Updated Speechmatics STT integration

examples/voice_agents/email_example.py

@@ -55,7 +55,7 @@ async def register_for_event(self, context: RunContext):
 async def entrypoint(ctx: JobContext):
     session = AgentSession(
         vad=silero.VAD.load(),
-        llm=inference.LLM("google/gemini-2.5-flash"),
+        llm=inference.LLM("openai/gpt-4.1-mini"),
         stt=inference.STT("deepgram/nova-3"),
         tts=inference.TTS("cartesia/sonic-3"),
     )

livekit-agents/livekit/agents/llm/chat_context.py

@@ -302,6 +302,7 @@ def copy(
         exclude_instructions: bool = False,
         exclude_empty_message: bool = False,
         exclude_handoff: bool = False,
+        exclude_config_update: bool = False,
         tools: NotGivenOr[Sequence[Tool | Toolset | str]] = NOT_GIVEN,
     ) -> ChatContext:
         items = []
@@ -343,6 +344,9 @@ def get_tool_names(
             if exclude_handoff and item.type == "agent_handoff":
                 continue
 
+            if exclude_config_update and item.type == "agent_config_update":
+                continue
+
             if (
                 is_given(tools)
                 and (item.type == "function_call" or item.type == "function_call_output")
@@ -389,6 +393,7 @@ def merge(
         *,
         exclude_function_call: bool = False,
         exclude_instructions: bool = False,
+        exclude_config_update: bool = False,
     ) -> ChatContext:
         """Add messages from `other_chat_ctx` into this one, avoiding duplicates, and keep items sorted by created_at."""
         existing_ids = {item.id for item in self._items}
@@ -407,6 +412,9 @@ def merge(
             ):
                 continue
 
+            if exclude_config_update and item.type == "agent_config_update":
+                continue
+
             if item.id not in existing_ids:
                 idx = self.find_insertion_index(created_at=item.created_at)
                 self._items.insert(idx, item)
@@ -422,6 +430,7 @@ def to_dict(
         exclude_timestamp: bool = True,
         exclude_function_call: bool = False,
         exclude_metrics: bool = False,
+        exclude_config_update: bool = False,
     ) -> dict[str, Any]:
         items: list[ChatItem] = []
         for item in self.items:
@@ -431,6 +440,9 @@ def to_dict(
             ]:
                 continue
 
+            if exclude_config_update and item.type == "agent_config_update":
+                continue
+
             if item.type == "message":
                 item = item.model_copy()
                 if exclude_image:

livekit-agents/livekit/agents/llm/realtime.py

@@ -182,10 +182,6 @@ def generate_reply(
     @abstractmethod
     def commit_audio(self) -> None: ...
 
-    # commit the user turn to the server
-    @abstractmethod
-    def commit_user_turn(self) -> None: ...
-
     # clear the input audio buffer to the server
     @abstractmethod
     def clear_audio(self) -> None: ...

livekit-agents/livekit/agents/stt/stt.py

@@ -64,6 +64,7 @@ class SpeechData:
 @dataclass
 class RecognitionUsage:
     audio_duration: float
+    """Incremental audio duration/usage in seconds"""
 
 
 @dataclass

livekit-agents/livekit/agents/voice/agent.py

@@ -756,6 +756,18 @@ def _handle_task_done(_: asyncio.Task[Any]) -> None:
         old_agent = old_activity.agent
         session = old_activity.session
 
+        old_allow_interruptions = True
+        if speech_handle:
+            if speech_handle.interrupted:
+                raise RuntimeError(
+                    f"{self.__class__.__name__} cannot be awaited inside a function tool that is already interrupted"
+                )
+
+            # lock the speech handle to prevent interruptions until the task is complete
+            # there should be no await before this line to avoid race conditions
+            old_allow_interruptions = speech_handle.allow_interruptions
+            speech_handle.allow_interruptions = False
+
         blocked_tasks = [current_task]
         if (
             old_activity._on_enter_task
@@ -790,6 +802,9 @@ def _handle_task_done(_: asyncio.Task[Any]) -> None:
             return await asyncio.shield(self.__fut)
 
         finally:
+            if speech_handle:
+                speech_handle.allow_interruptions = old_allow_interruptions
+
             # run_state could have changed after self.__fut
             run_state = session._global_run_state