diff --git a/examples/realtime/twilio_sip/README.md b/examples/realtime/twilio_sip/README.md
index a96e5d379..d74a5960b 100644
--- a/examples/realtime/twilio_sip/README.md
+++ b/examples/realtime/twilio_sip/README.md
@@ -25,7 +25,7 @@ This example shows how to handle OpenAI Realtime SIP calls with the Agents SDK.
 
 1. Install dependencies:
    ```bash
-   uv pip install -r examples/realtime/twilio-sip/requirements.txt
+   uv pip install -r examples/realtime/twilio_sip/requirements.txt
    ```
 2. Export required environment variables:
    ```bash
diff --git a/src/agents/realtime/openai_realtime.py b/src/agents/realtime/openai_realtime.py
index 3aad14c8a..2df33cb64 100644
--- a/src/agents/realtime/openai_realtime.py
+++ b/src/agents/realtime/openai_realtime.py
@@ -198,6 +198,7 @@ def __init__(self) -> None:
         self._playback_tracker: RealtimePlaybackTracker | None = None
         self._created_session: OpenAISessionCreateRequest | None = None
         self._server_event_type_adapter = get_server_event_type_adapter()
+        self._call_id: str | None = None
 
     async def connect(self, options: RealtimeModelConfig) -> None:
         """Establish a connection to the model and keep it alive."""
@@ -220,6 +221,7 @@ async def connect(self, options: RealtimeModelConfig) -> None:
         if model_name:
             self.model = model_name
 
+        self._call_id = call_id
         api_key = await get_api_key(options.get("api_key"))
 
         if "tracing" in model_settings:
@@ -833,10 +835,13 @@ def _get_session_config(
         speed = model_settings.get("speed")
         modalities = model_settings.get("modalities", DEFAULT_MODEL_SETTINGS.get("modalities"))
 
-        input_audio_format = model_settings.get(
-            "input_audio_format",
-            DEFAULT_MODEL_SETTINGS.get("input_audio_format"),
-        )
+        if self._call_id:
+            input_audio_format = model_settings.get("input_audio_format")
+        else:
+            input_audio_format = model_settings.get(
+                "input_audio_format",
+                DEFAULT_MODEL_SETTINGS.get("input_audio_format"),
+            )
         input_audio_transcription = model_settings.get(
             "input_audio_transcription",
             DEFAULT_MODEL_SETTINGS.get("input_audio_transcription"),
@@ -845,10 +850,13 @@ def _get_session_config(
             "turn_detection",
             DEFAULT_MODEL_SETTINGS.get("turn_detection"),
         )
-        output_audio_format = model_settings.get(
-            "output_audio_format",
-            DEFAULT_MODEL_SETTINGS.get("output_audio_format"),
-        )
+        if self._call_id:
+            output_audio_format = model_settings.get("output_audio_format")
+        else:
+            output_audio_format = model_settings.get(
+                "output_audio_format",
+                DEFAULT_MODEL_SETTINGS.get("output_audio_format"),
+            )
         input_audio_noise_reduction = model_settings.get(
             "input_audio_noise_reduction",
             DEFAULT_MODEL_SETTINGS.get("input_audio_noise_reduction"),
diff --git a/tests/realtime/test_openai_realtime.py b/tests/realtime/test_openai_realtime.py
index 08c45e5d7..f8eb725ff 100644
--- a/tests/realtime/test_openai_realtime.py
+++ b/tests/realtime/test_openai_realtime.py
@@ -606,6 +606,29 @@ def test_get_and_update_session_config(self, model):
         assert cfg.audio is not None and cfg.audio.output is not None
         assert cfg.audio.output.voice == "verse"
 
+    def test_session_config_defaults_audio_formats_when_not_call(self, model):
+        settings: dict[str, Any] = {}
+        cfg = model._get_session_config(settings)
+        assert cfg.audio is not None
+        assert cfg.audio.input is not None
+        assert cfg.audio.input.format is not None
+        assert cfg.audio.input.format.type == "audio/pcm"
+        assert cfg.audio.output is not None
+        assert cfg.audio.output.format is not None
+        assert cfg.audio.output.format.type == "audio/pcm"
+
+    def test_session_config_preserves_sip_audio_formats(self, model):
+        model._call_id = "call-123"
+        settings = {
+            "turn_detection": {"type": "semantic_vad", "interrupt_response": True},
+        }
+        cfg = model._get_session_config(settings)
+        assert cfg.audio is not None
+        assert cfg.audio.input is not None
+        assert cfg.audio.input.format is None
+        assert cfg.audio.output is not None
+        assert cfg.audio.output.format is None
+
     @pytest.mark.asyncio
     async def test_handle_error_event_success(self, model):
         """Test successful handling of error events."""