Preventing emitting the UserStartedSpeaking event multiple times.

src/pipecat/audio/turn/base_smart_turn.py

@@ -57,6 +57,10 @@ class BaseSmartTurn(ABC):
     def set_sample_rate(self, sample_rate: int):
         self._sample_rate = sample_rate
 
+    @property
+    def speech_triggered(self) -> bool:
+        return self._speech_triggered
+
     def append_audio(self, buffer: bytes, is_speech: bool) -> EndOfTurnState:
         # Convert raw audio to float32 format and append to the buffer
         audio_int16 = np.frombuffer(buffer, dtype=np.int16)

src/pipecat/transports/base_input.py

@@ -196,12 +196,18 @@ class BaseInputTransport(FrameProcessor):
             and new_vad_state != VADState.STOPPING
         ):
             frame = None
-            if new_vad_state == VADState.SPEAKING:
-                frame = UserStartedSpeakingFrame()
-            # TODO: need to double check if this is the expected behavior
-            # Not triggering the UserStoppedSpeakingFrame if the turn analyser is enabled
-            elif new_vad_state == VADState.QUIET and not self.end_of_turn_analyzer:
-                frame = UserStoppedSpeakingFrame()
+            # If the turn analyser is enabled, this will prevent:
+            # - Creating the UserStoppedSpeakingFrame
+            # - Creating the UserStartedSpeakingFrame multiple times
+            can_create_user_frames = (
+                self._params.end_of_turn_analyzer is None
+                or not self._params.end_of_turn_analyzer.speech_triggered
+            )
+            if can_create_user_frames:
+                if new_vad_state == VADState.SPEAKING:
+                    frame = UserStartedSpeakingFrame()
+                elif new_vad_state == VADState.QUIET:
+                    frame = UserStoppedSpeakingFrame()
 
             if frame:
                 await self._handle_user_interruption(frame)