    const msg = line.trim();
+    if (msg === 'READY') {
+      console.log('✅ Wake word engine ready — say "Hey Jarvis"!\n');
+    } else if (msg.startsWith('Loading')) {
+      console.log(`⏳ ${msg}`);
+    } else if (msg.startsWith('Wake word model loaded')) {
+      console.log(`✅ ${msg}`);
+    } else if (!msg.includes('onnxruntime') && !msg.includes('UserWarning') && !msg.includes('warnings.warn')) {
+      console.log(`   [py] ${msg}`);
+    }
+  }
+});
+
+py.on('exit', (code) => {
+  console.log(`\n❌ Python process exited with code ${code}`);
+  process.exit(code ?? 1);
+});
+
+// ── Open serial and forward AUDIO_UP to Python stdin ──
+let audioChunks = 0;
+
+const decoder = new FrameDecoder((frame) => {
+  if (frame.type === MsgType.AUDIO_UP) {
+    audioChunks++;
+    if (py.stdin && !py.stdin.destroyed) {
+      py.stdin.write(frame.payload);
+    }
+    if (recordStream) {
+      recordStream.write(frame.payload);
+    }
+    // Progress indicator every ~1s (assuming ~100ms chunks)
+    if (audioChunks % 10 === 0) {
+      process.stdout.write(`\r🎧 Audio chunks: ${audioChunks}  `);
+    }
+  }
+});
+
+const serial = new SerialPort({ path: serialPath, baudRate, autoOpen: false });
+
+serial.on('data', (chunk: Buffer) => decoder.feed(chunk));
+serial.on('error', (err) => {
+  console.error('Serial error:', err.message);
+  process.exit(1);
+});
+
+serial.open((err) => {
+  if (err) {
+    console.error(`Failed to open ${serialPath}:`, err.message);
+    process.exit(1);
+  }
+  console.log(`🔌 Serial port open: ${serialPath}`);
+
+  // Send heartbeat so ESP32 stays active
+  setInterval(() => {
+    if (serial.isOpen) serial.write(encodeFrame(MsgType.STATUS));
+  }, 1000);
+});
+
+// ── Graceful shutdown ──
+function cleanup() {
+  console.log('\n\nShutting down...');
+  if (recordStream) {
+    recordStream.end();
+    console.log(`📁 Recording saved`);
+  }
+  const control = py.stdio[3] as unknown as NodeJS.WritableStream | null;
+  if (control && !(control as any).destroyed) {
+    control.write('QUIT\n');
+  }
+  setTimeout(() => {
+    py.kill('SIGTERM');
+    serial.close();
+    process.exit(0);
+  }, 500);
+}
+
+process.on('SIGINT', cleanup);
+process.on('SIGTERM', cleanup);

apps/robot-client/src/config/hardware.config.ts

@@ -74,7 +74,7 @@ export function loadHardwareConfig(): HardwareConfig {
       pythonPath: process.env.WAKEWORD_PYTHON_PATH || 'python3',
       scriptPath: process.env.WAKEWORD_SCRIPT_PATH || './scripts/wake_word.py',
       modelName: process.env.WAKEWORD_MODEL || 'hey_ti_pote',
-      threshold: parseFloat(process.env.WAKEWORD_THRESHOLD || '0.5'),
+      threshold: parseFloat(process.env.WAKEWORD_THRESHOLD || '0.75'),
     },
     serial: {
       // The ESP32 is now the mic/speaker front-end — serial link is

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

@@ -268,8 +268,8 @@ export class OrchestratorService extends EventEmitter {
       }
     }
 
-    // After playback, continue listening for more speech (continuous conversation)
-    this.continueListening();
+    // After playback, return to idle and wait for a new wake word
+    this.returnToIdle();
   }
 
   /**

apps/robot-client/src/services/wake-word.service.ts

@@ -42,10 +42,12 @@ export class WakeWordService extends EventEmitter {
   /** Latched forwarder so we can detach it on stop / error. */
   private readonly forwardMicChunk = (chunk: Buffer): void => {
     if (!this.process || !this.process.stdin || this.process.stdin.destroyed) return;
-    // Node gracefully buffers writes if the pipe is full; we don't
-    // apply back-pressure here because dropping wake-word audio would
-    // just hurt detection accuracy for a few tens of ms.
-    this.process.stdin.write(chunk);
+    this.process.stdin.write(chunk, (err) => {
+      if (err && (err as NodeJS.ErrnoException).code === 'EPIPE') {
+        this.logger.warn('Wake word process stdin pipe broken — detaching audio');
+        this.detachHardware();
+      }
+    });
   };
 
   constructor(
@@ -142,7 +144,7 @@ export class WakeWordService extends EventEmitter {
         } else if (msg.startsWith('Matched device') || msg.startsWith('Using device') || msg.startsWith('Listening')) {
           this.logger.info(`🔊 ${msg}`);
         } else {
-          this.logger.warn({ msg }, 'Wake word stderr');
+          this.logger.warn('Wake word stderr: %s', msg);
         }
       }
     });