fix(stt): strip Qwen3-ASR response prefix

Normalize the structured <asr_text> marker after extracting text from string, SDK object, and dictionary transcription responses. Preserve the current provider-aware STT configuration architecture.

Refreshes #8773 on current main.

Co-authored-by: angelos <angelos@oikos.lan.home.malaiwah.com>

Assisted-by: Codex:gpt-5.6

tests/tools/test_managed_media_gateways.py

@@ -347,6 +347,33 @@ def test_transcription_uses_model_specific_response_formats(monkeypatch, tmp_pat
     assert json_capture["close_calls"] == 1
 
 
+@pytest.mark.parametrize(
+    ("transcription", "expected"),
+    [
+        ("language English<asr_text>Hello from Qwen.", "Hello from Qwen."),
+        (
+            types.SimpleNamespace(text="language Chinese<asr_text>Object response."),
+            "Object response.",
+        ),
+        (
+            {"text": "language English<asr_text>Dictionary response."},
+            "Dictionary response.",
+        ),
+    ],
+)
+def test_extract_transcript_text_strips_qwen3_asr_prefix(
+    transcription,
+    expected,
+):
+    _install_fake_tools_package()
+    transcription_tools = _load_tool_module(
+        "tools.transcription_tools",
+        "transcription_tools.py",
+    )
+
+    assert transcription_tools._extract_transcript_text(transcription) == expected
+
+
 PLUGINS_DIR = Path(__file__).resolve().parents[2] / "plugins"
 
 

tools/transcription_tools.py

@@ -2076,17 +2076,26 @@ def _resolve_openai_audio_client_config() -> tuple[str, str]:
 
 def _extract_transcript_text(transcription: Any) -> str:
     """Normalize text and JSON transcription responses to a plain string."""
-    if isinstance(transcription, str):
-        return transcription.strip()
+    text: Optional[str] = None
 
-    if hasattr(transcription, "text"):
+    if isinstance(transcription, str):
+        text = transcription.strip()
+
+    if text is None and hasattr(transcription, "text"):
         value = getattr(transcription, "text")
         if isinstance(value, str):
-            return value.strip()
+            text = value.strip()
 
-    if isinstance(transcription, dict):
+    if text is None and isinstance(transcription, dict):
         value = transcription.get("text")
         if isinstance(value, str):
-            return value.strip()
+            text = value.strip()
 
-    return str(transcription).strip()
+    if text is None:
+        text = str(transcription).strip()
+
+    marker = "<asr_text>"
+    if marker in text:
+        text = text.split(marker, 1)[1].strip()
+
+    return text