Starting to create a local smart turn

examples/foundational/38a-local-smart-turn.py

@@ -0,0 +1,108 @@
+#
+# Copyright (c) 2024–2025, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+import os
+
+from dotenv import load_dotenv
+from loguru import logger
+
+from pipecat.audio.turn.local_smart_turn import LocalSmartTurnAnalyzer
+from pipecat.audio.vad.silero import SileroVADAnalyzer
+from pipecat.pipeline.pipeline import Pipeline
+from pipecat.pipeline.runner import PipelineRunner
+from pipecat.pipeline.task import PipelineParams, PipelineTask
+from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
+from pipecat.services.cartesia.tts import CartesiaTTSService
+from pipecat.services.deepgram.stt import DeepgramSTTService
+from pipecat.services.openai.llm import OpenAILLMService
+from pipecat.transports.base_transport import TransportParams
+from pipecat.transports.network.small_webrtc import SmallWebRTCTransport
+from pipecat.transports.network.webrtc_connection import SmallWebRTCConnection
+
+load_dotenv(override=True)
+
+
+async def run_bot(webrtc_connection: SmallWebRTCConnection):
+    logger.info(f"Starting bot")
+
+    transport = SmallWebRTCTransport(
+        webrtc_connection=webrtc_connection,
+        params=TransportParams(
+            audio_in_enabled=True,
+            audio_out_enabled=True,
+            vad_enabled=True,
+            vad_analyzer=SileroVADAnalyzer(),
+            vad_audio_passthrough=True,
+            end_of_turn_analyzer=LocalSmartTurnAnalyzer(),
+        ),
+    )
+
+    stt = DeepgramSTTService(api_key=os.getenv("DEEPGRAM_API_KEY"))
+
+    tts = CartesiaTTSService(
+        api_key=os.getenv("CARTESIA_API_KEY"),
+        voice_id="71a7ad14-091c-4e8e-a314-022ece01c121",  # British Reading Lady
+    )
+
+    llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")
+
+    messages = [
+        {
+            "role": "system",
+            "content": "You are a helpful LLM in a WebRTC call. Your goal is to demonstrate your capabilities in a succinct way. Your output will be converted to audio so don't include special characters in your answers. Respond to what the user said in a creative and helpful way.",
+        },
+    ]
+
+    context = OpenAILLMContext(messages)
+    context_aggregator = llm.create_context_aggregator(context)
+
+    pipeline = Pipeline(
+        [
+            transport.input(),  # Transport user input
+            stt,
+            context_aggregator.user(),  # User responses
+            llm,  # LLM
+            tts,  # TTS
+            transport.output(),  # Transport bot output
+            context_aggregator.assistant(),  # Assistant spoken responses
+        ]
+    )
+
+    task = PipelineTask(
+        pipeline,
+        params=PipelineParams(
+            allow_interruptions=True,
+            enable_metrics=True,
+            enable_usage_metrics=True,
+            report_only_initial_ttfb=True,
+        ),
+    )
+
+    @transport.event_handler("on_client_connected")
+    async def on_client_connected(transport, client):
+        logger.info(f"Client connected")
+        # Kick off the conversation.
+        messages.append({"role": "system", "content": "Please introduce yourself to the user."})
+        await task.queue_frames([context_aggregator.user().get_context_frame()])
+
+    @transport.event_handler("on_client_disconnected")
+    async def on_client_disconnected(transport, client):
+        logger.info(f"Client disconnected")
+
+    @transport.event_handler("on_client_closed")
+    async def on_client_closed(transport, client):
+        logger.info(f"Client closed connection")
+        await task.cancel()
+
+    runner = PipelineRunner(handle_sigint=False)
+
+    await runner.run(task)
+
+
+if __name__ == "__main__":
+    from run import main
+
+    main()

src/pipecat/audio/turn/local_smart_turn.py

@@ -0,0 +1,51 @@
+#
+# Copyright (c) 2024–2025, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+
+import numpy as np
+from loguru import logger
+
+from pipecat.audio.turn.base_turn_analyzer import BaseEndOfTurnAnalyzer, EndOfTurnState
+
+
+class LocalSmartTurnAnalyzer(BaseEndOfTurnAnalyzer):
+    def __init__(self):
+        super().__init__()
+        self._audio_buffer = bytearray()
+
+        logger.debug("Loading Local Smart Turn model...")
+
+        # TODO: implement it
+
+        logger.debug("Loaded Local Smart Turn")
+
+    def analyze_audio(self, buffer: bytes) -> EndOfTurnState:
+        self._audio_buffer += buffer
+
+        # TODO: we probably don't need this
+        # Checking if we have at least 6 seconds of audio
+        # if len(self._audio_buffer) < 16000 * 2 * 6:
+        #    return EndOfTurnState.INCOMPLETE
+
+        audio_int16 = np.frombuffer(self._audio_buffer, dtype=np.int16)
+
+        # Divide by 32768 because we have signed 16-bit data.
+        audio_float32 = np.frombuffer(audio_int16, dtype=np.int16).astype(np.float32) / 32768.0
+
+        # TODO: implement to use the smart turn
+        # for now it is always returning as complete only for testing it
+        prediction = 1
+
+        state = EndOfTurnState.COMPLETE if prediction == 1 else EndOfTurnState.INCOMPLETE
+
+        if state == EndOfTurnState.COMPLETE:
+            # clears the buffer completely
+            self._audio_buffer = bytearray()
+        else:
+            # TODO: implement it
+            pass
+
+        return state

src/pipecat/transports/base_input.py

@@ -172,9 +172,16 @@ class BaseInputTransport(FrameProcessor):
         elif isinstance(frame, UserStoppedSpeakingFrame):
             logger.debug("User stopped speaking")
             await self.push_frame(frame)
+
+            # TODO check, we probably should change here as well.
+            # if the end of turn is enabled, we should only stop interruption after this point
             if self.interruptions_allowed:
                 await self._stop_interruption()
                 await self.push_frame(StopInterruptionFrame())
+        elif isinstance(frame, UserEndOfTurnFrame):
+            logger.debug("User end of turn")
+            await self.push_frame(frame)
+            # TODO: implement to handle interruptions
 
     #
     # Audio input
@@ -220,7 +227,7 @@ class BaseInputTransport(FrameProcessor):
     ):
         new_eot_state = await self._end_of_turn_analyze(audio_frame)
         if new_eot_state != end_of_turn_state:
-            await self.push_frame(UserEndOfTurnFrame())
+            await self._handle_user_interruption(UserEndOfTurnFrame())
         return new_eot_state
 
     async def _audio_task_handler(self):
@@ -239,9 +246,13 @@ class BaseInputTransport(FrameProcessor):
             # changes from QUIET to SPEAKING and vice versa.
             if self._params.vad_enabled:
                 vad_state = await self._handle_vad(frame, vad_state)
+                # TODO: need to check if we need to keep it later
+                if vad_state == VADState.QUIET:
+                    end_of_turn_state = EndOfTurnState.INCOMPLETE
                 audio_passthrough = self._params.vad_audio_passthrough
 
-            if self._params.end_of_turn_analyzer:
+            # We only need to check for completion if the user is speaking
+            if self._params.end_of_turn_analyzer and VADState.QUIET != vad_state:
                 end_of_turn_state = await self._handle_end_of_turn(frame, end_of_turn_state)
 
             # Push audio downstream if passthrough.