Handle "response.output_audio.done" in openai.Realtime to emit end of output

Author: dangusev

plugins/openai/tests/test_openai_realtime.py

@@ -10,6 +10,7 @@
 from vision_agents.core.llm.realtime import (
     RealtimeAgentTranscript,
     RealtimeAudioOutput,
+    RealtimeAudioOutputDone,
     RealtimeUserTranscript,
 )
 from vision_agents.plugins.openai import Realtime
@@ -127,10 +128,12 @@ async def test_simple_response_flow(self, realtime):
             pass
 
         await asyncio.sleep(3.0)
-        audio = [
-            i for i in realtime.output.peek() if isinstance(i, RealtimeAudioOutput)
-        ]
+        items = realtime.output.peek()
+        audio = [i for i in items if isinstance(i, RealtimeAudioOutput)]
+        done = [i for i in items if isinstance(i, RealtimeAudioOutputDone)]
         assert len(audio) > 0
+        assert len(done) >= 1
+        assert any(not d.interrupted for d in done)
 
     async def test_audio_sending_flow(self, realtime, mia_audio_16khz):
         # Wait for connection to be fully established

plugins/openai/vision_agents/plugins/openai/openai_realtime.py

@@ -336,12 +336,9 @@ async def _handle_openai_event(self, event: dict) -> None:
         elif et == "output_audio_buffer.stopped":
             # Output audio buffer stopped - acknowledgment of audio playback end
             pass
-        elif et == "response.audio.done":
-            # Audio generation complete for this response item
-            self._emit_audio_output_done_event()
-        elif et == "response.output_audio.done":
-            # Output audio generation complete for this response item
-            pass
+        elif et in ("response.output_audio.done", "response.audio.done"):
+            # Audio generation complete for this response item.
+            self._emit_audio_output_done_event(response_id=event.get("response_id"))
         elif et == "response.content_part.done":
             # Content part complete - contains full transcript
             pass