diff --git a/isr.py b/isr.py
index 01abfac..4b05ec0 100644
--- a/isr.py
+++ b/isr.py
@@ -186,13 +186,15 @@ class ALSAStream:
             self._process = subprocess.Popen(
                 cmd,
                 stdout=subprocess.PIPE,
-                stderr=subprocess.DEVNULL,
+                stderr=subprocess.PIPE,
             )
         except FileNotFoundError:
             raise RuntimeError("arecord not found - install alsa-utils: sudo apt install alsa-utils")
 
         self._thread = threading.Thread(target=self._read_loop, daemon=True)
+        self._stderr_thread = threading.Thread(target=self._log_stderr, daemon=True)
         self._thread.start()
+        self._stderr_thread.start()
         return self
 
     def __exit__(self, *args):
@@ -206,6 +208,12 @@ class ALSAStream:
         if self._thread:
             self._thread.join(timeout=1)
 
+    def _log_stderr(self):
+        for line in self._process.stderr:
+            line = line.decode('utf-8', errors='replace').rstrip()
+            if line:
+                self.logger.warning(f"arecord: {line}")
+
     def _read_loop(self):
         chunk_size = self.sample_rate * self.channels * 2 // 10  # 100 ms chunks
         while self._running and self._process.poll() is None:
@@ -306,6 +314,14 @@ class AudioSystem:
             if spec_lower in dev.name.lower():
                 return dev
 
+        # No hardware device matched.  Treat spec as a direct ALSA PCM name so
+        # virtual devices defined in asound.conf (dsnoop, plug, etc.) work without
+        # appearing in `arecord -l`.
+        if 'alsa' in self._backends:
+            return AudioDevice(
+                id=spec, name=spec, channels=2, sample_rate=44100, backend='alsa'
+            )
+
         return None