fix(realtime): omit null audio formats in SIP call attach updates (#2454)

Author: weiguangli-io

src/agents/realtime/openai_realtime.py

@@ -1043,7 +1043,9 @@ def _get_session_config(
                 input_format_source = model_settings.get(
                     "input_audio_format", DEFAULT_MODEL_SETTINGS.get("input_audio_format")
                 )
-        audio_input_args["format"] = to_realtime_audio_format(input_format_source)
+        input_format = to_realtime_audio_format(input_format_source)
+        if input_format is not None:
+            audio_input_args["format"] = input_format
 
         if "noise_reduction" in input_audio_config:
             audio_input_args["noise_reduction"] = input_audio_config.get("noise_reduction")
@@ -1085,7 +1087,9 @@ def _get_session_config(
                 output_format_source = model_settings.get(
                     "output_audio_format", DEFAULT_MODEL_SETTINGS.get("output_audio_format")
                 )
-        audio_output_args["format"] = to_realtime_audio_format(output_format_source)
+        output_format = to_realtime_audio_format(output_format_source)
+        if output_format is not None:
+            audio_output_args["format"] = output_format
 
         if "speed" in output_audio_config:
             audio_output_args["speed"] = output_audio_config.get("speed")

tests/realtime/test_realtime_model_settings.py

@@ -6,6 +6,7 @@
 from openai.types.realtime.realtime_session_create_request import (
     RealtimeSessionCreateRequest,
 )
+from openai.types.realtime.session_update_event import SessionUpdateEvent
 
 from agents.handoffs import Handoff
 from agents.realtime.agent import RealtimeAgent
@@ -14,6 +15,7 @@
 from agents.realtime.model import RealtimeModelConfig
 from agents.realtime.openai_realtime import (
     OpenAIRealtimeSIPModel,
+    OpenAIRealtimeWebSocketModel,
     _build_model_settings_from_agent,
     _collect_enabled_handoffs,
 )
@@ -129,3 +131,36 @@ def ping() -> str:
             tool_names.add(name)
     assert ping.name in tool_names
     assert f"transfer_to_{child_agent.name}" in tool_names
+
+
+def test_call_id_session_update_omits_null_audio_formats() -> None:
+    model = OpenAIRealtimeWebSocketModel()
+    model._call_id = "call_123"
+
+    session_config = model._get_session_config({})
+    payload = SessionUpdateEvent(type="session.update", session=session_config).model_dump(
+        exclude_unset=True
+    )
+
+    audio = payload["session"]["audio"]
+    assert "format" not in audio["input"]
+    assert "format" not in audio["output"]
+
+
+def test_call_id_session_update_includes_explicit_audio_formats() -> None:
+    model = OpenAIRealtimeWebSocketModel()
+    model._call_id = "call_123"
+
+    session_config = model._get_session_config(
+        {
+            "input_audio_format": "g711_ulaw",
+            "output_audio_format": "g711_ulaw",
+        }
+    )
+    payload = SessionUpdateEvent(type="session.update", session=session_config).model_dump(
+        exclude_unset=True
+    )
+
+    audio = payload["session"]["audio"]
+    assert audio["input"]["format"]["type"] == "audio/pcmu"
+    assert audio["output"]["format"]["type"] == "audio/pcmu"