From 9bf88bbf14eb15e50e84b8603b57eebbeeaadb14 Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 07:43:30 -0300
Subject: [PATCH 01/13] Fixing RivaTTSService error handler.

---
 src/pipecat/services/riva/tts.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/pipecat/services/riva/tts.py b/src/pipecat/services/riva/tts.py
index 07aa4d105..4d393943b 100644
--- a/src/pipecat/services/riva/tts.py
+++ b/src/pipecat/services/riva/tts.py
@@ -166,7 +166,6 @@ class RivaTTSService(TTSService):
                 add_response(None)
             except Exception as e:
                 logger.error(f"{self} exception: {e}")
-                yield ErrorFrame(error=f"{self} error: {e}")
                 add_response(None)
 
         await self.start_ttfb_metrics()
@@ -191,6 +190,7 @@ class RivaTTSService(TTSService):
                 resp = await asyncio.wait_for(queue.get(), timeout=RIVA_TTS_TIMEOUT_SECS)
         except asyncio.TimeoutError:
             logger.error(f"{self} timeout waiting for audio response")
+            yield ErrorFrame(error=f"{self} error: {e}")
 
         await self.start_tts_usage_metrics(text)
         yield TTSStoppedFrame()

From 77cd106795778fa22bcee3ba64588e7e79b9529a Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 09:54:08 -0300
Subject: [PATCH 02/13] Extracted the logic for retrying connections, and
 create a new send_with_retry method inside WebSocketService.

---
 src/pipecat/services/websocket_service.py | 74 +++++++++++++++++------
 1 file changed, 56 insertions(+), 18 deletions(-)

diff --git a/src/pipecat/services/websocket_service.py b/src/pipecat/services/websocket_service.py
index f9799b9c3..d19e6ad4d 100644
--- a/src/pipecat/services/websocket_service.py
+++ b/src/pipecat/services/websocket_service.py
@@ -36,6 +36,7 @@ class WebsocketService(ABC):
         """
         self._websocket: Optional[websockets.WebSocketClientProtocol] = None
         self._reconnect_on_error = reconnect_on_error
+        self._reconnect_in_progress: bool = False  # Add this flag
 
     async def _verify_connection(self) -> bool:
         """Verify the websocket connection is active and responsive.
@@ -66,6 +67,59 @@ class WebsocketService(ABC):
         await self._connect_websocket()
         return await self._verify_connection()
 
+    async def _try_reconnect(
+        self,
+        max_retries: int = 3,
+        report_error: Optional[Callable[[ErrorFrame], Awaitable[None]]] = None,
+    ) -> bool:
+        # Prevent concurrent reconnection attempts
+        if self._reconnect_in_progress:
+            logger.warning(f"{self} reconnect attempt aborted: already in progress")
+            return False
+
+        self._reconnect_in_progress = True
+        last_exception: Optional[Exception] = None
+        try:
+            for attempt in range(1, max_retries + 1):
+                try:
+                    logger.warning(f"{self} reconnecting, attempt {attempt}")
+                    if await self._reconnect_websocket(attempt):
+                        logger.info(f"{self} reconnected successfully on attempt {attempt}")
+                        return True
+                except Exception as e:
+                    last_exception = e
+                    logger.error(f"{self} reconnection attempt {attempt} failed: {e}")
+                    if report_error:
+                        await report_error(
+                            ErrorFrame(f"{self} reconnection attempt {attempt} failed: {e}")
+                        )
+                wait_time = exponential_backoff_time(attempt)
+                await asyncio.sleep(wait_time)
+            fatal_msg = f"{self} failed to reconnect after {max_retries} attempts"
+            if last_exception:
+                fatal_msg += f": {last_exception}"
+            logger.error(fatal_msg)
+            if report_error:
+                await report_error(ErrorFrame(fatal_msg, fatal=True))
+            return False
+        finally:
+            self._reconnect_in_progress = False
+
+    async def send_with_retry(self, message, report_error: Callable[[ErrorFrame], Awaitable[None]]):
+        """Attempt to send a message, retrying after reconnect if necessary."""
+        try:
+            await self._websocket.send(message)
+        except Exception as e:
+            logger.error(f"{self} send failed: {e}, will try to reconnect")
+            # Try to reconnect before retrying
+            success = await self._try_reconnect(report_error=report_error)
+            if success:
+                logger.info(f"{self} reconnected successfully, will retry send the message")
+                # trying to send the message one more time
+                await self._websocket.send(message)
+            else:
+                logger.error(f"{self} send failed; unable to reconnect")
+
     async def _receive_task_handler(self, report_error: Callable[[ErrorFrame], Awaitable[None]]):
         """Handle websocket message receiving with automatic retry logic.
 
@@ -76,13 +130,9 @@ class WebsocketService(ABC):
         Args:
             report_error: Callback function to report connection errors.
         """
-        retry_count = 0
-        MAX_RETRIES = 3
-
         while True:
             try:
                 await self._receive_messages()
-                retry_count = 0  # Reset counter on successful message receive
             except ConnectionClosedOK as e:
                 # Normal closure, don't retry
                 logger.debug(f"{self} connection closed normally: {e}")
@@ -92,21 +142,9 @@ class WebsocketService(ABC):
                 logger.error(message)
 
                 if self._reconnect_on_error:
-                    retry_count += 1
-                    if retry_count >= MAX_RETRIES:
-                        await report_error(ErrorFrame(message))
+                    success = await self._try_reconnect(report_error=report_error)
+                    if not success:
                         break
-
-                    logger.warning(f"{self} connection error, will retry: {e}")
-                    await report_error(ErrorFrame(message))
-
-                    try:
-                        if await self._reconnect_websocket(retry_count):
-                            retry_count = 0  # Reset counter on successful reconnection
-                        wait_time = exponential_backoff_time(retry_count)
-                        await asyncio.sleep(wait_time)
-                    except Exception as reconnect_error:
-                        logger.error(f"{self} reconnection failed: {reconnect_error}")
                 else:
                     await report_error(ErrorFrame(message))
                     break

From 19cc0177b85bea279692abb1d829567667d73027 Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 09:54:20 -0300
Subject: [PATCH 03/13] Refactored DeepgramFluxSTTService to automatically
 reconnect if sending a message fails.

---
 src/pipecat/services/deepgram/flux/stt.py | 114 ++++++++++++++++++----
 1 file changed, 95 insertions(+), 19 deletions(-)

diff --git a/src/pipecat/services/deepgram/flux/stt.py b/src/pipecat/services/deepgram/flux/stt.py
index ac6e8bb08..a0d45afe4 100644
--- a/src/pipecat/services/deepgram/flux/stt.py
+++ b/src/pipecat/services/deepgram/flux/stt.py
@@ -6,7 +6,9 @@
 
 """Deepgram Flux speech-to-text service implementation."""
 
+import asyncio
 import json
+import time
 from enum import Enum
 from typing import Any, AsyncGenerator, Dict, Optional
 from urllib.parse import urlencode
@@ -94,6 +96,7 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             mip_opt_out: Optional. Opts out requests from the Deepgram Model Improvement Program
                 (default False).
             tag: List of tags to label requests for identification during usage reporting.
+            min_confidence: Optional. Minimum confidence required confidence to create a TranscriptionFrame
         """
 
         eager_eot_threshold: Optional[float] = None
@@ -102,6 +105,7 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         keyterm: list = []
         mip_opt_out: Optional[bool] = None
         tag: list = []
+        min_confidence: Optional[float] = None  # New parameter
 
     def __init__(
         self,
@@ -163,6 +167,13 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         self._register_event_handler("on_end_of_turn")
         self._register_event_handler("on_eager_end_of_turn")
         self._register_event_handler("on_update")
+        self._connection_established_event = asyncio.Event()
+        # Watchdog task to prevent dangling tasks
+        # If we stop sending audio to Flux after we have received that the User has started speaking
+        # we never receive the user stopped speaking event unless we resume sending audio to it.
+        self._last_stt_time = None
+        self._watchdog_task = None
+        self._user_is_speaking = False
 
     async def _connect(self):
         """Connect to WebSocket and start background tasks.
@@ -172,9 +183,6 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         """
         await self._connect_websocket()
 
-        if self._websocket and not self._receive_task:
-            self._receive_task = self.create_task(self._receive_task_handler(self._report_error))
-
     async def _disconnect(self):
         """Disconnect from WebSocket and clean up tasks.
 
@@ -182,14 +190,7 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         and cleans up resources to prevent memory leaks.
         """
         try:
-            # Cancel background tasks BEFORE closing websocket
-            if self._receive_task:
-                await self.cancel_task(self._receive_task, timeout=2.0)
-                self._receive_task = None
-
-            # Now close the websocket
             await self._disconnect_websocket()
-
         except Exception as e:
             logger.error(f"{self} exception: {e}")
             await self.push_error(ErrorFrame(error=f"{self} error: {e}"))
@@ -197,6 +198,25 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             # Reset state only after everything is cleaned up
             self._websocket = None
 
+    async def _send_silence(self, duration_secs: float = 0.5):
+        """Send a block of silence of the specified duration (default 500 ms)."""
+        sample_width = 2  # bytes per sample for 16-bit PCM
+        num_channels = 1  # mono
+        num_samples = int(self.sample_rate * duration_secs)
+        silence = b"\x00" * (num_samples * sample_width * num_channels)
+        await self._websocket.send(silence)
+
+    async def _watchdog_task_handler(self):
+        while self._websocket and self._websocket.state is State.OPEN:
+            now = time.monotonic()
+            # More than 500 ms without sending new audio to Flux
+            if self._user_is_speaking and self._last_stt_time and now - self._last_stt_time > 0.5:
+                logger.warning("Sending silence to Flux to prevent dangling task")
+                await self._send_silence()
+                self._last_stt_time = time.monotonic()
+            # check every 100ms
+            await asyncio.sleep(0.1)
+
     async def _connect_websocket(self):
         """Establish WebSocket connection to API.
 
@@ -208,10 +228,26 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             if self._websocket and self._websocket.state is State.OPEN:
                 return
 
+            self._connection_established_event.clear()
+            self._user_is_speaking = False
             self._websocket = await websocket_connect(
                 self._websocket_url,
                 additional_headers={"Authorization": f"Token {self._api_key}"},
             )
+
+            # Creating the receiver task
+            if not self._receive_task:
+                self._receive_task = self.create_task(
+                    self._receive_task_handler(self._report_error)
+                )
+
+            # Creating the watchdog task
+            if not self._watchdog_task:
+                self._watchdog_task = self.create_task(self._watchdog_task_handler())
+
+            # Now wait for the connection established event
+            logger.debug("WebSocket connected, waiting for server confirmation...")
+            await self._connection_established_event.wait()
             logger.debug("Connected to Deepgram Flux Websocket")
             await self._call_event_handler("on_connected")
         except Exception as e:
@@ -227,6 +263,16 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         metrics collection. Handles disconnection errors gracefully.
         """
         try:
+            # Cancel background tasks BEFORE closing websocket
+            if self._receive_task:
+                await self.cancel_task(self._receive_task, timeout=2.0)
+                self._receive_task = None
+            if self._watchdog_task:
+                await self.cancel_task(self._watchdog_task, timeout=2.0)
+                self._watchdog_task = None
+                self._last_stt_time = None
+
+            self._connection_established_event.clear()
             await self.stop_all_metrics()
 
             if self._websocket:
@@ -340,7 +386,8 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             return
 
         try:
-            await self._websocket.send(audio)
+            self._last_stt_time = time.monotonic()
+            await self.send_with_retry(audio, self._report_error)
         except Exception as e:
             logger.error(f"{self} exception: {e}")
             yield ErrorFrame(error=f"{self} error: {e}")
@@ -463,6 +510,8 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         transcription processing.
         """
         logger.info("Connected to Flux - ready to stream audio")
+        # Notify connection is established
+        self._connection_established_event.set()
 
     async def _handle_fatal_error(self, data: Dict[str, Any]):
         """Handle fatal error messages from Deepgram Flux.
@@ -530,6 +579,7 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             transcript: maybe the first few words of the turn.
         """
         logger.debug("User started speaking")
+        self._user_is_speaking = True
         await self.push_interruption_task_frame_and_wait()
         await self.broadcast_frame(UserStartedSpeakingFrame)
         await self.start_metrics()
@@ -550,6 +600,22 @@ class DeepgramFluxSTTService(WebsocketSTTService):
         logger.trace(f"Received event TurnResumed: {event}")
         await self._call_event_handler("on_turn_resumed")
 
+    def _calculate_average_confidence(self, transcript_data) -> Optional[float]:
+        """Calculate the average confidence from transcript data.
+
+        Return None if the data is missing or invalid.
+        """
+        # Example: Assume transcript_data has a list of words with confidence
+        words = transcript_data.get("words")
+        if not words or not isinstance(words, list):
+            return None
+        confidences = [
+            w.get("confidence") for w in words if isinstance(w.get("confidence"), (float, int))
+        ]
+        if not confidences:
+            return None
+        return sum(confidences) / len(confidences)
+
     async def _handle_end_of_turn(self, transcript: str, data: Dict[str, Any]):
         """Handle EndOfTurn events from Deepgram Flux.
 
@@ -569,16 +635,26 @@ class DeepgramFluxSTTService(WebsocketSTTService):
             data: The TurnInfo message data containing event type, transcript and some extra metadata.
         """
         logger.debug("User stopped speaking")
+        self._user_is_speaking = False
 
-        await self.push_frame(
-            TranscriptionFrame(
-                transcript,
-                self._user_id,
-                time_now_iso8601(),
-                self._language,
-                result=data,
+        # Compute the average confidence
+        average_confidence = self._calculate_average_confidence(data)
+
+        if not self._params.min_confidence or average_confidence > self._params.min_confidence:
+            await self.push_frame(
+                TranscriptionFrame(
+                    transcript,
+                    self._user_id,
+                    time_now_iso8601(),
+                    self._language,
+                    result=data,
+                )
             )
-        )
+        else:
+            logger.warning(
+                f"Transcription confidence below min_confidence threshold: {average_confidence}"
+            )
+
         await self._handle_transcription(transcript, True, self._language)
         await self.stop_processing_metrics()
         await self.push_frame(UserStoppedSpeakingFrame(), FrameDirection.DOWNSTREAM)

From 04dbbabc036289fb3d90e5228a3e5af573d1ba21 Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 09:54:30 -0300
Subject: [PATCH 04/13] Introduced a minimum confidence parameter in
 DeepgramFluxSTTService to avoid generating transcriptions below a defined
 threshold.

---
 examples/foundational/07c-interruptible-deepgram-flux.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/examples/foundational/07c-interruptible-deepgram-flux.py b/examples/foundational/07c-interruptible-deepgram-flux.py
index f5b246acd..62579c2c5 100644
--- a/examples/foundational/07c-interruptible-deepgram-flux.py
+++ b/examples/foundational/07c-interruptible-deepgram-flux.py
@@ -52,7 +52,10 @@ transport_params = {
 async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     logger.info(f"Starting bot")
 
-    stt = DeepgramFluxSTTService(api_key=os.getenv("DEEPGRAM_API_KEY"))
+    stt = DeepgramFluxSTTService(
+        api_key=os.getenv("DEEPGRAM_API_KEY"),
+        params=DeepgramFluxSTTService.InputParams(min_confidence=0.3),
+    )
 
     tts = DeepgramTTSService(api_key=os.getenv("DEEPGRAM_API_KEY"), voice="aura-2-andromeda-en")
 

From b104a59b109168c01c0240ae885f7b3a0fd7e2ce Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 09:54:39 -0300
Subject: [PATCH 05/13] Mentioning the Deepgram Flux improvements in the
 changelog.

---
 CHANGELOG.md | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7cda253c7..90a7542de 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Added a watchdog to `DeepgramFluxSTTService` to prevent dangling tasks in case the
+  user was speaking and we stop receiving audio.
+
+- Introduced a minimum confidence parameter in `DeepgramFluxSTTService` to avoid
+  generating transcriptions below a defined threshold.
+
 - Added `ElevenLabsRealtimeSTTService` which implements the Realtime STT
   service from ElevenLabs.
 
@@ -18,6 +24,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- Extracted the logic for retrying connections, and create a new `send_with_retry`
+  method inside `WebSocketService`.
+
+- Refactored `DeepgramFluxSTTService` to automatically reconnect if sending a
+  message fails.
+
 - Updated all STT and TTS services to use consistent error handling pattern with
   `push_error()` method for better pipeline error event integration.
 

From f3b254e335be513623a6b30b80e1c60353d1f22e Mon Sep 17 00:00:00 2001
From: Paul Kompfner <paul@daily.co>
Date: Fri, 14 Nov 2025 21:39:18 -0500
Subject: [PATCH 06/13] D'oh! My TTS "inter-frame-spaces" logic was *way*
 overcomplicated (and fundamentally mistaken, though it happened to work)

Now:
- For TTS word-by-word output and `TTSSpeakFrames`: `TTSTextFrame`s' have `includes_inter_frame_spaces=False`.
- For all other TTS output: `TTSTextFrame` pass through the received text frames' `includes_inter_frame_spaces` value. So far, this value has always been `True`: LLMs send text chunks already containing all necessary spaces.
- `LLMTextFrame`s set `includes_inter_frame_spaces=False` at init time, per the aforementioned assumption.
---
 CHANGELOG.md                                  |  8 +---
 src/pipecat/frames/frames.py                  |  5 ++-
 src/pipecat/services/anthropic/llm.py         |  4 +-
 src/pipecat/services/asyncai/tts.py           | 18 --------
 src/pipecat/services/aws/llm.py               |  4 +-
 src/pipecat/services/aws/tts.py               |  9 ----
 src/pipecat/services/azure/tts.py             |  9 ----
 src/pipecat/services/deepgram/tts.py          | 18 --------
 src/pipecat/services/fish/tts.py              |  9 ----
 .../services/google/gemini_live/llm.py        |  2 -
 src/pipecat/services/google/llm.py            |  4 +-
 src/pipecat/services/google/tts.py            | 18 --------
 src/pipecat/services/groq/tts.py              |  9 ----
 src/pipecat/services/hume/tts.py              |  9 ----
 src/pipecat/services/inworld/tts.py           |  9 ----
 src/pipecat/services/lmnt/tts.py              |  9 ----
 src/pipecat/services/minimax/tts.py           |  9 ----
 src/pipecat/services/neuphonic/tts.py         | 18 --------
 src/pipecat/services/openai/base_llm.py       |  4 +-
 src/pipecat/services/openai/realtime/llm.py   |  2 -
 src/pipecat/services/openai/tts.py            |  9 ----
 src/pipecat/services/piper/tts.py             |  9 ----
 src/pipecat/services/rime/tts.py              |  9 ----
 src/pipecat/services/riva/tts.py              |  9 ----
 src/pipecat/services/sambanova/llm.py         |  4 +-
 src/pipecat/services/sarvam/tts.py            | 18 --------
 src/pipecat/services/speechmatics/tts.py      |  9 ----
 src/pipecat/services/tts_service.py           | 44 +++++++++----------
 28 files changed, 33 insertions(+), 255 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7cda253c7..6a4a64f2d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -12,10 +12,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added `ElevenLabsRealtimeSTTService` which implements the Realtime STT
   service from ElevenLabs.
 
-- Added a `TTSService.includes_inter_frame_spaces` property getter, so that TTS
-  services that subclass `TTSService` can indicate whether the text in the
-  `TTSTextFrame`s they push already contain any necessary inter-frame spaces.
-
 ### Changed
 
 - Updated all STT and TTS services to use consistent error handling pattern with
@@ -56,8 +52,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and 
-  example wiring; leverages the enhancement model for robust detection with no 
+- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and
+  example wiring; leverages the enhancement model for robust detection with no
   ONNX dependency or added processing complexity.
 
 ## [0.0.94] - 2025-11-10
diff --git a/src/pipecat/frames/frames.py b/src/pipecat/frames/frames.py
index 6f48f79f7..ddb4e5a14 100644
--- a/src/pipecat/frames/frames.py
+++ b/src/pipecat/frames/frames.py
@@ -352,7 +352,10 @@ class TextFrame(DataFrame):
 class LLMTextFrame(TextFrame):
     """Text frame generated by LLM services."""
 
-    pass
+    def __post_init__(self):
+        super().__post_init__()
+        # LLM services send text frames with all necessary spaces included
+        self.includes_inter_frame_spaces = True
 
 
 @dataclass
diff --git a/src/pipecat/services/anthropic/llm.py b/src/pipecat/services/anthropic/llm.py
index a5c9ee791..2e3e0272d 100644
--- a/src/pipecat/services/anthropic/llm.py
+++ b/src/pipecat/services/anthropic/llm.py
@@ -373,9 +373,7 @@ class AnthropicLLMService(LLMService):
 
                 if event.type == "content_block_delta":
                     if hasattr(event.delta, "text"):
-                        frame = LLMTextFrame(event.delta.text)
-                        frame.includes_inter_frame_spaces = True
-                        await self.push_frame(frame)
+                        await self.push_frame(LLMTextFrame(event.delta.text))
                         completion_tokens_estimate += self._estimate_tokens(event.delta.text)
                     elif hasattr(event.delta, "partial_json") and tool_use_block:
                         json_accumulator += event.delta.partial_json
diff --git a/src/pipecat/services/asyncai/tts.py b/src/pipecat/services/asyncai/tts.py
index 8df597c56..f916d9bdd 100644
--- a/src/pipecat/services/asyncai/tts.py
+++ b/src/pipecat/services/asyncai/tts.py
@@ -146,15 +146,6 @@ class AsyncAITTSService(InterruptibleTTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that AsyncAI TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that AsyncAI's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Async language format.
 
@@ -433,15 +424,6 @@ class AsyncAIHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that AsyncAI TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that AsyncAI's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Async language format.
 
diff --git a/src/pipecat/services/aws/llm.py b/src/pipecat/services/aws/llm.py
index 147a8c12a..ccbac43b7 100644
--- a/src/pipecat/services/aws/llm.py
+++ b/src/pipecat/services/aws/llm.py
@@ -1078,9 +1078,7 @@ class AWSBedrockLLMService(LLMService):
                     if "contentBlockDelta" in event:
                         delta = event["contentBlockDelta"]["delta"]
                         if "text" in delta:
-                            frame = LLMTextFrame(delta["text"])
-                            frame.includes_inter_frame_spaces = True
-                            await self.push_frame(frame)
+                            await self.push_frame(LLMTextFrame(delta["text"]))
                             completion_tokens_estimate += self._estimate_tokens(delta["text"])
                         elif "toolUse" in delta and "input" in delta["toolUse"]:
                             # Handle partial JSON for tool use
diff --git a/src/pipecat/services/aws/tts.py b/src/pipecat/services/aws/tts.py
index cbc35b123..f22c42399 100644
--- a/src/pipecat/services/aws/tts.py
+++ b/src/pipecat/services/aws/tts.py
@@ -209,15 +209,6 @@ class AWSPollyTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that AWS TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that AWS's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to AWS Polly language format.
 
diff --git a/src/pipecat/services/azure/tts.py b/src/pipecat/services/azure/tts.py
index 8ac8b70e3..a1040f312 100644
--- a/src/pipecat/services/azure/tts.py
+++ b/src/pipecat/services/azure/tts.py
@@ -151,15 +151,6 @@ class AzureBaseTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Azure TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Azure's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Azure language format.
 
diff --git a/src/pipecat/services/deepgram/tts.py b/src/pipecat/services/deepgram/tts.py
index f6045c1f3..f75d40b09 100644
--- a/src/pipecat/services/deepgram/tts.py
+++ b/src/pipecat/services/deepgram/tts.py
@@ -79,15 +79,6 @@ class DeepgramTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Deepgram TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Deepgram's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     @traced_tts
     async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         """Generate speech from text using Deepgram's TTS API.
@@ -177,15 +168,6 @@ class DeepgramHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Deepgram TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Deepgram's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     @traced_tts
     async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         """Generate speech from text using Deepgram's TTS API.
diff --git a/src/pipecat/services/fish/tts.py b/src/pipecat/services/fish/tts.py
index 0acb12a96..5fe129998 100644
--- a/src/pipecat/services/fish/tts.py
+++ b/src/pipecat/services/fish/tts.py
@@ -159,15 +159,6 @@ class FishAudioTTSService(InterruptibleTTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Fish Audio TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Fish Audio's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     async def set_model(self, model: str):
         """Set the TTS model and reconnect.
 
diff --git a/src/pipecat/services/google/gemini_live/llm.py b/src/pipecat/services/google/gemini_live/llm.py
index 11632968e..7e0b0f494 100644
--- a/src/pipecat/services/google/gemini_live/llm.py
+++ b/src/pipecat/services/google/gemini_live/llm.py
@@ -1452,8 +1452,6 @@ class GeminiLiveLLMService(LLMService):
             self._bot_text_buffer += text
             self._search_result_buffer += text  # Also accumulate for grounding
             frame = LLMTextFrame(text=text)
-            # Gemini Live text already includes any necessary inter-chunk spaces
-            frame.includes_inter_frame_spaces = True
             await self.push_frame(frame)
 
         # Check for grounding metadata in server content
diff --git a/src/pipecat/services/google/llm.py b/src/pipecat/services/google/llm.py
index ad5fd70a7..883932b76 100644
--- a/src/pipecat/services/google/llm.py
+++ b/src/pipecat/services/google/llm.py
@@ -920,9 +920,7 @@ class GoogleLLMService(LLMService):
                         for part in candidate.content.parts:
                             if not part.thought and part.text:
                                 search_result += part.text
-                                frame = LLMTextFrame(part.text)
-                                frame.includes_inter_frame_spaces = True
-                                await self.push_frame(frame)
+                                await self.push_frame(LLMTextFrame(part.text))
                             elif part.function_call:
                                 function_call = part.function_call
                                 id = function_call.id or str(uuid.uuid4())
diff --git a/src/pipecat/services/google/tts.py b/src/pipecat/services/google/tts.py
index 449b2bca2..cf03e2d52 100644
--- a/src/pipecat/services/google/tts.py
+++ b/src/pipecat/services/google/tts.py
@@ -596,15 +596,6 @@ class GoogleHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Google TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Google's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Google TTS language format.
 
@@ -803,15 +794,6 @@ class GoogleBaseTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Google and Gemini TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Google's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Google TTS language format.
 
diff --git a/src/pipecat/services/groq/tts.py b/src/pipecat/services/groq/tts.py
index d2dcb73a1..9026c4c4c 100644
--- a/src/pipecat/services/groq/tts.py
+++ b/src/pipecat/services/groq/tts.py
@@ -111,15 +111,6 @@ class GroqTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Groq TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Groq's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     @traced_tts
     async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         """Generate speech from text using Groq's TTS API.
diff --git a/src/pipecat/services/hume/tts.py b/src/pipecat/services/hume/tts.py
index 6760c8121..a3a7e9a4c 100644
--- a/src/pipecat/services/hume/tts.py
+++ b/src/pipecat/services/hume/tts.py
@@ -110,15 +110,6 @@ class HumeTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Hume TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Hume's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     async def start(self, frame: StartFrame) -> None:
         """Start the service.
 
diff --git a/src/pipecat/services/inworld/tts.py b/src/pipecat/services/inworld/tts.py
index 067ea6ab6..dc2282b91 100644
--- a/src/pipecat/services/inworld/tts.py
+++ b/src/pipecat/services/inworld/tts.py
@@ -250,15 +250,6 @@ class InworldTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Inworld TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Inworld's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     async def start(self, frame: StartFrame):
         """Start the Inworld TTS service.
 
diff --git a/src/pipecat/services/lmnt/tts.py b/src/pipecat/services/lmnt/tts.py
index d98d3f76c..ebcad0f20 100644
--- a/src/pipecat/services/lmnt/tts.py
+++ b/src/pipecat/services/lmnt/tts.py
@@ -124,15 +124,6 @@ class LmntTTSService(InterruptibleTTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that LMNT TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that LMNT's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to LMNT service language format.
 
diff --git a/src/pipecat/services/minimax/tts.py b/src/pipecat/services/minimax/tts.py
index b8e984eeb..05e2dac3b 100644
--- a/src/pipecat/services/minimax/tts.py
+++ b/src/pipecat/services/minimax/tts.py
@@ -194,15 +194,6 @@ class MiniMaxHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that MiniMax TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that MiniMax's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to MiniMax service language format.
 
diff --git a/src/pipecat/services/neuphonic/tts.py b/src/pipecat/services/neuphonic/tts.py
index 992abbe67..60b0ebcb1 100644
--- a/src/pipecat/services/neuphonic/tts.py
+++ b/src/pipecat/services/neuphonic/tts.py
@@ -151,15 +151,6 @@ class NeuphonicTTSService(InterruptibleTTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Neuphonic TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Neuphonic's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Neuphonic service language format.
 
@@ -449,15 +440,6 @@ class NeuphonicHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Neuphonic TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Neuphonic's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Neuphonic service language format.
 
diff --git a/src/pipecat/services/openai/base_llm.py b/src/pipecat/services/openai/base_llm.py
index 5a8c1ab31..d020e1106 100644
--- a/src/pipecat/services/openai/base_llm.py
+++ b/src/pipecat/services/openai/base_llm.py
@@ -390,9 +390,7 @@ class BaseOpenAILLMService(LLMService):
                     # Keep iterating through the response to collect all the argument fragments
                     arguments += tool_call.function.arguments
             elif chunk.choices[0].delta.content:
-                frame = LLMTextFrame(chunk.choices[0].delta.content)
-                frame.includes_inter_frame_spaces = True
-                await self.push_frame(frame)
+                await self.push_frame(LLMTextFrame(chunk.choices[0].delta.content))
 
             # When gpt-4o-audio / gpt-4o-mini-audio is used for llm or stt+llm
             # we need to get LLMTextFrame for the transcript
diff --git a/src/pipecat/services/openai/realtime/llm.py b/src/pipecat/services/openai/realtime/llm.py
index f66e6e8e1..8eaa3d6fa 100644
--- a/src/pipecat/services/openai/realtime/llm.py
+++ b/src/pipecat/services/openai/realtime/llm.py
@@ -678,8 +678,6 @@ class OpenAIRealtimeLLMService(LLMService):
         # the output modality is "text"
         if evt.delta:
             frame = LLMTextFrame(evt.delta)
-            # OpenAI Realtime text already includes any necessary inter-chunk spaces
-            frame.includes_inter_frame_spaces = True
             await self.push_frame(frame)
 
     async def _handle_evt_audio_transcript_delta(self, evt):
diff --git a/src/pipecat/services/openai/tts.py b/src/pipecat/services/openai/tts.py
index a5231170e..23cb75324 100644
--- a/src/pipecat/services/openai/tts.py
+++ b/src/pipecat/services/openai/tts.py
@@ -131,15 +131,6 @@ class OpenAITTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that OpenAI TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that OpenAI's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     async def set_model(self, model: str):
         """Set the TTS model to use.
 
diff --git a/src/pipecat/services/piper/tts.py b/src/pipecat/services/piper/tts.py
index 3752418b5..dd842ff11 100644
--- a/src/pipecat/services/piper/tts.py
+++ b/src/pipecat/services/piper/tts.py
@@ -66,15 +66,6 @@ class PiperTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Piper TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Piper's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     @traced_tts
     async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         """Generate speech from text using Piper's HTTP API.
diff --git a/src/pipecat/services/rime/tts.py b/src/pipecat/services/rime/tts.py
index 329aecd7d..7b62f20fa 100644
--- a/src/pipecat/services/rime/tts.py
+++ b/src/pipecat/services/rime/tts.py
@@ -501,15 +501,6 @@ class RimeHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Rime TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Rime's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> str | None:
         """Convert pipecat language to Rime language code.
 
diff --git a/src/pipecat/services/riva/tts.py b/src/pipecat/services/riva/tts.py
index 07aa4d105..6de6f9332 100644
--- a/src/pipecat/services/riva/tts.py
+++ b/src/pipecat/services/riva/tts.py
@@ -113,15 +113,6 @@ class RivaTTSService(TTSService):
             riva.client.proto.riva_tts_pb2.RivaSynthesisConfigRequest()
         )
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Riva TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Riva's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     async def set_model(self, model: str):
         """Attempt to set the TTS model.
 
diff --git a/src/pipecat/services/sambanova/llm.py b/src/pipecat/services/sambanova/llm.py
index 76f11e81c..5ed600457 100644
--- a/src/pipecat/services/sambanova/llm.py
+++ b/src/pipecat/services/sambanova/llm.py
@@ -176,9 +176,7 @@ class SambaNovaLLMService(OpenAILLMService):  # type: ignore
                     # Keep iterating through the response to collect all the argument fragments
                     arguments += tool_call.function.arguments
             elif chunk.choices[0].delta.content:
-                frame = LLMTextFrame(chunk.choices[0].delta.content)
-                frame.includes_inter_frame_spaces = True
-                await self.push_frame(frame)
+                await self.push_frame(LLMTextFrame(chunk.choices[0].delta.content))
 
             # When gpt-4o-audio / gpt-4o-mini-audio is used for llm or stt+llm
             # we need to get LLMTextFrame for the transcript
diff --git a/src/pipecat/services/sarvam/tts.py b/src/pipecat/services/sarvam/tts.py
index 6ce45d27d..127a0d589 100644
--- a/src/pipecat/services/sarvam/tts.py
+++ b/src/pipecat/services/sarvam/tts.py
@@ -195,15 +195,6 @@ class SarvamHttpTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Sarvam TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Sarvam's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Sarvam AI language format.
 
@@ -467,15 +458,6 @@ class SarvamTTSService(InterruptibleTTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Sarvam TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Sarvam's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     def language_to_service_language(self, language: Language) -> Optional[str]:
         """Convert a Language enum to Sarvam AI language format.
 
diff --git a/src/pipecat/services/speechmatics/tts.py b/src/pipecat/services/speechmatics/tts.py
index e115d5a7c..b8fe172e7 100644
--- a/src/pipecat/services/speechmatics/tts.py
+++ b/src/pipecat/services/speechmatics/tts.py
@@ -105,15 +105,6 @@ class SpeechmaticsTTSService(TTSService):
         """
         return True
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates that Speechmatics TTSTextFrames include necessary inter-frame spaces.
-
-        Returns:
-            True, indicating that Speechmatics's text frames include necessary inter-frame spaces.
-        """
-        return True
-
     @traced_tts
     async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         """Generate speech from text using Speechmatics' HTTP API.
diff --git a/src/pipecat/services/tts_service.py b/src/pipecat/services/tts_service.py
index 29c54f497..f0d602a40 100644
--- a/src/pipecat/services/tts_service.py
+++ b/src/pipecat/services/tts_service.py
@@ -142,6 +142,7 @@ class TTSService(AIService):
         self._voice_id: str = ""
         self._settings: Dict[str, Any] = {}
         self._text_aggregator: BaseTextAggregator = text_aggregator or SimpleTextAggregator()
+        self._aggregated_text_includes_inter_frame_spaces: bool = False
         self._text_filters: Sequence[BaseTextFilter] = text_filters or []
         self._transport_destination: Optional[str] = transport_destination
         self._tracing_enabled: bool = False
@@ -192,23 +193,6 @@ class TTSService(AIService):
         CHUNK_SECONDS = 0.5
         return int(self.sample_rate * CHUNK_SECONDS * 2)  # 2 bytes/sample
 
-    @property
-    def includes_inter_frame_spaces(self) -> bool:
-        """Indicates whether TTSTextFrames include necesary inter-frame spaces.
-
-        When True, the TTSTextFrame objects pushed by this service already
-        include all necessary spaces between subsequent frames. When False,
-        downstream processors (like the assistant context aggregator) may need
-        to add spacing.
-
-        Subclasses should override this property to return True if their text
-        generation process already includes necessary inter-frame spaces.
-
-        Returns:
-            False by default. Subclasses can override to return True.
-        """
-        return False
-
     async def set_model(self, model: str):
         """Set the TTS model to use.
 
@@ -369,9 +353,16 @@ class TTSService(AIService):
             await self._maybe_pause_frame_processing()
 
             sentence = self._text_aggregator.text
+            includes_inter_frame_spaces = self._aggregated_text_includes_inter_frame_spaces
+
+            # Reset aggregator state
             await self._text_aggregator.reset()
             self._processing_text = False
-            await self._push_tts_frames(sentence)
+            self._aggregated_text_includes_inter_frame_spaces = False
+
+            await self._push_tts_frames(
+                sentence, includes_inter_frame_spaces=includes_inter_frame_spaces
+            )
             if isinstance(frame, LLMFullResponseEndFrame):
                 if self._push_text_frames:
                     await self.push_frame(frame, direction)
@@ -380,7 +371,8 @@ class TTSService(AIService):
         elif isinstance(frame, TTSSpeakFrame):
             # Store if we were processing text or not so we can set it back.
             processing_text = self._processing_text
-            await self._push_tts_frames(frame.text)
+            # Assumption: text in TTSSpeakFrame does not include inter-frame spaces
+            await self._push_tts_frames(frame.text, includes_inter_frame_spaces=False)
             # We pause processing incoming frames because we are sending data to
             # the TTS. We pause to avoid audio overlapping.
             await self._maybe_pause_frame_processing()
@@ -474,11 +466,17 @@ class TTSService(AIService):
             text = frame.text
         else:
             text = await self._text_aggregator.aggregate(frame.text)
+            # Assumption: whether inter-frame spaces are included shouldn't
+            # change during aggregation, so we can just use the latest frame's
+            # value
+            self._aggregated_text_includes_inter_frame_spaces = frame.includes_inter_frame_spaces
 
         if text:
-            await self._push_tts_frames(text)
+            await self._push_tts_frames(
+                text, includes_inter_frame_spaces=frame.includes_inter_frame_spaces
+            )
 
-    async def _push_tts_frames(self, text: str):
+    async def _push_tts_frames(self, text: str, includes_inter_frame_spaces: bool):
         # Remove leading newlines only
         text = text.lstrip("\n")
 
@@ -508,7 +506,7 @@ class TTSService(AIService):
             # We send the original text after the audio. This way, if we are
             # interrupted, the text is not added to the assistant context.
             frame = TTSTextFrame(text)
-            frame.includes_inter_frame_spaces = self.includes_inter_frame_spaces
+            frame.includes_inter_frame_spaces = includes_inter_frame_spaces
             await self.push_frame(frame)
 
     async def _stop_frame_handler(self):
@@ -635,6 +633,8 @@ class WordTTSService(TTSService):
                 frame = TTSStoppedFrame()
                 frame.pts = last_pts
             else:
+                # Assumption: word-by-word text frames don't include spaces, so
+                # we can rely on the default includes_inter_frame_spaces=False
                 frame = TTSTextFrame(word)
                 frame.pts = self._initial_word_timestamp + timestamp
             if frame:

From 0c3c26b7b87863ecc79f148ad394adc4f939a31b Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 15:20:09 -0300
Subject: [PATCH 07/13] Passing the custom request_data to the
 SmallWebRTCRunnerArguments body.

---
 src/pipecat/runner/run.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/src/pipecat/runner/run.py b/src/pipecat/runner/run.py
index 55c70ed8a..ebca467df 100644
--- a/src/pipecat/runner/run.py
+++ b/src/pipecat/runner/run.py
@@ -264,7 +264,10 @@ def _setup_webrtc_routes(
         # Prepare runner arguments with the callback to run your bot
         async def webrtc_connection_callback(connection):
             bot_module = _get_bot_module()
-            runner_args = SmallWebRTCRunnerArguments(webrtc_connection=connection)
+
+            runner_args = SmallWebRTCRunnerArguments(
+                webrtc_connection=connection, body=request.request_data
+            )
             background_tasks.add_task(bot_module.bot, runner_args)
 
         # Delegate handling to SmallWebRTCRequestHandler
@@ -326,7 +329,8 @@ def _setup_webrtc_routes(
                         type=request_data["type"],
                         pc_id=request_data.get("pc_id"),
                         restart_pc=request_data.get("restart_pc"),
-                        request_data=request_data,
+                        request_data=request_data.get("request_data")
+                        or request_data.get("requestData"),
                     )
                     return await offer(webrtc_request, background_tasks)
                 elif request.method == HTTPMethod.PATCH.value:

From 74154b26a2a97840bd60a4f8ddb676da7a26f77f Mon Sep 17 00:00:00 2001
From: Filipi Fuchter <filipi87@gmail.com>
Date: Mon, 17 Nov 2025 15:39:07 -0300
Subject: [PATCH 08/13] Mentioning the SmallWebRTCTransport fix in the readme.

---
 CHANGELOG.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 33f7fdc6b..bca4a5d6f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -48,6 +48,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- Fixed an issue in the `Runner` where, when using `SmallWebRTCTransport`, the
+  `request_data` was not being passed to the `SmallWebRTCRunnerArguments` body.
+
 - Fixed subtle issue of assistant context messages ending up with double spaces
   between words or sentences.
 

From 7eedb33d5076de8584452dd311c844bb46122cc6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Conchillo=20Flaqu=C3=A9?= <aleix@daily.co>
Date: Mon, 17 Nov 2025 11:19:04 -0800
Subject: [PATCH 09/13] BaseTextFilter: only require subclasses to implement
 filter()

---
 CHANGELOG.md                               | 12 ++++++------
 src/pipecat/utils/text/base_text_filter.py |  3 ---
 2 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index bca4a5d6f..d4fce0917 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -18,8 +18,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added `ElevenLabsRealtimeSTTService` which implements the Realtime STT
   service from ElevenLabs.
 
+- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and
+  example wiring; leverages the enhancement model for robust detection with no
+  ONNX dependency or added processing complexity.
+
 ### Changed
 
+- `BaseTextFilter` only require subclasses to implement the `filter()` method.
+
 - Extracted the logic for retrying connections, and create a new `send_with_retry`
   method inside `WebSocketService`.
 
@@ -65,12 +71,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Prevented `HeyGenVideoService` from automatically disconnecting after 5 minutes.
 
-### Added
-
-- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and
-  example wiring; leverages the enhancement model for robust detection with no
-  ONNX dependency or added processing complexity.
-
 ## [0.0.94] - 2025-11-10
 
 ### Changed
diff --git a/src/pipecat/utils/text/base_text_filter.py b/src/pipecat/utils/text/base_text_filter.py
index 1a18a38a6..0ede2137c 100644
--- a/src/pipecat/utils/text/base_text_filter.py
+++ b/src/pipecat/utils/text/base_text_filter.py
@@ -26,7 +26,6 @@ class BaseTextFilter(ABC):
     behavior, settings management, and interruption handling logic.
     """
 
-    @abstractmethod
     async def update_settings(self, settings: Mapping[str, Any]):
         """Update the filter's configuration settings.
 
@@ -53,7 +52,6 @@ class BaseTextFilter(ABC):
         """
         pass
 
-    @abstractmethod
     async def handle_interruption(self):
         """Handle interruption events in the processing pipeline.
 
@@ -62,7 +60,6 @@ class BaseTextFilter(ABC):
         """
         pass
 
-    @abstractmethod
     async def reset_interruption(self):
         """Reset the filter state after an interruption has been handled.
 

From 5095fc6a64da534abfedab0c438761288547087e Mon Sep 17 00:00:00 2001
From: Paul Kompfner <paul@daily.co>
Date: Mon, 17 Nov 2025 15:16:19 -0500
Subject: [PATCH 10/13] Update Moondream example so that Moondream service
 output makes it into the context, even if the TTS service is disabled

---
 .../14d-function-calling-moondream-video.py   | 40 +++++++++++++++++--
 1 file changed, 37 insertions(+), 3 deletions(-)

diff --git a/examples/foundational/14d-function-calling-moondream-video.py b/examples/foundational/14d-function-calling-moondream-video.py
index 6aeb2b892..9544818b9 100644
--- a/examples/foundational/14d-function-calling-moondream-video.py
+++ b/examples/foundational/14d-function-calling-moondream-video.py
@@ -15,14 +15,21 @@ from pipecat.audio.turn.smart_turn.base_smart_turn import SmartTurnParams
 from pipecat.audio.turn.smart_turn.local_smart_turn_v3 import LocalSmartTurnAnalyzerV3
 from pipecat.audio.vad.silero import SileroVADAnalyzer
 from pipecat.audio.vad.vad_analyzer import VADParams
-from pipecat.frames.frames import LLMRunFrame, UserImageRequestFrame
+from pipecat.frames.frames import (
+    Frame,
+    LLMFullResponseEndFrame,
+    LLMFullResponseStartFrame,
+    LLMRunFrame,
+    TextFrame,
+    UserImageRequestFrame,
+)
 from pipecat.pipeline.parallel_pipeline import ParallelPipeline
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.runner import PipelineRunner
 from pipecat.pipeline.task import PipelineTask
 from pipecat.processors.aggregators.llm_context import LLMContext
 from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair
-from pipecat.processors.frame_processor import FrameDirection
+from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
 from pipecat.runner.types import RunnerArguments
 from pipecat.runner.utils import (
     create_transport,
@@ -66,6 +73,27 @@ async def fetch_user_image(params: FunctionCallParams):
     # await params.result_callback({"result": "Image is being captured."})
 
 
+class MoondreamTextFrameWrapper(FrameProcessor):
+    """Wraps Moondream-provided TextFrames with LLM response start/end frames.
+
+    This processor detects TextFrames and automatically wraps them with
+    LLMFullResponseStartFrame and LLMFullResponseEndFrame to provide proper
+    response boundaries for downstream processors.
+    """
+
+    async def process_frame(self, frame: Frame, direction: FrameDirection):
+        await super().process_frame(frame, direction)
+
+        # If we receive a TextFrame, wrap it with response start/end frames
+        if isinstance(frame, TextFrame):
+            await self.push_frame(LLMFullResponseStartFrame(), direction)
+            await self.push_frame(frame, direction)
+            await self.push_frame(LLMFullResponseEndFrame(), direction)
+        else:
+            # For all other frames, just pass them through
+            await self.push_frame(frame, direction)
+
+
 # We store functions so objects (e.g. SileroVADAnalyzer) don't get
 # instantiated. The function will be called when the desired transport gets
 # selected.
@@ -130,6 +158,12 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     # If you run into weird description, try with use_cpu=True
     moondream = MoondreamService()
 
+    # Wrap TextFrames with LLM response start/end frames, which makes Moondream
+    # output be treated like LLM responses for the purpose of context
+    # aggregation. Without this, the assistant context aggregator would ignore
+    # Moondream output (if the TTS service is disabled).
+    moondream_text_wrapper = MoondreamTextFrameWrapper()
+
     pipeline = Pipeline(
         [
             transport.input(),  # Transport user input
@@ -137,7 +171,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
             context_aggregator.user(),  # User responses
             ParallelPipeline(
                 [llm],  # LLM
-                [moondream],
+                [moondream, moondream_text_wrapper],
             ),
             tts,  # TTS
             transport.output(),  # Transport bot output

From 4835617b16a1f4a5fbbb2f70562454c75a568573 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Conchillo=20Flaqu=C3=A9?= <aleix@daily.co>
Date: Fri, 14 Nov 2025 15:20:01 -0800
Subject: [PATCH 11/13] ConsumerProcessor: queue frames internally instead of
 pushing them

---
 CHANGELOG.md                                 | 4 ++++
 src/pipecat/processors/consumer_processor.py | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index d4fce0917..04d789370 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -24,6 +24,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- `ConsumerProcessor` now queues frames from the producer internally instead of
+  pushing them directly. This allows us to subclass consumer processors and
+  manipulate frames before they are pushed.
+
 - `BaseTextFilter` only require subclasses to implement the `filter()` method.
 
 - Extracted the logic for retrying connections, and create a new `send_with_retry`
diff --git a/src/pipecat/processors/consumer_processor.py b/src/pipecat/processors/consumer_processor.py
index 5445b492d..3654194ec 100644
--- a/src/pipecat/processors/consumer_processor.py
+++ b/src/pipecat/processors/consumer_processor.py
@@ -83,4 +83,4 @@ class ConsumerProcessor(FrameProcessor):
         while True:
             frame = await self._queue.get()
             new_frame = await self._transformer(frame)
-            await self.push_frame(new_frame, self._direction)
+            await self.queue_frame(new_frame, self._direction)

From 3132e1226571228bda6fc6a323128a2c3095e003 Mon Sep 17 00:00:00 2001
From: Mark Backman <mark@daily.co>
Date: Fri, 14 Nov 2025 10:48:52 -0500
Subject: [PATCH 12/13] Add camera and screen capture support to dev runner for
 SmallWebRTC

---
 CHANGELOG.md                |  8 ++++++++
 src/pipecat/runner/utils.py | 16 ++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 04d789370..de1f520a7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Added ai-coustics integrated VAD (`AICVADAnalyzer`) with `AICFilter` factory and
+  example wiring; leverages the enhancement model for robust detection with no
+  ONNX dependency or added processing complexity.
+
 - Added a watchdog to `DeepgramFluxSTTService` to prevent dangling tasks in case the
   user was speaking and we stop receiving audio.
 
@@ -39,6 +43,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Updated all STT and TTS services to use consistent error handling pattern with
   `push_error()` method for better pipeline error event integration.
 
+- Added support for `maybe_capture_participant_camera()` and
+  `maybe_capture_participant_screen()` for `SmallWebRTCTransport` in the runner
+  utils.
+
 - Added Hindi support for Rime TTS services.
 
 - Updated `GeminiTTSService` to use Google Cloud Text-to-Speech streaming API
diff --git a/src/pipecat/runner/utils.py b/src/pipecat/runner/utils.py
index f9fd0c14a..76a6fa82f 100644
--- a/src/pipecat/runner/utils.py
+++ b/src/pipecat/runner/utils.py
@@ -281,6 +281,14 @@ async def maybe_capture_participant_camera(
     except ImportError:
         pass
 
+    try:
+        from pipecat.transports.smallwebrtc.transport import SmallWebRTCTransport
+
+        if isinstance(transport, SmallWebRTCTransport):
+            await transport.capture_participant_video(video_source="camera")
+    except ImportError:
+        pass
+
 
 async def maybe_capture_participant_screen(
     transport: BaseTransport, client: Any, framerate: int = 0
@@ -303,6 +311,14 @@ async def maybe_capture_participant_screen(
     except ImportError:
         pass
 
+    try:
+        from pipecat.transports.smallwebrtc.transport import SmallWebRTCTransport
+
+        if isinstance(transport, SmallWebRTCTransport):
+            await transport.capture_participant_video(video_source="screenVideo")
+    except ImportError:
+        pass
+
 
 def _smallwebrtc_sdp_cleanup_ice_candidates(text: str, pattern: str) -> str:
     """Clean up ICE candidates in SDP text for SmallWebRTC.

From 9f45ad4d2e2ec9713a95c9d7c39b2bba5dd258cf Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aleix=20Conchillo=20Flaqu=C3=A9?= <aleix@daily.co>
Date: Tue, 18 Nov 2025 08:38:45 -0800
Subject: [PATCH 13/13] LLMContext: create_image_message/create_audio_message
 are now async

---
 CHANGELOG.md                                  |  5 ++
 .../foundational/12-describe-image-openai.py  |  2 +-
 .../12a-describe-image-anthropic.py           |  2 +-
 .../foundational/12b-describe-image-aws.py    |  2 +-
 .../12c-describe-image-gemini-flash.py        |  2 +-
 .../processors/aggregators/llm_context.py     | 46 ++++++++++++-------
 6 files changed, 38 insertions(+), 21 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index de1f520a7..82da28b26 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -28,6 +28,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- ⚠️ Breaking change: `LLMContext.create_image_message()` and
+  `LLMContext.create_audio_message()` are now async methods. This fixes and
+  issue where the asyncio event loop would be blocked while encoding audio or
+  images.
+
 - `ConsumerProcessor` now queues frames from the producer internally instead of
   pushing them directly. This allows us to subclass consumer processors and
   manipulate frames before they are pushed.
diff --git a/examples/foundational/12-describe-image-openai.py b/examples/foundational/12-describe-image-openai.py
index 8c72075e8..477803da6 100644
--- a/examples/foundational/12-describe-image-openai.py
+++ b/examples/foundational/12-describe-image-openai.py
@@ -110,7 +110,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
         # Kick off the conversation.
         image = Image.open(image_path)
-        message = LLMContext.create_image_message(
+        message = await LLMContext.create_image_message(
             image=image.tobytes(),
             format="RGB",
             size=image.size,
diff --git a/examples/foundational/12a-describe-image-anthropic.py b/examples/foundational/12a-describe-image-anthropic.py
index 6a9891712..ac4e8f01c 100644
--- a/examples/foundational/12a-describe-image-anthropic.py
+++ b/examples/foundational/12a-describe-image-anthropic.py
@@ -110,7 +110,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
         # Kick off the conversation.
         image = Image.open(image_path)
-        message = LLMContext.create_image_message(
+        message = await LLMContext.create_image_message(
             image=image.tobytes(),
             format="RGB",
             size=image.size,
diff --git a/examples/foundational/12b-describe-image-aws.py b/examples/foundational/12b-describe-image-aws.py
index 441c49cfd..cf1ce66a0 100644
--- a/examples/foundational/12b-describe-image-aws.py
+++ b/examples/foundational/12b-describe-image-aws.py
@@ -117,7 +117,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
         # Kick off the conversation.
         image = Image.open(image_path)
-        message = LLMContext.create_image_message(
+        message = await LLMContext.create_image_message(
             image=image.tobytes(),
             format="RGB",
             size=image.size,
diff --git a/examples/foundational/12c-describe-image-gemini-flash.py b/examples/foundational/12c-describe-image-gemini-flash.py
index 919bf3553..bfd7f5146 100644
--- a/examples/foundational/12c-describe-image-gemini-flash.py
+++ b/examples/foundational/12c-describe-image-gemini-flash.py
@@ -110,7 +110,7 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
 
         # Kick off the conversation.
         image = Image.open(image_path)
-        message = LLMContext.create_image_message(
+        message = await LLMContext.create_image_message(
             image=image.tobytes(),
             format="RGB",
             size=image.size,
diff --git a/src/pipecat/processors/aggregators/llm_context.py b/src/pipecat/processors/aggregators/llm_context.py
index d9280f9c0..b9216103a 100644
--- a/src/pipecat/processors/aggregators/llm_context.py
+++ b/src/pipecat/processors/aggregators/llm_context.py
@@ -14,6 +14,7 @@ translation from this universal context into whatever format it needs, using a
 service-specific adapter.
 """
 
+import asyncio
 import base64
 import io
 import wave
@@ -137,7 +138,7 @@ class LLMContext:
         return {"role": role, "content": content}
 
     @staticmethod
-    def create_image_message(
+    async def create_image_message(
         *,
         role: str = "user",
         format: str,
@@ -154,15 +155,21 @@ class LLMContext:
             image: Raw image bytes.
             text: Optional text to include with the image.
         """
-        buffer = io.BytesIO()
-        Image.frombytes(format, size, image).save(buffer, format="JPEG")
-        encoded_image = base64.b64encode(buffer.getvalue()).decode("utf-8")
+
+        def encode_image():
+            buffer = io.BytesIO()
+            Image.frombytes(format, size, image).save(buffer, format="JPEG")
+            encoded_image = base64.b64encode(buffer.getvalue()).decode("utf-8")
+            return encoded_image
+
+        encoded_image = await asyncio.to_thread(encode_image)
+
         url = f"data:image/jpeg;base64,{encoded_image}"
 
         return LLMContext.create_image_url_message(role=role, url=url, text=text)
 
     @staticmethod
-    def create_audio_message(
+    async def create_audio_message(
         *, role: str = "user", audio_frames: list[AudioRawFrame], text: str = "Audio follows"
     ) -> LLMContextMessage:
         """Create a context message containing audio.
@@ -172,21 +179,26 @@ class LLMContext:
             audio_frames: List of audio frame objects to include.
             text: Optional text to include with the audio.
         """
-        sample_rate = audio_frames[0].sample_rate
-        num_channels = audio_frames[0].num_channels
 
-        content = []
-        content.append({"type": "text", "text": text})
-        data = b"".join(frame.audio for frame in audio_frames)
+        def encode_audio():
+            sample_rate = audio_frames[0].sample_rate
+            num_channels = audio_frames[0].num_channels
 
-        with io.BytesIO() as buffer:
-            with wave.open(buffer, "wb") as wf:
-                wf.setsampwidth(2)
-                wf.setnchannels(num_channels)
-                wf.setframerate(sample_rate)
-                wf.writeframes(data)
+            content = []
+            content.append({"type": "text", "text": text})
+            data = b"".join(frame.audio for frame in audio_frames)
 
-        encoded_audio = base64.b64encode(buffer.getvalue()).decode("utf-8")
+            with io.BytesIO() as buffer:
+                with wave.open(buffer, "wb") as wf:
+                    wf.setsampwidth(2)
+                    wf.setnchannels(num_channels)
+                    wf.setframerate(sample_rate)
+                    wf.writeframes(data)
+
+            encoded_audio = base64.b64encode(buffer.getvalue()).decode("utf-8")
+            return encoded_audio
+
+        encoded_audio = asyncio.to_thread(encode_audio)
 
         content.append(
             {