feat: add voice session confirmation guardrails

backend/tests/test_voice_sessions.py

@@ -6,6 +6,7 @@ from app.core.config import settings
 from app.db.database import get_db
 from app.main import app
 from app.services.adapters.text.models import StoryOutput
+from app.services.voice_transcription_service import VoiceTranscriptionResult
 
 
 async def test_voice_session_create_and_fallback_turn_returns_audio(
@@ -272,6 +273,82 @@ async def test_voice_session_uploaded_audio_turn_uses_demo_transcript_hint(
             app.dependency_overrides.clear()
 
 
+async def test_voice_session_low_confidence_turn_requests_confirmation(
+    db_session,
+    auth_token,
+):
+    async def override_get_db():
+        yield db_session
+
+    app.dependency_overrides[get_db] = override_get_db
+
+    with (
+        patch(
+            "app.services.voice_session_service.generate_story_content",
+            new_callable=AsyncMock,
+        ) as mock_generate,
+        patch(
+            "app.services.voice_session_service.text_to_speech",
+            new_callable=AsyncMock,
+        ) as mock_tts,
+        patch(
+            "app.services.voice_session_service.transcribe_voice_audio",
+            new_callable=AsyncMock,
+        ) as mock_transcribe,
+    ):
+        mock_tts.return_value = b"confirmation-audio"
+        mock_transcribe.return_value = VoiceTranscriptionResult(
+            transcript_text="我想听一个会发光的小恐龙故事",
+            confidence=0.41,
+            provider="openai",
+        )
+
+        transport = ASGITransport(app=app)
+        try:
+            async with AsyncClient(transport=transport, base_url="http://test") as client:
+                client.cookies.set("access_token", auth_token)
+
+                response = await client.post("/api/voice-sessions", json={})
+                assert response.status_code == 201
+                session_id = response.json()["id"]
+
+                response = await client.post(
+                    f"/api/voice-sessions/{session_id}/turns",
+                    files={
+                        "audio_file": ("turn.webm", b"fake-webm-audio", "audio/webm"),
+                    },
+                )
+                assert response.status_code == 202
+                turn_id = response.json()["turn_id"]
+
+                response = await client.get(
+                    f"/api/voice-sessions/{session_id}/turns/{turn_id}"
+                )
+                assert response.status_code == 200
+                turn_data = response.json()
+                assert turn_data["status"] == "audio_ready"
+                assert turn_data["requires_confirmation"] is True
+                assert turn_data["understanding_summary"].startswith("本轮系统理解为")
+                assert "请家长帮忙确认" in turn_data["confirmation_message"]
+                assert turn_data["assistant_text"] == turn_data["confirmation_message"]
+
+                response = await client.get(f"/api/voice-sessions/{session_id}")
+                assert response.status_code == 200
+                session_data = response.json()
+                assert session_data["latest_requires_confirmation"] is True
+                assert "请家长帮忙确认" in session_data["latest_confirmation_message"]
+                assert session_data["can_finalize"] is False
+                assert session_data["story_state"]["narrative_segments"] == []
+                assert any(
+                    event["event_type"] == "turn_confirmation_requested"
+                    for event in session_data["events"]
+                )
+
+                mock_generate.assert_not_awaited()
+        finally:
+            app.dependency_overrides.clear()
+
+
 async def test_voice_session_list_orders_recent_sessions_first(
     db_session,
     auth_token,