From b075502c4c6908b89359c7798a14dd8f38422c08 Mon Sep 17 00:00:00 2001
From: marcus-daily <111281783+marcus-daily@users.noreply.github.com>
Date: Fri, 9 Jan 2026 13:44:44 +0000
Subject: [PATCH] Addressing code review comments

---
 src/pipecat/audio/turn/base_turn_analyzer.py   | 12 ++++++++++++
 .../audio/turn/smart_turn/base_smart_turn.py   | 10 +++++-----
 .../turn_analyzer_user_turn_stop_strategy.py   | 18 +++++-------------
 3 files changed, 22 insertions(+), 18 deletions(-)

diff --git a/src/pipecat/audio/turn/base_turn_analyzer.py b/src/pipecat/audio/turn/base_turn_analyzer.py
index 64510b518..942214738 100644
--- a/src/pipecat/audio/turn/base_turn_analyzer.py
+++ b/src/pipecat/audio/turn/base_turn_analyzer.py
@@ -116,6 +116,18 @@ class BaseTurnAnalyzer(ABC):
         """
         pass
 
+    def on_vad_start_secs_updated(self, vad_start_secs: float):
+        """Invoked when the VAD start trigger time is updated.
+
+        The turn analyzer may choose to change its buffer size depending
+        on this value.
+
+        Args:
+            vad_start_secs (float): The number of seconds of voice activity
+                before triggering the user speaking event.
+        """
+        pass
+
     @abstractmethod
     def clear(self):
         """Reset the turn analyzer to its initial state."""
diff --git a/src/pipecat/audio/turn/smart_turn/base_smart_turn.py b/src/pipecat/audio/turn/smart_turn/base_smart_turn.py
index 11d3573bf..e71731446 100644
--- a/src/pipecat/audio/turn/smart_turn/base_smart_turn.py
+++ b/src/pipecat/audio/turn/smart_turn/base_smart_turn.py
@@ -35,15 +35,11 @@ class SmartTurnParams(BaseTurnParams):
     Parameters:
         stop_secs: Maximum silence duration in seconds before ending turn.
         pre_speech_ms: Milliseconds of audio to include before speech starts.
-        vad_start_secs: Seconds VAD waits before confirming speech start (e.g. VAD STARTING window).
-            This is added to `pre_speech_ms` at inference slicing time so Smart Turn can include
-            the initial audio that occurred while VAD was still confirming speech.
         max_duration_secs: Maximum duration in seconds for audio segments.
     """
 
     stop_secs: float = STOP_SECS
     pre_speech_ms: float = PRE_SPEECH_MS
-    vad_start_secs: float = 0.0
     max_duration_secs: float = MAX_DURATION_SECONDS
 
 
@@ -82,6 +78,7 @@ class BaseSmartTurn(BaseTurnAnalyzer):
         # Thread executor that will run the model. We only need one thread per
         # analyzer because one analyzer just handles one audio stream.
         self._executor = ThreadPoolExecutor(max_workers=1)
+        self._vad_start_secs: float = 0.0
 
     @property
     def speech_triggered(self) -> bool:
@@ -165,6 +162,9 @@ class BaseSmartTurn(BaseTurnAnalyzer):
         logger.debug(f"End of Turn result: {state}")
         return state, result
 
+    def on_vad_start_secs_updated(self, vad_start_secs: float):
+        self._vad_start_secs = vad_start_secs
+
     def clear(self):
         """Reset the turn analyzer to its initial state."""
         self._clear(EndOfTurnState.COMPLETE)
@@ -185,7 +185,7 @@ class BaseSmartTurn(BaseTurnAnalyzer):
             return state, None
 
         # Extract recent audio segment for prediction
-        effective_pre_speech_ms = self._params.pre_speech_ms + (self._params.vad_start_secs * 1000)
+        effective_pre_speech_ms = self._params.pre_speech_ms + (self._vad_start_secs * 1000)
         start_time = self._speech_start_time - (effective_pre_speech_ms / 1000)
         start_index = 0
         for i, (t, _) in enumerate(audio_buffer):
diff --git a/src/pipecat/turns/user_stop/turn_analyzer_user_turn_stop_strategy.py b/src/pipecat/turns/user_stop/turn_analyzer_user_turn_stop_strategy.py
index 28fa9e761..16802795a 100644
--- a/src/pipecat/turns/user_stop/turn_analyzer_user_turn_stop_strategy.py
+++ b/src/pipecat/turns/user_stop/turn_analyzer_user_turn_stop_strategy.py
@@ -108,20 +108,12 @@ class TurnAnalyzerUserTurnStopStrategy(BaseUserTurnStopStrategy):
     async def _handle_speech_control_params(self, frame: SpeechControlParamsFrame):
         """Sync Smart Turn pre-speech buffering with VAD start delay.
 
-        In the new user-turn-strategies pipeline, `VADUserStartedSpeakingFrame`
-        is emitted only once VAD has *confirmed* speech (after `vad_params.start_secs`).
-        Smart Turn should still include the initial audio collected during that
-        confirmation window, so we record it in `SmartTurnParams.vad_start_secs` and
-        add it at inference slicing time (preserving `pre_speech_ms` semantics).
+        `VADUserStartedSpeakingFrame` is emitted only once VAD has confirmed speech
+        (after `vad_params.start_secs`). Smart Turn should still include the initial
+        audio collected during that confirmation window, so we let the analyzer know
+        when this value has changed.
         """
-        if not frame.vad_params:
-            return
-
-        params = self._turn_analyzer.params
-        if not isinstance(params, SmartTurnParams):
-            return
-
-        params.vad_start_secs = frame.vad_params.start_secs
+        self._turn_analyzer.on_vad_start_secs_updated(frame.vad_params.start_secs)
 
     async def _handle_input_audio(self, frame: InputAudioRawFrame):
         """Handle input audio to check if the turn is completed."""