Smart Turn analyzer now uses the full context of the turn rather than just the audio since VAD last triggered (fixes #3094)

CHANGELOG.md

@@ -40,6 +40,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Updated Smart Turn model weights to v3.1.
 
+- Smart Turn analyzer now uses the full context of the turn rather than just the
+  audio since VAD last triggered.
+
 ### Deprecated
 
 - Package `pipecat.sync` is deprecated, use `pipecat.utils.sync` instead.

src/pipecat/audio/turn/smart_turn/base_smart_turn.py

@@ -28,7 +28,6 @@ from pipecat.metrics.metrics import MetricsData, SmartTurnMetricsData
 STOP_SECS = 3
 PRE_SPEECH_MS = 0
 MAX_DURATION_SECONDS = 8  # Max allowed segment duration
-USE_ONLY_LAST_VAD_SEGMENT = True
 
 
 class SmartTurnParams(BaseTurnParams):
@@ -43,8 +42,6 @@ class SmartTurnParams(BaseTurnParams):
     stop_secs: float = STOP_SECS
     pre_speech_ms: float = PRE_SPEECH_MS
     max_duration_secs: float = MAX_DURATION_SECONDS
-    # not exposing this for now yet until the model can handle it.
-    # use_only_last_vad_segment: bool = USE_ONLY_LAST_VAD_SEGMENT
 
 
 class SmartTurnTimeoutException(Exception):
@@ -160,7 +157,7 @@ class BaseSmartTurn(BaseTurnAnalyzer):
         state, result = await loop.run_in_executor(
             self._executor, self._process_speech_segment, self._audio_buffer
         )
-        if state == EndOfTurnState.COMPLETE or USE_ONLY_LAST_VAD_SEGMENT:
+        if state == EndOfTurnState.COMPLETE:
             self._clear(state)
         logger.debug(f"End of Turn result: {state}")
         return state, result