dv-pipecat-ai 0.0.85.dev830__py3-none-any.whl → 0.0.85.dev837__py3-none-any.whl

{dv_pipecat_ai-0.0.85.dev830.dist-info → dv_pipecat_ai-0.0.85.dev837.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-ai
-Version: 0.0.85.dev830
+Version: 0.0.85.dev837
 Summary: An open source framework for voice (and multimodal) assistants
 License-Expression: BSD-2-Clause
 Project-URL: Source, https://github.com/pipecat-ai/pipecat

{dv_pipecat_ai-0.0.85.dev830.dist-info → dv_pipecat_ai-0.0.85.dev837.dist-info}/RECORD

@@ -1,4 +1,4 @@
-dv_pipecat_ai-0.0.85.dev830.dist-info/licenses/LICENSE,sha256=DWY2QGf2eMCFhuu2ChairtT6CB7BEFffNVhXWc4Od08,1301
+dv_pipecat_ai-0.0.85.dev837.dist-info/licenses/LICENSE,sha256=DWY2QGf2eMCFhuu2ChairtT6CB7BEFffNVhXWc4Od08,1301
 pipecat/__init__.py,sha256=j0Xm6adxHhd7D06dIyyPV_GlBYLlBnTAERVvD_jAARQ,861
 pipecat/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/adapters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -107,7 +107,7 @@ pipecat/pipeline/to_be_updated/merge_pipeline.py,sha256=jLEWdufIW3z1xZhdoLowdJ_S
 pipecat/processors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/processors/async_generator.py,sha256=qPOZxk5eOad_NrF_Z06vWZ6deXIxb9AKZKYO2e5pkJs,2385
 pipecat/processors/consumer_processor.py,sha256=DrWCKnfblknZJ0bLmR_unIeJ1axQw4IPUn2IB3KLGGA,3228
-pipecat/processors/dtmf_aggregator.py,sha256=mo_IXUlsnVl-_Xn8sbTGnRF4Lkts0h6E3uauGbeFyWs,10204
+pipecat/processors/dtmf_aggregator.py,sha256=k3xYncUr_8y5lrYfeX8PxqlF7jqFLshg_HB6HiFg7TA,10193
 pipecat/processors/frame_processor.py,sha256=uBu6Waa0_diMXdQXMZ5V5a_KwaaPzcieyuv5gO9u-ME,33841
 pipecat/processors/idle_frame_processor.py,sha256=z8AuhGap61lA5K35P6XCaOpn4kkmK_9NZNppbpQxheU,3124
 pipecat/processors/logger.py,sha256=8xa4KKekXQIETlQR7zoGnwUpLNo8CeDVm7YjyXePN-w,2385
@@ -122,7 +122,7 @@ pipecat/processors/aggregators/gated.py,sha256=tii0sRrBkRW6y9Xq5iTWPnqlOEejU4VqP
 pipecat/processors/aggregators/gated_llm_context.py,sha256=CPv6sMA8irD1zZ3fU1gSv6D7qcPvCA0MdpFhBtJ_ekI,3007
 pipecat/processors/aggregators/gated_open_ai_llm_context.py,sha256=DgqmdPj1u3fP_SVmxtfP7NjHqnyhN_RVVTDfmjbkxAs,361
 pipecat/processors/aggregators/llm_context.py,sha256=wNbZA0Vt0FzNc5cu06xiv1z7DIClIlfqR1ZD8EusbVw,11085
-pipecat/processors/aggregators/llm_response.py,sha256=V6wBTzfUGLJfMuI34fkf5VTR0I66AWIW8btxKI8_3IM,48795
+pipecat/processors/aggregators/llm_response.py,sha256=--6D736k5mNnIhmauRbA7ZG7H9tBR16okniz3Mpypns,48573
 pipecat/processors/aggregators/llm_response_universal.py,sha256=5PqmpATpekD8BVWyBExZgatKHsNbZem8M-A7_VwTbiQ,34334
 pipecat/processors/aggregators/openai_llm_context.py,sha256=cC8DXdVPERRN04i0i-1Ys6kusvnbMALeH-Z8Pu5K684,12999
 pipecat/processors/aggregators/sentence.py,sha256=E7e3knfQl6HEGpYMKPklF1aO_gOn-rr7SnynErwfkQk,2235
@@ -153,17 +153,18 @@ pipecat/runner/livekit.py,sha256=in-2Io3FUZV-VcZZ-gQCx9L1WnKp5sHqmm7tDYlFNl4,458
 pipecat/runner/run.py,sha256=McalzMoFYEJJRXyoD5PBAyUhHCdsEeeZJk8lBvplRck,30054
 pipecat/runner/types.py,sha256=zHjbAiU17fG0ypLXCEzPu7bpDOutAg-4gE7TESvK8n0,1761
 pipecat/runner/utils.py,sha256=Ve9rjRvbt1o8e9by0nIrCJzUDGcuJUeYYhkqycmgHXc,18682
-pipecat/serializers/__init__.py,sha256=xcmbbR7YYU5C4HPbo2WVgPij-Bl_qlrLcnunCdpcZkg,804
-pipecat/serializers/asterisk.py,sha256=bPuGuLiCf04_H0d9Gc-5BpEtqD9BRNWnpZZq5MZ1fDY,6091
+pipecat/serializers/__init__.py,sha256=z0V5GflCoPt4k2Yqm4ivuzKDh9VsYYAgK2UXZTw10aU,863
+pipecat/serializers/asterisk.py,sha256=QLJMXkU3DZ0sgFw3Vq2Zf8PHKkQQguL_v-l2Io4lZ_M,6729
 pipecat/serializers/base_serializer.py,sha256=OyBUZccs2ZT9mfkBbq2tGsUJMvci6o-j90Cl1sicPaI,2030
-pipecat/serializers/convox.py,sha256=Irby_iZywgBtevlxiC8nE2GY3eh4yNNRi2YC-0vnNTY,11155
-pipecat/serializers/custom.py,sha256=O0gHTyoSb1AZ_tEmE9VgRViYckmsNzjwCAqt-Xc2CaM,9081
+pipecat/serializers/convox.py,sha256=fj9NkFTB74B9k8qWEuICQNGUQtEV0DusaHohkOqNLa8,11145
+pipecat/serializers/custom.py,sha256=clUEqOazGe3B2XoUFRN9zkFpMd6aIZeVRTqBRHAzavM,9071
 pipecat/serializers/exotel.py,sha256=B04LtNnRMzKmaS61gPZbUjc2nbki3FmpCfUMww6cOe4,5953
 pipecat/serializers/livekit.py,sha256=OMaM7yUiHfeTPbpNxE2TrmIzjmbNQIjNvlujt81dsRI,3285
 pipecat/serializers/plivo.py,sha256=ie6VUhZDTJ7KlAuJyHNeIeMtJ3ScDq_2js1SZtz7jLI,9256
 pipecat/serializers/protobuf.py,sha256=L0jSqvgTdkfxsu6JWjYK8QSTVji9nhzmgRsEEbGU7xY,5223
 pipecat/serializers/telnyx.py,sha256=eFkC7dExDFildYLR8DPvgfHbgXlCwdSPd1vc11yxyok,10847
 pipecat/serializers/twilio.py,sha256=0emSzXVw8DU_N5RPruMekbBKku9Q429-0z1PMuYejSk,10823
+pipecat/serializers/vi.py,sha256=Q7kMXvKM493RIuOUc99LKZWgVmvd8_owAzIK_oEktfw,11150
 pipecat/services/__init__.py,sha256=8e3Ta-8_BOPozhDB3l0GJkNXs5PWhib6yqZQUof2Kvw,1209
 pipecat/services/ai_service.py,sha256=yE386fm2Id-yD4fCNfkmEMtg0lTA7PB17n2x_A_jwTg,5896
 pipecat/services/ai_services.py,sha256=_RrDWfM8adV17atzY9RxK0nXRVM5kbUkKrvN90GAWYM,795
@@ -205,7 +206,7 @@ pipecat/services/azure/realtime/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm
 pipecat/services/azure/realtime/llm.py,sha256=MnDiw-YJP3kll1gbkta4z4vsWfWZ5oBprZCinMP9O0M,2385
 pipecat/services/cartesia/__init__.py,sha256=vzh0jBnfPwWdxFfV-tu0x1HFoOTgr9s91GYmD-CJUtY,284
 pipecat/services/cartesia/stt.py,sha256=00k9gQYo_xPKb-RRJ-RNV4LPFw-7xXiFU7ACFLYttWY,12388
-pipecat/services/cartesia/tts.py,sha256=EdpVJoDhZn7N5hj-VDsCaO-W2MsA78UzOdrHR4G7w08,24355
+pipecat/services/cartesia/tts.py,sha256=I_OZCINywkDXmYzFL35MjSN8cAuNEaJs7nj0YB_obtc,27008
 pipecat/services/cerebras/__init__.py,sha256=5zBmqq9Zfcl-HC7ylekVS5qrRedbl1mAeEwUT-T-c_o,259
 pipecat/services/cerebras/llm.py,sha256=-yzSe_6YDGigwzES-LZS4vNXMPugmvsIYEpTySyr5nA,3047
 pipecat/services/deepgram/__init__.py,sha256=IjRtMI7WytRDdmYVpk2qDWClXUiNgdl7ZkvEAWg1eYE,304
@@ -324,7 +325,7 @@ pipecat/services/sambanova/llm.py,sha256=5XVfPLEk__W8ykFqLdV95ZUhlGGkAaJwmbciLdZ
 pipecat/services/sambanova/stt.py,sha256=ZZgEZ7WQjLFHbCko-3LNTtVajjtfUvbtVLtFcaNadVQ,2536
 pipecat/services/sarvam/__init__.py,sha256=B4TN_tTHV9fWg0aSoPvfQlXISA0nJaQ9-u08I9UWvH4,280
 pipecat/services/sarvam/stt.py,sha256=p9Iq4loMwnftNZ_S0WoFSoX7iBbRKyja6RsVWbpj508,19314
-pipecat/services/sarvam/tts.py,sha256=wzfa0vvmd0wtuzqFSjRbTmHHS8H0L8nP9jkXwqFUJ3A,27638
+pipecat/services/sarvam/tts.py,sha256=lrwfdC53kZ7f2QPgNRxzryISNkrJCvNtlZ-19-iXg94,27610
 pipecat/services/simli/__init__.py,sha256=cbDcqOaGsEgKbGYKpJ1Vv7LN4ZjOWA04sE84WW5vgQI,257
 pipecat/services/simli/video.py,sha256=Zu2XLvl2Y6VHaWzT9wEdzW9d0EYoZyzYLxjQFyV8vho,8320
 pipecat/services/soniox/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -338,7 +339,7 @@ pipecat/services/together/llm.py,sha256=VSayO-U6g9Ld0xK9CXRQPUsd5gWJKtiA8qDAyXgs
 pipecat/services/ultravox/__init__.py,sha256=EoHCSXI2o0DFQslELgkhAGZtxDj63gZi-9ZEhXljaKE,259
 pipecat/services/ultravox/stt.py,sha256=uCQm_-LbycXdXRV6IE1a6Mymis6tyww7V8PnPzAQtx8,16586
 pipecat/services/vistaar/__init__.py,sha256=UFfSWFN5rbzl6NN-E_OH_MFaSYodZWNlenAU0wk-rAI,110
-pipecat/services/vistaar/llm.py,sha256=8jp9BxGYOysmD6CFyof7m2AJRbTDx4KT4kFuUc95wcc,19335
+pipecat/services/vistaar/llm.py,sha256=GNVKaelbpNH7NW7iOpBj2rJjmhMVUsPqfnBI-YgIjjw,19326
 pipecat/services/whisper/__init__.py,sha256=smADmw0Fv98k7cGRuHTEcljKTO2WdZqLpJd0qsTCwH8,281
 pipecat/services/whisper/base_stt.py,sha256=VhslESPnYIeVbmnQTzmlZPV35TH49duxYTvJe0epNnE,7850
 pipecat/services/whisper/stt.py,sha256=9Qd56vWMzg3LtHikQnfgyMtl4odE6BCHDbpAn3HSWjw,17480
@@ -353,7 +354,7 @@ pipecat/transcriptions/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3
 pipecat/transcriptions/language.py,sha256=-mWI1MiZbasuoqZTOBH69dAmoM7-UJzWq9rSCcrnmh4,8228
 pipecat/transports/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/transports/base_input.py,sha256=WGtoXXlF3GIjYgjtYnAgi8nZozd5abNlGNjwRnz8FRs,20138
-pipecat/transports/base_output.py,sha256=mNlIOo7tETlbYPbDyOtA2H-TkBGFKmjuCMDzQUtiwmk,35423
+pipecat/transports/base_output.py,sha256=7WoXtAQAi-3OC9PC_zk61lCWlBTk5-NuTLUbsQUAI_U,36723
 pipecat/transports/base_transport.py,sha256=JlNiH0DysTfr6azwHauJqY_Z9HJC702O29Q0qrsLrg4,7530
 pipecat/transports/daily/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 pipecat/transports/daily/transport.py,sha256=VanO33ff9g6px-vwGgT6M7cMVg786pOGfMU7Okm7a78,91917
@@ -415,7 +416,7 @@ pipecat/utils/tracing/service_decorators.py,sha256=fwzxFpi8DJl6BJbK74G0UEB4ccMJg
 pipecat/utils/tracing/setup.py,sha256=7TEgPNpq6M8lww8OQvf0P9FzYc5A30xICGklVA-fua0,2892
 pipecat/utils/tracing/turn_context_provider.py,sha256=ikon3plFOx0XbMrH6DdeHttNpb-U0gzMZIm3bWLc9eI,2485
 pipecat/utils/tracing/turn_trace_observer.py,sha256=dma16SBJpYSOE58YDWy89QzHyQFc_9gQZszKeWixuwc,9725
-dv_pipecat_ai-0.0.85.dev830.dist-info/METADATA,sha256=wPJAPffJo_L5wKNWKbIxlaBG09JAGKUTFl_qkLwmoPw,32924
-dv_pipecat_ai-0.0.85.dev830.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-dv_pipecat_ai-0.0.85.dev830.dist-info/top_level.txt,sha256=kQzG20CxGf-nSsHmtXHx3hY2-8zHA3jYg8jk0TajqXc,8
-dv_pipecat_ai-0.0.85.dev830.dist-info/RECORD,,
+dv_pipecat_ai-0.0.85.dev837.dist-info/METADATA,sha256=dQC8Y4gHZ3jPBKpybN1R9aKRUbb9mQpb0cPuLQo5KUc,32924
+dv_pipecat_ai-0.0.85.dev837.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+dv_pipecat_ai-0.0.85.dev837.dist-info/top_level.txt,sha256=kQzG20CxGf-nSsHmtXHx3hY2-8zHA3jYg8jk0TajqXc,8
+dv_pipecat_ai-0.0.85.dev837.dist-info/RECORD,,

pipecat/processors/aggregators/llm_response.py

@@ -49,7 +49,6 @@ from pipecat.frames.frames import (
     OpenAILLMContextAssistantTimestampFrame,
     SpeechControlParamsFrame,
     StartFrame,
-    StartInterruptionFrame,
     TextFrame,
     TranscriptDropFrame,
     TranscriptionFrame,
@@ -473,8 +472,8 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
             frame: The frame to process.
             direction: The direction of frame flow in the pipeline.
         """
-        if isinstance(frame, StartInterruptionFrame):
-            self.logger.debug("Received StartInterruptionFrame")
+        if isinstance(frame, InterruptionFrame):
+            self.logger.debug("Received InterruptionFrame")
         await super().process_frame(frame, direction)
 
         if isinstance(frame, StartFrame):
@@ -560,7 +559,7 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
                         "Triggering interruption - pushing BotInterruptionFrame and aggregation"
                     )
                     # await self.push_frame(BotInterruptionFrame(), FrameDirection.UPSTREAM)
-                    await self.push_frame(StartInterruptionFrame(), FrameDirection.DOWNSTREAM)
+                    await self.push_frame(InterruptionFrame(), FrameDirection.DOWNSTREAM)
                     self.logger.debug("Pushed BotInterruptionFrame")
                 # No interruption config - normal behavior (always push aggregation)
                 await self._process_aggregation()
@@ -596,12 +595,8 @@ class LLMUserContextAggregator(LLMContextResponseAggregator):
         """Notify upstream processors that pending transcripts should be dropped."""
         if self._pending_transcription_ids:
             drop_frame = TranscriptDropFrame(transcript_ids=list(self._pending_transcription_ids))
-            self.logger.debug(
-                f"Dropping {len(self._pending_transcription_ids)} transcript chunk(s) due to {reason}"
-            )
             await self.push_frame(drop_frame, FrameDirection.UPSTREAM)
         self._pending_transcription_ids.clear()
-        self._aggregation = ""
 
     async def _start(self, frame: StartFrame):
         self._create_aggregation_task()

pipecat/processors/dtmf_aggregator.py

@@ -1,3 +1,5 @@
+"""DTMF aggregator processor for collecting and flushing DTMF input digits."""
+
 import asyncio
 
 from pipecat.frames.frames import (
@@ -8,8 +10,8 @@ from pipecat.frames.frames import (
     EndFrame,
     Frame,
     InputDTMFFrame,
+    InterruptionFrame,
     StartDTMFCaptureFrame,
-    StartInterruptionFrame,
     TranscriptionFrame,
     WaitForDTMFFrame,
 )
@@ -19,10 +21,11 @@ from pipecat.utils.time import time_now_iso8601
 
 class DTMFAggregator(FrameProcessor):
     """Aggregates DTMF frames using idle wait logic.
+
     The aggregator accumulates digits from incoming InputDTMFFrame instances.
     It flushes the aggregated digits by emitting a TranscriptionFrame when:
       - No new digit arrives within the specified timeout period,
-      - The termination digit (“#”) is received, or
+      - The termination digit ("#") is received, or
       - The number of digits aggregated equals the configured 'digits' value.
     """
 
@@ -34,7 +37,9 @@ class DTMFAggregator(FrameProcessor):
         digits: int = None,
         **kwargs,
     ):
-        """:param timeout: Idle timeout in seconds before flushing the aggregated digits.
+        """Initialize the DTMF aggregator.
+
+        :param timeout: Idle timeout in seconds before flushing the aggregated digits.
         :param digits: Number of digits to aggregate before flushing.
         """
         super().__init__(**kwargs)
@@ -48,6 +53,7 @@ class DTMFAggregator(FrameProcessor):
         self._dtmf_capture_active = False
 
     async def process_frame(self, frame: Frame, direction: FrameDirection) -> None:
+        """Process incoming frames and handle DTMF input aggregation."""
         # Handle DTMF frames.
         await super().process_frame(frame, direction)
 
@@ -69,8 +75,8 @@ class DTMFAggregator(FrameProcessor):
             self._digit_event.set()  # Trigger the timeout handler
             await self._start_dtmf_capture()
             await self.push_frame(frame, direction)
-        elif isinstance(frame, StartInterruptionFrame):
-            self.logger.debug("Received StartInterruptionFrame")
+        elif isinstance(frame, InterruptionFrame):
+            self.logger.debug("Received InterruptionFrame")
             if self._aggregation:
                 await self.flush_aggregation()
             await self._end_dtmf_capture()
@@ -108,9 +114,7 @@ class DTMFAggregator(FrameProcessor):
         if "digits" in settings:
             new_digits = settings["digits"]
             if new_digits != self._digits:
-                self.logger.debug(
-                    f"Updating DTMF digits from {self._digits} to {new_digits}"
-                )
+                self.logger.debug(f"Updating DTMF digits from {self._digits} to {new_digits}")
                 self._digits = new_digits
                 settings_changed = True
 
@@ -125,9 +129,7 @@ class DTMFAggregator(FrameProcessor):
                 new_end_on = set(end_value)
 
             if new_end_on != self._end_on:
-                self.logger.debug(
-                    f"Updating DTMF end_on from {self._end_on} to {new_end_on}"
-                )
+                self.logger.debug(f"Updating DTMF end_on from {self._end_on} to {new_end_on}")
                 self._end_on = new_end_on
                 settings_changed = True
 
@@ -142,9 +144,7 @@ class DTMFAggregator(FrameProcessor):
                 new_reset_on = set(reset_value)
 
             if new_reset_on != self._reset_on:
-                self.logger.debug(
-                    f"Updating DTMF reset_on from {self._reset_on} to {new_reset_on}"
-                )
+                self.logger.debug(f"Updating DTMF reset_on from {self._reset_on} to {new_reset_on}")
                 self._reset_on = new_reset_on
                 settings_changed = True
 
@@ -183,9 +183,7 @@ class DTMFAggregator(FrameProcessor):
     def _create_aggregation_task(self, raise_timeout: bool = False) -> None:
         """Creates the aggregation task if it hasn't been created yet."""
         if not self._aggregation_task:
-            self._aggregation_task = self.create_task(
-                self._aggregation_task_handler(raise_timeout)
-            )
+            self._aggregation_task = self.create_task(self._aggregation_task_handler(raise_timeout))
 
     async def _stop_aggregation_task(self) -> None:
         """Stops the aggregation task."""
@@ -198,9 +196,7 @@ class DTMFAggregator(FrameProcessor):
         while True:
             try:
                 # Wait for a new digit signal with a timeout.
-                await asyncio.wait_for(
-                    self._digit_event.wait(), timeout=self._idle_timeout
-                )
+                await asyncio.wait_for(self._digit_event.wait(), timeout=self._idle_timeout)
                 self._digit_event.clear()
             except asyncio.TimeoutError:
                 # No new digit arrived within the timeout period; flush if needed
@@ -216,7 +212,7 @@ class DTMFAggregator(FrameProcessor):
             aggregated_frame.metadata["push_aggregation"] = True
 
             # Send interruption frame (as per original design)
-            await self.push_frame(StartInterruptionFrame(), FrameDirection.DOWNSTREAM)
+            await self.push_frame(InterruptionFrame(), FrameDirection.DOWNSTREAM)
 
             # Push the transcription frame
             await self.push_frame(aggregated_frame, FrameDirection.DOWNSTREAM)

pipecat/serializers/__init__.py

@@ -5,6 +5,7 @@ from .exotel import ExotelFrameSerializer
 from .plivo import PlivoFrameSerializer
 from .telnyx import TelnyxFrameSerializer
 from .twilio import TwilioFrameSerializer
+from .vi import VIFrameSerializer
 
 __all__ = [
     "FrameSerializer",
@@ -15,6 +16,7 @@ __all__ = [
     "PlivoFrameSerializer",
     "TelnyxFrameSerializer",
     "TwilioFrameSerializer",
+    "VIFrameSerializer",
 ]
 
 # Optional imports

pipecat/serializers/asterisk.py

@@ -1,4 +1,6 @@
 # asterisk_ws_serializer.py
+"""Frame serializer for Asterisk WebSocket communication."""
+
 import base64
 import json
 from typing import Literal, Optional
@@ -12,8 +14,8 @@ from pipecat.frames.frames import (
     EndFrame,
     Frame,
     InputAudioRawFrame,
+    InterruptionFrame,
     StartFrame,
-    StartInterruptionFrame,
     TransportMessageFrame,
     TransportMessageUrgentFrame,
 )
@@ -21,6 +23,8 @@ from pipecat.serializers.base_serializer import FrameSerializer, FrameSerializer
 
 
 class AsteriskFrameSerializer(FrameSerializer):
+    """Serializes Pipecat frames to/from Asterisk WebSocket JSON messages."""
+
     class InputParams(BaseModel):
         """Configuration parameters for AsteriskFrameSerializer.
 
@@ -39,6 +43,12 @@ class AsteriskFrameSerializer(FrameSerializer):
         auto_hang_up: bool = False  # no-op here; adapter handles hangup
 
     def __init__(self, stream_id: str, params: Optional[InputParams] = None):
+        """Initialize the Asterisk frame serializer.
+
+        Args:
+            stream_id: Unique identifier for the media stream.
+            params: Configuration parameters for the serializer.
+        """
         self._stream_id = stream_id
         self._params = params or AsteriskFrameSerializer.InputParams()
         self._tel_rate = self._params.telephony_sample_rate
@@ -49,13 +59,16 @@ class AsteriskFrameSerializer(FrameSerializer):
 
     @property
     def type(self) -> FrameSerializerType:
+        """Return the serializer type (TEXT for JSON messages)."""
         return FrameSerializerType.TEXT  # we send/recv JSON strings
 
     async def setup(self, frame: StartFrame):
+        """Setup the serializer with audio parameters from the StartFrame."""
         self._sample_rate = self._params.sample_rate or frame.audio_in_sample_rate
 
     # Pipecat -> Adapter (play to caller)
     async def serialize(self, frame: Frame) -> str | bytes | None:
+        """Serialize Pipecat frames to Asterisk WebSocket JSON messages."""
         # On pipeline end, ask bridge to hang up
         if (
             self._params.auto_hang_up
@@ -64,7 +77,7 @@ class AsteriskFrameSerializer(FrameSerializer):
         ):
             self._hangup_sent = True
             return json.dumps({"event": "hangup"})
-        if isinstance(frame, StartInterruptionFrame):
+        if isinstance(frame, InterruptionFrame):
             return json.dumps({"event": "clear", "streamId": self._stream_id})
         if isinstance(frame, AudioRawFrame):
             pcm = frame.audio
@@ -114,6 +127,7 @@ class AsteriskFrameSerializer(FrameSerializer):
 
     # Adapter -> Pipecat (audio from caller)
     async def deserialize(self, data: str | bytes) -> Frame | None:
+        """Deserialize Asterisk WebSocket JSON messages to Pipecat frames."""
         try:
             msg = json.loads(data)
         except Exception:

pipecat/serializers/convox.py

@@ -22,9 +22,9 @@ from pipecat.frames.frames import (
     Frame,
     InputAudioRawFrame,
     InputDTMFFrame,
+    InterruptionFrame,
     KeypadEntry,
     StartFrame,
-    StartInterruptionFrame,
     TransportMessageFrame,
     TransportMessageUrgentFrame,
 )
@@ -117,7 +117,7 @@ class ConVoxFrameSerializer(FrameSerializer):
             self._call_ended = True
             # Return the callEnd event to be sent via the WebSocket
             return await self._send_call_end_event()
-        elif isinstance(frame, StartInterruptionFrame):
+        elif isinstance(frame, InterruptionFrame):
             # Clear/interrupt command for ConVox
             message = {
                 "event": "clear",

pipecat/serializers/custom.py

@@ -28,8 +28,8 @@ from pipecat.frames.frames import (
     EndFrame,
     Frame,
     InputAudioRawFrame,
+    InterruptionFrame,
     StartFrame,
-    StartInterruptionFrame,
     TransportMessageFrame,
     TransportMessageUrgentFrame,
 )
@@ -121,7 +121,7 @@ class CustomFrameSerializer(FrameSerializer):
         Returns:
             Serialized data as JSON string, or None if the frame isn't handled.
         """
-        if isinstance(frame, StartInterruptionFrame):
+        if isinstance(frame, InterruptionFrame):
             # Send clear event to instruct client to discard buffered audio
             answer = {"event": "clear", "stream_sid": self._stream_sid}
             return json.dumps(answer)

pipecat/serializers/vi.py

@@ -0,0 +1,324 @@
+#
+# Copyright (c) 2024–2025, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+"""Vodafone Idea (VI) WebSocket frame serializer for audio streaming and call management."""
+
+import base64
+import json
+from datetime import datetime, timezone
+from typing import Optional
+
+from loguru import logger
+from pydantic import BaseModel
+
+from pipecat.audio.utils import create_default_resampler
+from pipecat.frames.frames import (
+    AudioRawFrame,
+    CancelFrame,
+    EndFrame,
+    Frame,
+    InputAudioRawFrame,
+    InputDTMFFrame,
+    KeypadEntry,
+    StartFrame,
+    StartInterruptionFrame,
+    TransportMessageFrame,
+    TransportMessageUrgentFrame,
+)
+from pipecat.serializers.base_serializer import FrameSerializer, FrameSerializerType
+
+
+class VIFrameSerializer(FrameSerializer):
+    """Serializer for Vodafone Idea (VI) WebSocket protocol.
+
+    This serializer handles converting between Pipecat frames and VI's WebSocket
+    protocol for bidirectional audio streaming. It supports audio conversion, DTMF events,
+    and real-time communication with VI telephony systems.
+
+    VI WebSocket protocol requirements:
+    - PCM audio format at 8kHz sample rate
+    - 16-bit Linear PCM encoding
+    - Base64 encoded audio payloads
+    - JSON message format for control and media events
+    - Bitrate: 128 Kbps
+
+    Events (VI → Endpoint):
+    - connected: WebSocket connection established
+    - start: Stream session started with call/stream IDs
+    - media: Audio data in Base64-encoded PCM
+    - dtmf: Keypad digit pressed
+    - stop: Stream ended
+    - mark: Audio playback checkpoint confirmation
+
+    Events (Endpoint → VI):
+    - media: Send audio back to VI
+    - mark: Request acknowledgment for audio playback
+    - clear: Clear queued audio (interruption)
+    - exit: Terminate session gracefully
+    """
+
+    class InputParams(BaseModel):
+        """Configuration parameters for VIFrameSerializer.
+
+        Attributes:
+            vi_sample_rate: Sample rate used by VI, defaults to 8000 Hz (telephony standard).
+            sample_rate: Optional override for pipeline input sample rate.
+            auto_hang_up: Whether to automatically terminate call on EndFrame.
+        """
+
+        vi_sample_rate: int = 8000
+        sample_rate: Optional[int] = None
+        auto_hang_up: bool = False
+
+    def __init__(
+        self,
+        stream_id: str,
+        call_id: Optional[str] = None,
+        params: Optional[InputParams] = None,
+    ):
+        """Initialize the VIFrameSerializer.
+
+        Args:
+            stream_id: The VI stream identifier.
+            call_id: The associated VI call identifier.
+            params: Configuration parameters.
+        """
+        self._stream_id = stream_id
+        self._call_id = call_id
+        self._params = params or VIFrameSerializer.InputParams()
+
+        self._vi_sample_rate = self._params.vi_sample_rate
+        self._sample_rate = 0  # Pipeline input rate
+        self._call_ended = False
+
+        self._resampler = create_default_resampler()
+
+    @property
+    def type(self) -> FrameSerializerType:
+        """Gets the serializer type.
+
+        Returns:
+            The serializer type as TEXT for JSON WebSocket messages.
+        """
+        return FrameSerializerType.TEXT
+
+    async def setup(self, frame: StartFrame):
+        """Sets up the serializer with pipeline configuration.
+
+        Args:
+            frame: The StartFrame containing pipeline configuration.
+        """
+        self._sample_rate = self._params.sample_rate or frame.audio_in_sample_rate
+
+    async def serialize(self, frame: Frame) -> str | bytes | None:
+        """Serializes a Pipecat frame to VI WebSocket format.
+
+        Handles conversion of various frame types to VI WebSocket messages.
+        For EndFrames, initiates call termination if auto_hang_up is enabled.
+
+        Args:
+            frame: The Pipecat frame to serialize.
+
+        Returns:
+            Serialized data as JSON string, or None if the frame isn't handled.
+        """
+        if (
+            self._params.auto_hang_up
+            and not self._call_ended
+            and isinstance(frame, (EndFrame, CancelFrame))
+        ):
+            self._call_ended = True
+            # Return the exit event to terminate the VI session
+            return await self._send_exit_event()
+
+        elif isinstance(frame, StartInterruptionFrame):
+            # Clear/interrupt command for VI - clears queued audio
+            message = {
+                "event": "clear",
+                "stream_id": self._stream_id,
+                "call_id": self._call_id,
+            }
+            logger.debug(f"VI: Sending clear event for stream_id: {self._stream_id}")
+            return json.dumps(message)
+
+        elif isinstance(frame, AudioRawFrame):
+            if self._call_ended:
+                logger.debug("VI SERIALIZE: Skipping audio - call has ended")
+                return None
+
+            # Convert PCM audio to VI format
+            data = frame.audio
+
+            # Resample to VI sample rate (8kHz)
+            serialized_data = await self._resampler.resample(
+                data, frame.sample_rate, self._vi_sample_rate
+            )
+
+            # Encode as base64 for transmission
+            payload = base64.b64encode(serialized_data).decode("ascii")
+
+            # VI expects media event format with Base64-encoded PCM audio
+            timestamp = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
+
+            message = {
+                "event": "media",
+                "stream_id": self._stream_id,
+                "media": {
+                    "timestamp": timestamp,
+                    "chunk": len(serialized_data),  # Chunk size in bytes
+                    "payload": payload,
+                },
+            }
+
+            return json.dumps(message)
+
+        elif isinstance(frame, (TransportMessageFrame, TransportMessageUrgentFrame)):
+            # Pass through transport messages (for mark events, etc.)
+            return json.dumps(frame.message)
+
+        return None
+
+    async def _send_exit_event(self):
+        """Send an exit event to VI to terminate the session gracefully.
+
+        This method is called when auto_hang_up is enabled and an EndFrame or
+        CancelFrame is received. The exit event allows IVR logic to continue
+        after the WebSocket session ends.
+        """
+        try:
+            exit_event = {
+                "event": "exit",
+                "stream_id": self._stream_id,
+                "call_id": self._call_id,
+                "timestamp": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
+            }
+
+            logger.info(
+                f"VI auto_hang_up: Sending exit event for stream_id: {self._stream_id}, call_id: {self._call_id}"
+            )
+            return json.dumps(exit_event)
+        except Exception as e:
+            logger.error(f"VI auto_hang_up: Failed to create exit event: {e}")
+            return None
+
+    async def deserialize(self, data: str | bytes) -> Frame | None:
+        """Deserializes VI WebSocket data to Pipecat frames.
+
+        Handles conversion of VI media events to appropriate Pipecat frames.
+
+        Args:
+            data: The raw WebSocket data from VI.
+
+        Returns:
+            A Pipecat frame corresponding to the VI event, or None if unhandled.
+        """
+        try:
+            message = json.loads(data)
+        except json.JSONDecodeError:
+            logger.error(f"Invalid JSON received from VI: {data}")
+            return None
+
+        # Log all incoming events for debugging and monitoring
+        event = message.get("event")
+        logger.debug(
+            f"VI INCOMING EVENT: {event} - stream_id: {self._stream_id}, call_id: {self._call_id}"
+        )
+
+        if event == "media":
+            # Handle incoming audio data from VI
+            media = message.get("media", {})
+            payload_base64 = media.get("payload")
+
+            if not payload_base64:
+                logger.warning("VI DESERIALIZE: No payload in VI media message")
+                return None
+
+            try:
+                payload = base64.b64decode(payload_base64)
+                chunk_size = len(payload)
+
+                # Log chunk info (optional)
+                logger.debug(
+                    f"VI DESERIALIZE: Received audio from VI - {chunk_size} bytes at {self._vi_sample_rate}Hz"
+                )
+
+            except Exception as e:
+                logger.error(f"VI DESERIALIZE: Error decoding VI audio payload: {e}")
+                return None
+
+            # Convert from VI sample rate (8kHz) to pipeline sample rate
+            deserialized_data = await self._resampler.resample(
+                payload,
+                self._vi_sample_rate,
+                self._sample_rate,
+            )
+
+            audio_frame = InputAudioRawFrame(
+                audio=deserialized_data,
+                num_channels=1,  # VI uses mono audio
+                sample_rate=self._sample_rate,
+            )
+            return audio_frame
+
+        elif event == "dtmf":
+            # Handle DTMF events
+            dtmf_data = message.get("dtmf", {})
+            digit = dtmf_data.get("digit")
+
+            if digit:
+                try:
+                    logger.info(f"VI: Received DTMF digit: {digit}")
+                    return InputDTMFFrame(KeypadEntry(digit))
+                except ValueError:
+                    logger.warning(f"Invalid DTMF digit from VI: {digit}")
+                    return None
+
+        elif event == "connected":
+            # Handle connection event
+            logger.info(f"VI connection established: {message}")
+            return None
+
+        elif event == "start":
+            # Handle stream start event
+            logger.info(f"VI stream started: {message}")
+            return None
+
+        elif event == "stop":
+            # Handle stream stop event
+            logger.info(f"VI stream stopped: {message}")
+            # Don't end the call here, wait for explicit exit or call end
+            return None
+
+        elif event == "mark":
+            # Handle mark event - checkpoint confirming audio playback completion
+            mark_data = message.get("mark", {})
+            mark_name = mark_data.get("name", "unknown")
+            logger.info(f"VI mark event received: {mark_name}")
+            # Mark events are informational, no frame to return
+            return None
+
+        elif event == "error":
+            # Handle error events
+            error_msg = message.get("error", "Unknown error")
+            logger.error(f"VI error: {error_msg}")
+            return None
+
+        elif event == "exit":
+            # Handle exit event from VI
+            logger.info("VI exit event received - terminating session")
+            self._call_ended = True
+            return CancelFrame()
+
+        elif event == "call_end" or event == "callEnd":
+            # Handle call end event (if VI sends this)
+            logger.info("VI call end event received")
+            self._call_ended = True
+            return CancelFrame()
+
+        else:
+            logger.debug(f"VI UNHANDLED EVENT: {event}")
+
+        return None

pipecat/services/cartesia/tts.py

@@ -15,7 +15,6 @@ from typing import AsyncGenerator, List, Literal, Optional, Union
 from loguru import logger
 from pydantic import BaseModel, Field
 
-
 from pipecat.frames.frames import (
     CancelFrame,
     EndFrame,
@@ -49,6 +48,26 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+class GenerationConfig(BaseModel):
+    """Configuration for Cartesia Sonic-3 generation parameters.
+
+    Sonic-3 interprets these parameters as guidance to ensure natural speech.
+    Test against your content for best results.
+
+    Parameters:
+        volume: Volume multiplier for generated speech. Valid range: [0.5, 2.0]. Default is 1.0.
+        speed: Speed multiplier for generated speech. Valid range: [0.6, 1.5]. Default is 1.0.
+        emotion: Single emotion string to guide the emotional tone. Examples include neutral,
+            angry, excited, content, sad, scared. Over 60 emotions are supported. For best
+            results, use with recommended voices: Leo, Jace, Kyle, Gavin, Maya, Tessa, Dana,
+            and Marian.
+    """
+
+    volume: Optional[float] = None
+    speed: Optional[float] = None
+    emotion: Optional[str] = None
+
+
 def language_to_cartesia_language(language: Language) -> Optional[str]:
     """Convert a Language enum to Cartesia language code.
 
@@ -74,6 +93,33 @@ def language_to_cartesia_language(language: Language) -> Optional[str]:
         Language.SV: "sv",
         Language.TR: "tr",
         Language.ZH: "zh",
+        Language.TL: "tl",
+        Language.BG: "bg",
+        Language.RO: "ro",
+        Language.AR: "ar",
+        Language.CS: "cs",
+        Language.EL: "el",
+        Language.FI: "fi",
+        Language.HR: "hr",
+        Language.MS: "ms",
+        Language.SK: "sk",
+        Language.DA: "da",
+        Language.TA: "ta",
+        Language.UK: "uk",
+        Language.HU: "hu",
+        Language.NO: "no",
+        Language.VI: "vi",
+        Language.BN: "bn",
+        Language.TH: "th",
+        Language.HE: "he",
+        Language.KA: "ka",
+        Language.ID: "id",
+        Language.TE: "te",
+        Language.GU: "gu",
+        Language.KN: "kn",
+        Language.ML: "ml",
+        Language.MR: "mr",
+        Language.PA: "pa",
     }
 
     result = BASE_LANGUAGES.get(language)
@@ -102,16 +148,20 @@ class CartesiaTTSService(AudioContextWordTTSService):
 
         Parameters:
             language: Language to use for synthesis.
-            speed: Voice speed control.
-            emotion: List of emotion controls.
+            speed: Voice speed control for non-Sonic-3 models (literal values).
+            emotion: List of emotion controls for non-Sonic-3 models.
 
                 .. deprecated:: 0.0.68
                         The `emotion` parameter is deprecated and will be removed in a future version.
+
+            generation_config: Generation configuration for Sonic-3 models. Includes volume,
+                speed (numeric), and emotion (string) parameters.
         """
 
         language: Optional[Language] = Language.EN
         speed: Optional[Literal["slow", "normal", "fast"]] = None
         emotion: Optional[List[str]] = []
+        generation_config: Optional[GenerationConfig] = None
 
     def __init__(
         self,
@@ -120,7 +170,7 @@ class CartesiaTTSService(AudioContextWordTTSService):
         voice_id: str,
         cartesia_version: str = "2025-04-16",
         url: str = "wss://api.cartesia.ai/tts/websocket",
-        model: str = "sonic-2",
+        model: str = "sonic-3",
         sample_rate: Optional[int] = None,
         encoding: str = "pcm_s16le",
         container: str = "raw",
@@ -136,7 +186,7 @@ class CartesiaTTSService(AudioContextWordTTSService):
             voice_id: ID of the voice to use for synthesis.
             cartesia_version: API version string for Cartesia service.
             url: WebSocket URL for Cartesia TTS API.
-            model: TTS model to use (e.g., "sonic-2").
+            model: TTS model to use (e.g., "sonic-3").
             sample_rate: Audio sample rate. If None, uses default.
             encoding: Audio encoding format.
             container: Audio container format.
@@ -180,6 +230,7 @@ class CartesiaTTSService(AudioContextWordTTSService):
             else "en",
             "speed": params.speed,
             "emotion": params.emotion,
+            "generation_config": params.generation_config,
         }
         self.set_model_name(model)
         self.set_voice(voice_id)
@@ -298,6 +349,11 @@ class CartesiaTTSService(AudioContextWordTTSService):
         if self._settings["speed"]:
             msg["speed"] = self._settings["speed"]
 
+        if self._settings["generation_config"]:
+            msg["generation_config"] = self._settings["generation_config"].model_dump(
+                exclude_none=True
+            )
+
         return json.dumps(msg)
 
     async def start(self, frame: StartFrame):
@@ -419,7 +475,6 @@ class CartesiaTTSService(AudioContextWordTTSService):
                 logger.error(f"{self} error: {msg}")
                 await self.push_frame(TTSStoppedFrame())
                 await self.stop_all_metrics()
-
                 await self.push_error(ErrorFrame(f"{self} error: {msg['error']}"))
                 self._context_id = None
             else:
@@ -484,23 +539,27 @@ class CartesiaHttpTTSService(TTSService):
 
         Parameters:
             language: Language to use for synthesis.
-            speed: Voice speed control.
-            emotion: List of emotion controls.
+            speed: Voice speed control for non-Sonic-3 models (literal values).
+            emotion: List of emotion controls for non-Sonic-3 models.
 
                 .. deprecated:: 0.0.68
                         The `emotion` parameter is deprecated and will be removed in a future version.
+
+            generation_config: Generation configuration for Sonic-3 models. Includes volume,
+                speed (numeric), and emotion (string) parameters.
         """
 
         language: Optional[Language] = Language.EN
         speed: Optional[Literal["slow", "normal", "fast"]] = None
         emotion: Optional[List[str]] = Field(default_factory=list)
+        generation_config: Optional[GenerationConfig] = None
 
     def __init__(
         self,
         *,
         api_key: str,
         voice_id: str,
-        model: str = "sonic-2",
+        model: str = "sonic-3",
         base_url: str = "https://api.cartesia.ai",
         cartesia_version: str = "2024-11-13",
         sample_rate: Optional[int] = None,
@@ -514,7 +573,7 @@ class CartesiaHttpTTSService(TTSService):
         Args:
             api_key: Cartesia API key for authentication.
             voice_id: ID of the voice to use for synthesis.
-            model: TTS model to use (e.g., "sonic-2").
+            model: TTS model to use (e.g., "sonic-3").
             base_url: Base URL for Cartesia HTTP API.
             cartesia_version: API version string for Cartesia service.
             sample_rate: Audio sample rate. If None, uses default.
@@ -541,6 +600,7 @@ class CartesiaHttpTTSService(TTSService):
             else "en",
             "speed": params.speed,
             "emotion": params.emotion,
+            "generation_config": params.generation_config,
         }
         self.set_voice(voice_id)
         self.set_model_name(model)
@@ -634,6 +694,11 @@ class CartesiaHttpTTSService(TTSService):
             if self._settings["speed"]:
                 payload["speed"] = self._settings["speed"]
 
+            if self._settings["generation_config"]:
+                payload["generation_config"] = self._settings["generation_config"].model_dump(
+                    exclude_none=True
+                )
+
             yield TTSStartedFrame()
 
             session = await self._client._get_session()

pipecat/services/sarvam/tts.py

@@ -23,7 +23,6 @@ from pipecat.frames.frames import (
     InterruptionFrame,
     LLMFullResponseEndFrame,
     StartFrame,
-    StartInterruptionFrame,
     TTSAudioRawFrame,
     TTSStartedFrame,
     TTSStoppedFrame,

pipecat/services/vistaar/llm.py

@@ -14,15 +14,15 @@ from loguru import logger
 from pydantic import BaseModel, Field
 
 from pipecat.frames.frames import (
-    EndFrame,
     CancelFrame,
+    EndFrame,
     Frame,
+    InterruptionFrame,
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
     LLMMessagesFrame,
     LLMTextFrame,
     LLMUpdateSettingsFrame,
-    StartInterruptionFrame,
 )
 from pipecat.processors.aggregators.llm_response import (
     LLMAssistantAggregatorParams,
@@ -391,7 +391,7 @@ class VistaarLLMService(LLMService):
             )
             await self.push_frame(frame, direction)
             return
-        elif isinstance(frame, StartInterruptionFrame):
+        elif isinstance(frame, InterruptionFrame):
             await self._handle_interruption()
             await self.push_frame(frame, direction)
             return
@@ -467,4 +467,4 @@ class VistaarLLMService(LLMService):
 
     def can_generate_metrics(self) -> bool:
         """Check if this service can generate processing metrics."""
-        return True
+        return True

pipecat/transports/base_output.py

@@ -50,6 +50,11 @@ from pipecat.utils.time import nanoseconds_to_seconds
 
 # TODO: When we use GeminiMultimodalLiveLLMService, we need to change this to 0.35 but that creates issue for faster TTS.
 BOT_VAD_STOP_SECS = 0.30
+# For the very first bot utterance (e.g., intro), we can safely
+# detect end-of-speech sooner to improve responsiveness for the
+# user’s first short reply. Keep conservative to avoid mid-utterance
+# false stops when TTS streams quickly.
+FIRST_BOT_VAD_STOP_SECS = 0.08
 
 
 class BaseOutputTransport(FrameProcessor):
@@ -406,6 +411,9 @@ class BaseOutputTransport(FrameProcessor):
             self._bot_speaking_frame_period = 0.2
             # Last time the bot actually spoke.
             self._bot_speech_last_time = 0
+            # Before the first stop event, we use a shorter silence
+            # threshold to make the first turn more responsive.
+            self._first_stop_pending = True
 
             self._audio_task: Optional[asyncio.Task] = None
             self._video_task: Optional[asyncio.Task] = None
@@ -631,6 +639,10 @@ class BaseOutputTransport(FrameProcessor):
 
             self._bot_speaking = False
 
+            # Mark that the first stop has been completed so subsequent
+            # stops use the regular (longer) VAD stop threshold.
+            self._first_stop_pending = False
+
             # Clean audio buffer (there could be tiny left overs if not multiple
             # to our output chunk size).
             self._audio_buffer = bytearray()
@@ -690,9 +702,14 @@ class BaseOutputTransport(FrameProcessor):
             async def without_mixer(vad_stop_secs: float) -> AsyncGenerator[Frame, None]:
                 while True:
                     try:
-                        frame = await asyncio.wait_for(
-                            self._audio_queue.get(), timeout=vad_stop_secs
+                        # Use a shorter timeout only for the first bot stop to
+                        # accelerate the initial turn handoff right after the intro.
+                        timeout = (
+                            FIRST_BOT_VAD_STOP_SECS
+                            if getattr(self, "_first_stop_pending", True)
+                            else BOT_VAD_STOP_SECS
                         )
+                        frame = await asyncio.wait_for(self._audio_queue.get(), timeout=timeout)
                         yield frame
                         self._audio_queue.task_done()
                     except asyncio.TimeoutError:
@@ -713,7 +730,13 @@ class BaseOutputTransport(FrameProcessor):
                     except asyncio.QueueEmpty:
                         # Notify the bot stopped speaking upstream if necessary.
                         diff_time = time.time() - last_frame_time
-                        if diff_time > vad_stop_secs:
+                        # Use a shorter threshold for the first stop only.
+                        current_stop_secs = (
+                            FIRST_BOT_VAD_STOP_SECS
+                            if getattr(self, "_first_stop_pending", True)
+                            else BOT_VAD_STOP_SECS
+                        )
+                        if diff_time > current_stop_secs:
                             await self._bot_stopped_speaking()
                         # Generate an audio frame with only the mixer's part.
                         frame = OutputAudioRawFrame(