diff --git a/src/utils/files.py b/src/utils/files.py
index 56212da6..b57a691c 100644
--- a/src/utils/files.py
+++ b/src/utils/files.py
@@ -151,7 +151,8 @@ class AudioProcessor:
         suffix = self.get_output_suffix(filename, content_type)
         normalized_filename = self.ensure_audio_filename(filename, suffix)
         normalized_content_type = self.normalize_content_type(filename, content_type)
-        self._probe_audio_duration_seconds(
+        await asyncio.to_thread(
+            self._probe_audio_duration_seconds,
             content,
             suffix,
         )
diff --git a/tests/utils/test_audio_processing.py b/tests/utils/test_audio_processing.py
index e8768ed8..f951ed14 100644
--- a/tests/utils/test_audio_processing.py
+++ b/tests/utils/test_audio_processing.py
@@ -149,6 +149,31 @@ async def test_audio_processor_extract_text_transcribes_directly():
     assert "transcription_fallback_used" not in extracted.metadata
 
 
+@pytest.mark.asyncio
+async def test_audio_processor_extract_text_probes_in_background_thread():
+    processor = AudioProcessor()
+    mock_probe = AsyncMock(return_value=1.0)
+    to_thread = AsyncMock(return_value=1.0)
+
+    with (
+        patch.object(processor, "_probe_audio_duration_seconds", mock_probe),
+        patch("src.utils.files.asyncio.to_thread", to_thread),
+        patch("src.utils.files.transcribe_audio", new=AsyncMock(return_value="ok")),
+    ):
+        extracted = await processor.extract_text(
+            b"bytes",
+            filename="voice-note.mp3",
+            content_type="audio/mpeg",
+        )
+
+    to_thread.assert_awaited_once_with(
+        mock_probe,
+        b"bytes",
+        ".mp3",
+    )
+    assert extracted.text == "ok"
+
+
 @pytest.mark.asyncio
 async def test_audio_processor_extract_text_allows_empty_transcript():
     processor = AudioProcessor()