LLMUserContextAggregator: don't reset timer with interim transcription

It turns out that in some cases we only get interim transcriptions (e.g. someone
is speaking very very softly or someone is talking in the background). In those
cases we don't want to interrupt the bot because there's really nothing to
interrupt the bot for.

We originally thought we should interrupt the bot right at the time we got an
interim frame, but this is causing too many false positives. It's actually
better to simply wait for a real transcription before interrupting (in case VAD
didn't interrupt).

CHANGELOG.md

@@ -36,7 +36,12 @@ stt = DeepgramSTTService(..., live_options=LiveOptions(model="nova-2-general"))
 
 ### Fixed
 
-- Fixed an issue where `EndTaskFrame` was not triggering `on_client_disconnected` or closing the WebSocket in FastAPI.
+- Fixed an issue that would cause undesired interruptions via
+  `EmulateUserStartedSpeakingFrame` when only interim transcriptions (i.e. no
+  final transcriptions) where received.
+
+- Fixed an issue where `EndTaskFrame` was not triggering
+  `on_client_disconnected` or closing the WebSocket in FastAPI.
 
 - Fixed a context aggregator issue that would not append the LLM text response
   to the context if a function call happened in the same LLM turn.

src/pipecat/processors/aggregators/llm_response.py

@@ -301,8 +301,6 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
 
     async def _handle_interim_transcription(self, _: InterimTranscriptionFrame):
         self._seen_interim_results = True
-        # Reset aggregation timer.
-        self._aggregation_event.set()
 
     def _create_aggregation_task(self):
         self._aggregation_task = self.create_task(self._aggregation_task_handler())