dv-pipecat-ai 0.0.93.dev136__tar.gz → 0.0.93.dev137__tar.gz

{dv_pipecat_ai-0.0.93.dev136/src/dv_pipecat_ai.egg-info → dv_pipecat_ai-0.0.93.dev137}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-ai
-Version: 0.0.93.dev136
+Version: 0.0.93.dev137
 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-0.0.93.dev136 → dv_pipecat_ai-0.0.93.dev137/src/dv_pipecat_ai.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dv-pipecat-ai
-Version: 0.0.93.dev136
+Version: 0.0.93.dev137
 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-0.0.93.dev136 → dv_pipecat_ai-0.0.93.dev137}/src/dv_pipecat_ai.egg-info/scm_file_list.json

@@ -873,6 +873,7 @@
     "tests/test_soniox_stt_connect.py",
     "tests/test_stt_mute_filter.py",
     "tests/test_transcript_processor.py",
+    "tests/test_tts_service_pause.py",
     "tests/test_turn_tracking_observer.py",
     "tests/test_user_aggregator_customizations.py",
     "tests/test_user_idle_processor.py",

dv_pipecat_ai-0.0.93.dev137/src/dv_pipecat_ai.egg-info/scm_version.json

@@ -0,0 +1,8 @@
+{
+  "tag": "0.0.92",
+  "distance": 137,
+  "node": "gedc7f9a34063d6750b0dd0ca7b735ac22eb69d16",
+  "dirty": false,
+  "branch": "dv-stage",
+  "node_date": "2026-09-01"
+}

{dv_pipecat_ai-0.0.93.dev136 → dv_pipecat_ai-0.0.93.dev137}/src/pipecat/processors/frame_processor.py

@@ -922,6 +922,21 @@ class FrameProcessor(BaseObject):
             await self.cancel_task(self.__process_frame_task)
             self.__process_frame_task = None
 
+    async def _process_frame_now(
+        self,
+        frame: Frame,
+        direction: FrameDirection,
+        callback: Optional[FrameCallback] = None,
+    ):
+        """Process a frame immediately, bypassing the input/process queues.
+
+        Subclasses use this to let a frame skip the queues (and therefore any
+        active `pause_processing_frames()` gate) while still firing the usual
+        `on_before_process_frame` / `on_after_process_frame` events and the
+        frame callback.
+        """
+        await self.__process_frame(frame, direction, callback)
+
     async def __process_frame(
         self, frame: Frame, direction: FrameDirection, callback: Optional[FrameCallback]
     ):

{dv_pipecat_ai-0.0.93.dev136 → dv_pipecat_ai-0.0.93.dev137}/src/pipecat/services/tts_service.py

@@ -36,6 +36,7 @@ from pipecat.frames.frames import (
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
     StartFrame,
+    SystemFrame,
     TextFrame,
     TranscriptionFrame,
     TTSAudioRawFrame,
@@ -45,7 +46,7 @@ from pipecat.frames.frames import (
     TTSTextFrame,
     TTSUpdateSettingsFrame,
 )
-from pipecat.processors.frame_processor import FrameDirection
+from pipecat.processors.frame_processor import FrameCallback, FrameDirection
 from pipecat.services.ai_service import AIService
 from pipecat.services.websocket_service import WebsocketService
 from pipecat.transcriptions.language import Language
@@ -393,6 +394,39 @@ class TTSService(AIService):
         else:
             await self.push_frame(frame, direction)
 
+    async def queue_frame(
+        self,
+        frame: Frame,
+        direction: FrameDirection = FrameDirection.DOWNSTREAM,
+        callback: Optional[FrameCallback] = None,
+    ):
+        """Queue a frame, letting upstream frames skip the speaking pause.
+
+        While the service is synthesizing, `_maybe_pause_frame_processing()`
+        blocks the non-system frame queue until `BotStoppedSpeakingFrame`. That
+        gate exists to keep *audio* from overlapping, and it only ever needs to
+        hold back the downstream text -> audio flow.
+
+        Upstream frames don't feed that flow, but they do have to traverse this
+        service: in the usual pipeline the assistant context aggregator sits
+        after the output transport, so the context frame it pushes upstream
+        after a function-call result must pass through here to reach the LLM.
+        Queuing it means the next inference cannot even start until playback of
+        the current utterance ends — and because a `TTSSpeakFrame` re-arms the
+        pause, it can take a second `BotStoppedSpeakingFrame` to release. The
+        bot ends up silently owing the user a reply for the length of an entire
+        utterance or more.
+
+        System frames already bypass the pause, so they keep the normal path
+        (ordering with interruptions matters); everything else travelling
+        upstream is processed immediately.
+        """
+        if direction == FrameDirection.UPSTREAM and not isinstance(frame, SystemFrame):
+            await self._process_frame_now(frame, direction, callback)
+            return
+
+        await super().queue_frame(frame, direction, callback)
+
     async def push_frame(self, frame: Frame, direction: FrameDirection = FrameDirection.DOWNSTREAM):
         """Push a frame downstream with TTS-specific handling.
 

dv_pipecat_ai-0.0.93.dev136/src/dv_pipecat_ai.egg-info/scm_version.json

@@ -1,8 +0,0 @@
-{
-  "tag": "0.0.92",
-  "distance": 136,
-  "node": "gc5eead3484ac34c95382dd840f21fea051699b29",
-  "dirty": false,
-  "branch": "dv-stage",
-  "node_date": "2026-08-31"
-}