Merge pull request #3391 from pipecat-ai/filipi/krisp_followup_improvements

Krisp VIVA follow-up improvements

changelog/3391.added.md

@@ -0,0 +1 @@
+- Added `KrispVivaTurn` analyzer for end of turn detection using the Krisp VIVA SDK (requires `krisp_audio`).

changelog/3391.changed.md

@@ -0,0 +1 @@
+- Improved Krisp SDK management to allow `KrispVivaTurn` and `KrispVivaFilter` to share a single SDK instance within the same process.

changelog/3391.fixed.md

@@ -0,0 +1 @@
+- Fixed potential memory leaks and initialization issues in `KrispVivaFilter` by improving SDK lifecycle management.

examples/foundational/07p-interruptible-krisp-viva.py

@@ -4,6 +4,23 @@
 # SPDX-License-Identifier: BSD 2-Clause License
 #
 
+"""Interruptible bot with Krisp VIVA noise filtering and turn detection.
+
+This example demonstrates a conversational bot with:
+- Krisp VIVA noise reduction on incoming audio
+- Krisp VIVA Turn detection for natural interruptions
+- Voice activity detection (VAD)
+
+Required environment variables:
+- KRISP_VIVA_FILTER_MODEL_PATH: Path to the Krisp noise filter model file (.kef)
+- KRISP_VIVA_TURN_MODEL_PATH: Path to the Krisp turn detection model file (.kef)
+- DEEPGRAM_API_KEY: Deepgram API key for STT/TTS
+- OPENAI_API_KEY: OpenAI API key for LLM
+
+Optional environment variables:
+- KRISP_NOISE_SUPPRESSION_LEVEL: Noise suppression level 0-100 (default: 100)
+  Higher values = more aggressive noise reduction
+"""
 
 import os
 
@@ -11,7 +28,7 @@ from dotenv import load_dotenv
 from loguru import logger
 
 from pipecat.audio.filters.krisp_viva_filter import KrispVivaFilter
-from pipecat.audio.turn.smart_turn.local_smart_turn_v3 import LocalSmartTurnAnalyzerV3
+from pipecat.audio.turn.krisp_viva_turn import KrispTurnParams, KrispVivaTurn
 from pipecat.audio.vad.silero import SileroVADAnalyzer
 from pipecat.audio.vad.vad_analyzer import VADParams
 from pipecat.frames.frames import LLMRunFrame
@@ -82,7 +99,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         context,
         user_params=LLMUserAggregatorParams(
             user_turn_strategies=UserTurnStrategies(
-                stop=[TurnAnalyzerUserTurnStopStrategy(turn_analyzer=LocalSmartTurnAnalyzerV3())]
+                stop=[TurnAnalyzerUserTurnStopStrategy(turn_analyzer=KrispVivaTurn())]
             ),
         ),
     )

src/pipecat/audio/turn/base_turn_analyzer.py

@@ -132,3 +132,7 @@ class BaseTurnAnalyzer(ABC):
     def clear(self):
         """Reset the turn analyzer to its initial state."""
         pass
+
+    async def cleanup(self):
+        """Cleanup the turn analyzer."""
+        pass

src/pipecat/audio/turn/krisp_viva_turn.py

@@ -133,7 +133,7 @@ class KrispVivaTurn(BaseTurnAnalyzer):
                 self._sdk_acquired = False
             raise
 
-    def __del__(self):
+    async def cleanup(self):
         """Release SDK reference when analyzer is destroyed."""
         if self._sdk_acquired:
             try:
@@ -192,8 +192,7 @@ class KrispVivaTurn(BaseTurnAnalyzer):
         # Create session when sample rate is set
         try:
             self._tt_session = self._create_tt_session(self._sample_rate)
-            # Clear buffer when sample rate changes
-            self._audio_buffer.clear()
+            self.clear()
         except Exception as e:
             logger.error(f"Failed to create turn detection session: {e}", exc_info=True)
             self._tt_session = None
@@ -310,7 +309,7 @@ class KrispVivaTurn(BaseTurnAnalyzer):
                 # confirms with sufficient confidence
                 if self._speech_triggered and prob >= self._params.threshold:
                     state = EndOfTurnState.COMPLETE
-                    self._clear(state)
+                    self.clear()
                     break
 
             # Store the last state for analyze_end_of_turn()
@@ -336,18 +335,6 @@ class KrispVivaTurn(BaseTurnAnalyzer):
 
     def clear(self):
         """Reset the turn analyzer to its initial state."""
-        self._clear(EndOfTurnState.COMPLETE)
-
-    def _clear(self, turn_state: EndOfTurnState):
-        """Clear internal state based on turn completion status.
-
-        Args:
-            turn_state: The end-of-turn state to use for clearing.
-        """
-        # If the state is still incomplete, keep the _speech_triggered as True
-        self._speech_triggered = turn_state == EndOfTurnState.INCOMPLETE
-        # Clear audio buffer on turn completion
-        if turn_state == EndOfTurnState.COMPLETE:
-            self._audio_buffer.clear()
-        # Reset last state when clearing
+        self._speech_triggered = False
+        self._audio_buffer.clear()
         self._last_state = EndOfTurnState.INCOMPLETE

src/pipecat/turns/user_stop/turn_analyzer_user_turn_stop_strategy.py

@@ -73,6 +73,7 @@ class TurnAnalyzerUserTurnStopStrategy(BaseUserTurnStopStrategy):
     async def cleanup(self):
         """Cleanup the strategy."""
         await super().cleanup()
+        await self._turn_analyzer.cleanup()
         if self._task:
             await self.task_manager.cancel_task(self._task)
             self._task = None