fix: #1661 Preserve realtime session voice settings when updating agents (#1684)

## Summary

Resolves #1661

- merge run and initial realtime session settings when constructing a
session
- reuse the merged base when recomputing settings so the active voice is
preserved on handoffs or agent updates
- add a regression test that ensures the initial voice and output audio
format remain intact when refreshing session settings

## Testing
- make format
- make lint
- make mypy *(fails: optional dependencies numpy, sqlalchemy, and
litellm are not installed in the test environment)*
- make tests *(fails: optional dependencies numpy and litellm are not
installed in the test environment)*

Author: seratch

src/agents/realtime/session.py

@@ -95,6 +95,12 @@ def __init__(
         self._history: list[RealtimeItem] = []
         self._model_config = model_config or {}
         self._run_config = run_config or {}
+        initial_model_settings = self._model_config.get("initial_model_settings")
+        run_config_settings = self._run_config.get("model_settings")
+        self._base_model_settings: RealtimeSessionModelSettings = {
+            **(run_config_settings or {}),
+            **(initial_model_settings or {}),
+        }
         self._event_queue: asyncio.Queue[RealtimeSessionEvent] = asyncio.Queue()
         self._closed = False
         self._stored_exception: Exception | None = None
@@ -619,9 +625,8 @@ async def _get_updated_model_settings_from_agent(
         starting_settings: RealtimeSessionModelSettings | None,
         agent: RealtimeAgent,
     ) -> RealtimeSessionModelSettings:
-        # Start with run config model settings as base
-        run_config_settings = self._run_config.get("model_settings", {})
-        updated_settings: RealtimeSessionModelSettings = run_config_settings.copy()
+        # Start with the merged base settings from run and model configuration.
+        updated_settings = self._base_model_settings.copy()
 
         instructions, tools, handoffs = await asyncio.gather(
             agent.get_system_prompt(self._context_wrapper),

tests/realtime/test_session.py

@@ -1606,6 +1606,47 @@ async def mock_get_handoffs(cls, agent, context_wrapper):
             assert model_settings["tool_choice"] == "required"
             assert model_settings["output_audio_format"] == "g711_ulaw"
 
+    @pytest.mark.asyncio
+    async def test_model_settings_preserve_initial_settings_on_updates(self):
+        """Initial model settings should persist when we recompute settings for updates."""
+
+        agent = RealtimeAgent(name="test_agent", instructions="test")
+        agent.handoffs = []
+        agent.get_system_prompt = AsyncMock(return_value="test_prompt")  # type: ignore
+        agent.get_all_tools = AsyncMock(return_value=[])  # type: ignore
+
+        mock_model = Mock(spec=RealtimeModel)
+
+        initial_settings: RealtimeSessionModelSettings = {
+            "voice": "initial_voice",
+            "output_audio_format": "pcm16",
+        }
+
+        session = RealtimeSession(
+            model=mock_model,
+            agent=agent,
+            context=None,
+            model_config={"initial_model_settings": initial_settings},
+            run_config={},
+        )
+
+        async def mock_get_handoffs(cls, agent, context_wrapper):
+            return []
+
+        with pytest.MonkeyPatch().context() as m:
+            m.setattr(
+                "agents.realtime.session.RealtimeSession._get_handoffs",
+                mock_get_handoffs,
+            )
+
+            model_settings = await session._get_updated_model_settings_from_agent(
+                starting_settings=None,
+                agent=agent,
+            )
+
+        assert model_settings["voice"] == "initial_voice"
+        assert model_settings["output_audio_format"] == "pcm16"
+
 
 class TestUpdateAgentFunctionality:
     """Tests for update agent functionality in RealtimeSession"""