infinicode 2.8.88 → 2.8.89

package/docs/ROBOPARK_PRODUCTION_MEMORY.md

@@ -60,6 +60,10 @@ ROBOVOICE checkout, and rebuilds Compose once.
 The preview agent previously used PyAudio for the selected speaker. BMW's
 proven output is ALSA `aplay -D plughw:2,0`. Infinicode `2.8.86` uses that
 same endpoint for live TTS whenever a Linux output selection contains `hw:`.
+Motion cues must use that same ALSA endpoint and finish before the greeting
+starts. Running a PyAudio cue concurrently with ALSA TTS creates an exclusive
+device race on `hw:2,0`. `playback_started` is valid only after PCM has been
+written to the physical speaker, not merely received from LiveKit.
 
 ### Silent Conversation After Greeting
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "infinicode",
-  "version": "2.8.88",
+  "version": "2.8.89",
   "description": "OpenKernel — provider-agnostic AI execution kernel. Native coding agent + mission-driven execution runtime.",
   "type": "module",
   "main": "./dist/kernel/index.js",

package/packages/robopark/scheduler/preview_agent.py

@@ -83,22 +83,51 @@ def _normalize_livekit_url(url: Optional[str]) -> Optional[str]:
     return url
 
 
-def _play_audio_effect(selected_output: str | None, effect: str) -> None:
-    """Play a short local cue without involving the voice pipeline."""
-    try:
-        import pyaudio
-    except Exception:
-        return
-    sample_rate = 48000
-    channels = 2
+def _play_audio_effect(selected_output: str | None, effect: str) -> None:
+    """Play a short local cue without involving the voice pipeline."""
+    sample_rate = 48000
+    channels = 2
     if effect == "motion":
         notes = ((880, 0.09), (1320, 0.13))
     else:
         notes = ((660, 0.10), (440, 0.16))
-    pa = pyaudio.PyAudio()
-    device_index = None
-    selected = str(selected_output or "default")
-    try:
+    selected = str(selected_output or "default")
+    frames = bytearray()
+    for frequency, duration in notes:
+        count = int(sample_rate * duration)
+        for n in range(count):
+            envelope = min(1.0, n / 240.0, (count - n) / 1200.0)
+            value = int(5000 * envelope * math.sin(2 * math.pi * frequency * n / sample_rate))
+            frames.extend(struct.pack("<hh", value, value))
+
+    if sys.platform.startswith("linux") and "hw:" in selected:
+        import re
+        import subprocess
+
+        match = re.search(r"\b(hw:\d+,\d+)\b", selected)
+        if not match:
+            return
+        result = subprocess.run(
+            [
+                "aplay", "-q", "-D", f"plug{match.group(1)}", "-t", "raw",
+                "-f", "S16_LE", "-r", str(sample_rate), "-c", str(channels),
+            ],
+            input=bytes(frames), capture_output=True, timeout=3.0,
+        )
+        if result.returncode:
+            logger.warning(
+                "audio effect failed on %s: %s",
+                match.group(1), result.stderr.decode("utf-8", errors="replace").strip(),
+            )
+        return
+
+    try:
+        import pyaudio
+    except Exception:
+        return
+    pa = pyaudio.PyAudio()
+    device_index = None
+    try:
         if selected.strip().isdigit():
             device_index = int(selected.strip())
         elif selected.lower() == "default":
@@ -118,14 +147,7 @@ def _play_audio_effect(selected_output: str | None, effect: str) -> None:
             output=True,
             output_device_index=device_index,
         )
-        for frequency, duration in notes:
-            count = int(sample_rate * duration)
-            frames = bytearray()
-            for n in range(count):
-                envelope = min(1.0, n / 240.0, (count - n) / 1200.0)
-                value = int(5000 * envelope * math.sin(2 * math.pi * frequency * n / sample_rate))
-                frames.extend(struct.pack("<hh", value, value))
-            stream.write(bytes(frames))
+        stream.write(bytes(frames))
         stream.stop_stream()
         stream.close()
     except Exception as e:
@@ -1041,9 +1063,10 @@ class PreviewAgent:
         # for the LiveKit publisher.
         await asyncio.sleep(0.5)
         # I/O. Camera/mic join and the preset greeting are the critical path.
-        asyncio.create_task(
-            asyncio.to_thread(_play_audio_effect, self.audio_playback_device, "motion")
-        )
+        # The cue and TTS share one physical ALSA output. Finish and release
+        # the cue before LiveKit can deliver the greeting; background playback
+        # races aplay and makes greeting delivery intermittently fail busy.
+        await asyncio.to_thread(_play_audio_effect, self.audio_playback_device, "motion")
         try:
             r = await self._session.post(
                 f"{self.scheduler_url.rstrip('/')}/api/devices/{self.device_id}/request-session",
@@ -1350,17 +1373,27 @@ class LiveKitPublisher:
         # steady pace, independent of how unevenly frames actually arrive.
         import queue
         import threading
-        write_queue: "queue.Queue[Optional[bytes]]" = queue.Queue()
-        written_frames = [0]
-        PREBUFFER_CHUNKS = 5
-        stream_failed = threading.Event()
+        write_queue: "queue.Queue[Optional[bytes]]" = queue.Queue()
+        written_frames = [0]
+        PREBUFFER_CHUNKS = 5
+        stream_failed = threading.Event()
+        playback_reported = threading.Event()
+        event_loop = asyncio.get_running_loop()
 
         def _safe_write(chunk: bytes) -> bool:
             if stream_failed.is_set():
                 return False
-            try:
-                out.write(chunk)
-                return True
+            try:
+                out.write(chunk)
+                if not playback_reported.is_set():
+                    playback_reported.set()
+                    asyncio.run_coroutine_threadsafe(
+                        self.agent._report_pipeline(
+                            "playback_started", "ok", "First TTS audio chunk written to robot speaker"
+                        ),
+                        event_loop,
+                    )
+                return True
             except Exception as e:
                 stream_failed.set()
                 logger.warning(f"audio out stream closed; disabling playback for this track: {e}")
@@ -1420,7 +1453,6 @@ class LiveKitPublisher:
                         f"audio out: first frame received for {sid} "
                         f"(rate={af.sample_rate}, channels={getattr(af, 'num_channels', 1)})"
                     )
-                    await self.agent._report_pipeline("playback_started", "ok", "First TTS audio frame reached the robot")
                 in_rate = af.sample_rate
                 in_channels = int(getattr(af, "num_channels", 1) or 1)
                 raw = bytes(af.data)