Merge listen-robustness: transcription retries + missing-mic message

src/mcspeak/audio.py

@@ -217,6 +217,27 @@ async def _force_pwplay_volume_100() -> None:
             return
 
 
+async def _terminate_quietly(proc) -> None:
+    """Stop a recorder subprocess that may have ALREADY exited.
+
+    When the mic is disconnected, pw-record exits instantly — a later
+    proc.terminate() then raises ProcessLookupError, which otherwise masks the
+    friendly "no audio captured / is a mic connected?" error. Swallow that.
+    """
+    try:
+        proc.terminate()
+    except ProcessLookupError:
+        return
+    try:
+        await asyncio.wait_for(proc.wait(), timeout=2.0)
+    except asyncio.TimeoutError:
+        try:
+            proc.kill()
+        except ProcessLookupError:
+            return
+        await proc.wait()
+
+
 async def record_audio_until_silence(
     out_path: Path,
     silence_threshold_ms: int = 1500,
@@ -322,12 +343,7 @@ async def record_audio_until_silence(
             # The inner read returned no data — break outer loop
             break
     finally:
-        proc.terminate()
-        try:
-            await asyncio.wait_for(proc.wait(), timeout=2.0)
-        except asyncio.TimeoutError:
-            proc.kill()
-            await proc.wait()
+        await _terminate_quietly(proc)
 
     if not pcm_buf:
         raise PlaybackError(
@@ -414,14 +430,12 @@ async def record_audio(
         # If we get here, pw-record exited early (mic gone, permission, etc.)
     except asyncio.TimeoutError:
         # Expected path — stop recording cleanly.
-        proc.terminate()
-        try:
-            await asyncio.wait_for(proc.wait(), timeout=2.0)
-        except asyncio.TimeoutError:
-            proc.kill()
-            await proc.wait()
+        await _terminate_quietly(proc)
     except asyncio.CancelledError:
-        proc.kill()
+        try:
+            proc.kill()
+        except ProcessLookupError:
+            pass
         await proc.wait()
         raise
 
@@ -482,14 +496,12 @@ async def _record_fixed_stream(
     try:
         await asyncio.wait_for(_drain(), timeout=duration_seconds)
     except asyncio.TimeoutError:
-        proc.terminate()
-        try:
-            await asyncio.wait_for(proc.wait(), timeout=2.0)
-        except asyncio.TimeoutError:
-            proc.kill()
-            await proc.wait()
+        await _terminate_quietly(proc)
     except asyncio.CancelledError:
-        proc.kill()
+        try:
+            proc.kill()
+        except ProcessLookupError:
+            pass
         await proc.wait()
         raise
 

src/mcspeak/server.py

@@ -43,6 +43,10 @@ OUTPUT_DIR_CONTAINER = Path("/output")
 # the discipline _resolve_output_path uses for writes.
 _TRANSCRIBE_INPUT_DIRS = (Path("/output"), Path("/tmp/mcspeak"))
 
+# listen() retries transcription this many times (the Parakeet gateway
+# intermittently times out; the recording is saved, so retrying is free).
+_LISTEN_TRANSCRIBE_ATTEMPTS = 3
+
 
 def _validate_readable_audio_path(audio_path: str) -> Path:
     """Resolve audio_path against the allowed input dirs and check existence.
@@ -893,19 +897,37 @@ async def listen(
             await ctx.info(f"Recording saved failed (non-fatal): {e}")
 
     await ctx.info(f"Transcribing {rec_path.name}...")
-    try:
-        result = await transcribe_audio(
-            rec_path,
-            response_format=response_format,
-            timestamp_granularities=timestamp_granularities,
-            diarize=diarize,
-            num_speakers=num_speakers,
-            punctuation=punctuation,
-            min_confidence=min_confidence,
-        )
-    except TranscriptionError as e:
+    # The Parakeet gateway serializes inference on one slot and intermittently
+    # times out when the backend is busy/cold. The recording is already saved,
+    # so retry a couple of times before giving up rather than making the caller
+    # re-run transcribe() by hand.
+    result = None
+    last_err: TranscriptionError | None = None
+    for attempt in range(_LISTEN_TRANSCRIBE_ATTEMPTS):
+        try:
+            result = await transcribe_audio(
+                rec_path,
+                response_format=response_format,
+                timestamp_granularities=timestamp_granularities,
+                diarize=diarize,
+                num_speakers=num_speakers,
+                punctuation=punctuation,
+                min_confidence=min_confidence,
+            )
+            break
+        except TranscriptionError as e:
+            last_err = e
+            if attempt < _LISTEN_TRANSCRIBE_ATTEMPTS - 1:
+                await ctx.info(
+                    f"Transcription attempt {attempt + 1} failed ({e}); retrying..."
+                )
+                await asyncio.sleep(0.75)
+    if result is None:
         return {
-            "error": str(e),
+            "error": (
+                f"Transcription failed after {_LISTEN_TRANSCRIBE_ATTEMPTS} attempts: "
+                f"{last_err}"
+            ),
             "recorded": str(rec_path),
             "saved_to": saved_to,
         }