Merge pull request #10 from csheaff/fix/socat-shutdown-timeout

Fix socat dropping results on longer recordings

Author: csheaff

backends/moonshine-server

@@ -62,7 +62,7 @@ case "${1:-}" in
         if [ ! -S "$SOCK" ]; then
             "$0" start >&2 || exit 1
         fi
-        echo "$2" | socat -T 30 - UNIX-CONNECT:"$SOCK"
+        echo "$2" | socat -t 120 -T 120 - UNIX-CONNECT:"$SOCK"
         ;;
     *)
         echo "Usage: moonshine-server {start|stop|transcribe <audio.wav>}" >&2

backends/parakeet-server

@@ -61,7 +61,7 @@ case "${1:-}" in
         if [ ! -S "$SOCK" ]; then
             "$0" start >&2 || exit 1
         fi
-        echo "$2" | socat -T 30 - UNIX-CONNECT:"$SOCK"
+        echo "$2" | socat -t 120 -T 120 - UNIX-CONNECT:"$SOCK"
         ;;
     *)
         echo "Usage: parakeet-server {start|stop|transcribe <audio.wav>}" >&2

transcribe-server

@@ -61,13 +61,20 @@ case "${1:-}" in
         ;;
     transcribe)
         # Ensure daemon is alive (not just a stale socket from a crash)
-        if [ -S "$SOCK" ] && [ -f "$PIDFILE" ] && ! kill -0 "$(cat "$PIDFILE")" 2>/dev/null; then
+        if [ -f "$PIDFILE" ] && ! kill -0 "$(cat "$PIDFILE")" 2>/dev/null; then
             rm -f "$PIDFILE" "$SOCK"
         fi
         if [ ! -S "$SOCK" ]; then
             "$0" start >&2 || exit 1
         fi
-        echo "$2" | socat -T 30 - UNIX-CONNECT:"$SOCK"
+        TEXT=$(echo "$2" | socat -t 120 -T 120 - UNIX-CONNECT:"$SOCK" 2>/dev/null) || true
+        if [ -z "$TEXT" ] && [ -f "$PIDFILE" ] && ! kill -0 "$(cat "$PIDFILE")" 2>/dev/null; then
+            # Daemon died during transcription — restart and retry once
+            rm -f "$PIDFILE" "$SOCK"
+            "$0" start >&2 || exit 1
+            TEXT=$(echo "$2" | socat -t 120 -T 120 - UNIX-CONNECT:"$SOCK" 2>/dev/null) || true
+        fi
+        printf '%s' "$TEXT"
         ;;
     *)
         echo "Usage: transcribe-server {start|stop|transcribe <audio.wav>}" >&2

whisper-daemon.py

@@ -3,6 +3,8 @@
 import sys
 import socket
 import signal
+import logging
+import time
 from faster_whisper import WhisperModel
 
 SOCK_PATH = sys.argv[1]
@@ -11,10 +13,19 @@
 DEVICE = sys.argv[4]
 COMPUTE = sys.argv[5]
 
+LOG_PATH = os.path.join(os.environ.get("XDG_RUNTIME_DIR", "/tmp"), "talktype-whisper.log")
+logging.basicConfig(
+    filename=LOG_PATH, level=logging.INFO,
+    format="%(asctime)s %(levelname)s %(message)s",
+)
+log = logging.getLogger("whisper-daemon")
+
 # Load model once
+log.info("Loading faster-whisper %s (device=%s, compute=%s)...", MODEL_NAME, DEVICE, COMPUTE)
 print(f"Loading faster-whisper {MODEL_NAME}...", flush=True)
 model = WhisperModel(MODEL_NAME, device=DEVICE, compute_type=COMPUTE)
 print("Model loaded.", flush=True)
+log.info("Model loaded.")
 
 
 def transcribe(audio_path):
@@ -23,6 +34,7 @@ def transcribe(audio_path):
 
 
 def cleanup(*_):
+    log.info("Shutting down (signal).")
     try:
         os.unlink(SOCK_PATH)
     except OSError:
@@ -45,12 +57,22 @@ def cleanup(*_):
     try:
         audio_path = conn.recv(4096).decode().strip()
         if audio_path and os.path.isfile(audio_path):
+            file_size = os.path.getsize(audio_path)
+            log.info("Transcribing %s (%d bytes)...", audio_path, file_size)
+            t0 = time.monotonic()
             text = transcribe(audio_path)
+            elapsed = time.monotonic() - t0
+            log.info("Done in %.1fs, %d chars: %s", elapsed, len(text),
+                      text[:200] if text else "(empty)")
             conn.sendall(text.encode())
         else:
+            log.warning("Bad path: %r", audio_path)
             conn.sendall(b"")
     except Exception as e:
-        print(f"Error: {e}", file=sys.stderr, flush=True)
-        conn.sendall(b"")
+        log.exception("Error during transcription")
+        try:
+            conn.sendall(b"")
+        except Exception:
+            pass
     finally:
         conn.close()