fix: 优化长语音上传交互并修复输入框高度跳变

api/routers/websocket.py

@@ -314,6 +314,7 @@ async def _process_audio_segment_async(
                 "data": {
                     "text": transcript_text or "",
                     "audio_duration": audio_duration,
+                    "voice_session_id": voice_session_id,
                     "segment_index": segment_index,
                     "is_last": is_last,
                 },
@@ -383,18 +384,6 @@ async def _process_audio_segment_async(
                     user=user,
                 )
 
-            if is_last:
-                await manager.send_message(conversation_id, {
-                    "type": MessageType.AGENT_RESPONSE,
-                    "conversation_id": conversation_id,
-                    "data": {
-                        "text": "最后一段语音已收到，我会继续完善这一轮总结。",
-                        "transition": True,
-                        "is_last": True,
-                        "segment_index": segment_index,
-                    },
-                    "timestamp": datetime.now(timezone.utc).isoformat(),
-                })
             break
 
     except Exception as e:

api/tests/test_websocket_baseline.py

@@ -463,6 +463,7 @@ class WebSocketBaselineTest(unittest.IsolatedAsyncioTestCase):
                     "type": "audio_segment",
                     "data": {
                         "audio_base64": "seg-1",
+                        "voice_session_id": "voice-session-1",
                         "segment_index": 1,
                         "duration": 12,
                         "is_last": False,
@@ -472,6 +473,7 @@ class WebSocketBaselineTest(unittest.IsolatedAsyncioTestCase):
                     "type": "audio_segment",
                     "data": {
                         "audio_base64": "seg-0",
+                        "voice_session_id": "voice-session-1",
                         "segment_index": 0,
                         "duration": 10,
                         "is_last": False,
@@ -523,6 +525,15 @@ class WebSocketBaselineTest(unittest.IsolatedAsyncioTestCase):
         ]
         self.assertEqual(ordered_messages, ["这是第 0 段", "这是第 1 段"])
         self.assertEqual(len([obj for obj in fake_db.added if isinstance(obj, Segment)]), 2)
+        transcript_msgs = [
+            item["message"]
+            for item in fake_manager.sent_messages
+            if item["message"]["type"] == ws_router.MessageType.TRANSCRIPT
+        ]
+        self.assertEqual(
+            [msg["data"]["voice_session_id"] for msg in transcript_msgs],
+            ["voice-session-1", "voice-session-1"],
+        )
 
     async def test_audio_segment_duplicate_index_is_idempotent(self):
         user = _make_user()
@@ -727,6 +738,68 @@ class WebSocketBaselineTest(unittest.IsolatedAsyncioTestCase):
         ]
         self.assertGreaterEqual(len(transition_msgs), 1)
 
+    async def test_audio_segment_last_segment_does_not_emit_terminal_transition(self):
+        user = _make_user()
+        conversation = Conversation(id="conv-1", user_id=user.id, status="active")
+        fake_db = _FakeAsyncDB(user=user, conversation=conversation)
+        fake_manager = _FakeManager()
+        fake_websocket = _FakeWebSocket(
+            messages=[
+                {
+                    "type": "audio_segment",
+                    "data": {
+                        "audio_base64": "last-seg-0",
+                        "voice_session_id": "voice-session-last",
+                        "client_segment_id": "voice-session-last-0",
+                        "segment_index": 0,
+                        "duration": 15,
+                        "is_last": True,
+                    },
+                },
+                WebSocketDisconnect(),
+            ]
+        )
+
+        process_user_message_mock = AsyncMock()
+        transcribe_mock = AsyncMock(return_value="最后一段转写")
+
+        with ExitStack() as stack:
+            stack.enter_context(
+                patch.object(
+                    ws_router,
+                    "verify_token",
+                    return_value={"type": "access", "sub": user.id},
+                )
+            )
+            stack.enter_context(
+                patch.object(ws_router, "get_async_db", _db_provider(fake_db))
+            )
+            stack.enter_context(patch.object(ws_router, "manager", fake_manager))
+            stack.enter_context(
+                patch("routers.quota.get_segment_count", new=AsyncMock(return_value=0))
+            )
+            stack.enter_context(
+                patch("routers.quota.check_can_send_message", return_value=(True, ""))
+            )
+            stack.enter_context(
+                patch.object(ws_router, "process_user_message", process_user_message_mock)
+            )
+            stack.enter_context(
+                patch.object(ws_router.asr_service, "transcribe", transcribe_mock)
+            )
+
+            await ws_router.websocket_endpoint(fake_websocket, "conv-1")
+            await asyncio.sleep(0.05)
+
+        transition_msgs = [
+            item["message"]
+            for item in fake_manager.sent_messages
+            if item["message"]["type"] == ws_router.MessageType.AGENT_RESPONSE
+            and item["message"].get("data", {}).get("transition") is True
+        ]
+        self.assertEqual(len(transition_msgs), 1)
+        self.assertIsNone(transition_msgs[0]["data"].get("is_last"))
+
     async def test_audio_segment_continues_after_reconnect_with_existing_previous_segment(self):
         user = _make_user()
         conversation = Conversation(id="conv-1", user_id=user.id, status="active")