Merge pull request #3440 from pipecat-ai/changelog-0.0.99

Release 0.0.99 - Changelog Update
2026-01-13 17:02:41 -08:00
parent 6fd5847f84 7e1b4a4e90
commit 86ed485711
74 changed files with 512 additions and 230 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,518 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 <!-- towncrier release notes start -->
 ## [0.0.99] - 2026-01-13
 ### Added
 - Introducing user turn strategies. User turn strategies indicate when the user
  turn starts or stops. In conversational agents, these are often referred to
  as start/stop speaking or turn-taking plans or policies.
  User turn start strategies indicate when the user starts speaking (e.g.
  using VAD events or when a user says one or more words).
  User turn stop strategies indicate when the user stops speaking (e.g. using
  an end-of-turn detection model or by observing incoming transcriptions).
  A list of strategies can be specified for both strategies; strategies are
  evaluated in order until one evaluates to true.
    Available user turn start strategies:
      - VADUserTurnStartStrategy
      - TranscriptionUserTurnStartStrategy
      - MinWordsUserTurnStartStrategy
      - ExternalUserTurnStartStrategy
    Available user turn stop strategies:
      - TranscriptionUserTurnStopStrategy
      - TurnAnalyzerUserTurnStopStrategy
      - ExternalUserTurnStopStrategy
    The default strategies are:
      - start: [VADUserTurnStartStrategy, TranscriptionUserTurnStartStrategy]
      - stop: [TranscriptionUserTurnStopStrategy]
  urn strategies are configured when setting up `LLMContextAggregatorPair`.
  For example:
    ```python
    context_aggregator = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(
            user_turn_strategies=UserTurnStrategies(
                stop=[
                    TurnAnalyzerUserTurnStopStrategy(
 turn_analyzer=LocalSmartTurnAnalyzerV3(params=SmartTurnParams())
                    )
                ],
            )
        ),
    )
    ```
  In order to use the user turn strategies you must update to the new
  universal `LLMContext` and `LLMContextAggregatorPair`.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - Added `RNNoiseFilter` for real-time noise suppression using RNNoise neural
  network via pyrnnoise library.
  (PR [#3205](https://github.com/pipecat-ai/pipecat/pull/3205))
 - Added `GrokRealtimeLLMService` for xAI's Grok Voice Agent API with real-time
  voice conversations:
    - Support for real-time audio streaming with WebSocket connection
    - Built-in server-side VAD (Voice Activity Detection)
    - Multiple voice options: Ara, Rex, Sal, Eve, Leo
    - Built-in tools support: web_search, x_search, file_search
    - Custom function calling with standard Pipecat tools schema
    - Configurable audio formats (PCM at 8kHz-48kHz)
  (PR [#3267](https://github.com/pipecat-ai/pipecat/pull/3267))
 - Added an approximation of TTFB for Ultravox.
  (PR [#3268](https://github.com/pipecat-ai/pipecat/pull/3268))
 - Added a new `AudioContextTTSService` to the TTS service base classes. The
  `AudioContextWordTTSService` now inherits from `AudioContextTTSService` and
  `WebsocketWordTTSService`.
  (PR [#3289](https://github.com/pipecat-ai/pipecat/pull/3289))
 - `LLMUserAggregator` now exposes the following events:
     - `on_user_turn_started`: triggered when a user turn starts
     - `on_user_turn_stopped`: triggered when a user turn ends
     - `on_user_turn_stop_timeout`: triggered when a user turn does not stop
        and times out
  (PR [#3291](https://github.com/pipecat-ai/pipecat/pull/3291))
 - Introducing user mute strategies. User mute strategies indicate when user
  input should be muted based on the current system state.
  In conversational agents, user mute strategies are used to prevent user
  input from interrupting bot speech, tool execution, or other critical system
  operations.
  A list of strategies can be specified; all strategies are evaluated for
  every frame so that each strategy can maintain its internal state. A user
  frame is muted if any of the configured strategies indicates it should be
  muted.
    Available user mute strategies:
    * `FirstSpeechUserMuteStrategy`
    * `MuteUntilFirstBotCompleteUserMuteStrategy`
    * `AlwaysUserMuteStrategy`
    * `FunctionCallUserMuteStrategy`
  User mute strategies replace the legacy `STTMuteFilter` and provide a more
  flexible and composable approach to muting user input.
  User mute strategies are configured when setting up the
  `LLMContextAggregatorPair`. For example:
    ```python
    context_aggregator = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(
            user_mute_strategies=[
                FirstSpeechUserMuteStrategy(),
            ]
        ),
    )
    ```
  In order to use user mute strategies you should update to the new universal
  `LLMContext` and `LLMContextAggregatorPair`.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))
 - Added `use_ssl` parameter to `NvidiaSTTService`, `NvidiaSegmentedSTTService`
  and `NvidiaTTSService`.
  (PR [#3300](https://github.com/pipecat-ai/pipecat/pull/3300))
 - Added `enable_interruptions` constructor argument to all user turn
  strategies. This tells the `LLMUserAggregator` to push or not push an
  `InterruptionFrame`.
  (PR [#3316](https://github.com/pipecat-ai/pipecat/pull/3316))
 - Added `split_sentences` parameter to `SpeechmaticsSTTService` to control
  sentence splitting behavior for finals on sentence boundaries.
  (PR [#3328](https://github.com/pipecat-ai/pipecat/pull/3328))
 - Added word-level timestamp support to `AzureTTSService` for accurate
  text-to-audio synchronization.
  (PR [#3334](https://github.com/pipecat-ai/pipecat/pull/3334))
 - Added `pronunciation_dict_id` parameter to `CartesiaTTSService.InputParams`
  and `CartesiaHttpTTSService.InputParams` to support Cartesia's pronunciation
  dictionary feature for custom pronunciations.
  (PR [#3346](https://github.com/pipecat-ai/pipecat/pull/3346))
 - Added support for using the HeyGen LiveAvatar API with the `HeyGenTransport`
  (see https://www.liveavatar.com/).
  (PR [#3357](https://github.com/pipecat-ai/pipecat/pull/3357))
 - Added image support to `OpenAIRealtimeLLMService` via `InputImageRawFrame`:
    - New `start_video_paused` parameter to control initial video input state
    - New `video_frame_detail` parameter to set image processing quality
  ("auto",
      "low", or "high"). This corresponds to OpenAI Realtime's `image_detail`
      parameter.
    - `set_video_input_paused()` method to pause/resume video input at runtime
    - `set_video_frame_detail()` method to adjust video frame quality
       dynamically
    - Automatic rate limiting (1 frame per second) to prevent API overload
  (PR [#3360](https://github.com/pipecat-ai/pipecat/pull/3360))
 - Added `UserTurnProcessor`, a frame processor built on `UserTurnController`
  that pushes `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` frames
  and interruptions based on the controller's user turn strategies.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))
 - Added `UserTurnController` to manage user turns. It emits
  `on_user_turn_started`, `on_user_turn_stopped`, and
  `on_user_turn_stop_timeout` events, and can be integrated into processors to
  detect and handle user turns. `LLMUserAggregator` and `UserTurnProcessor` are
  implemented using this controller.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))
 - Added `should_interrupt` property to `DeepgramFluxSTTService`,
  `DeepgramSTTService`, and `SpeechmaticsSTTService` to configure whether the
  bot should be interrupted when the external service detects user speech.
  (PR [#3374](https://github.com/pipecat-ai/pipecat/pull/3374))
 - `LLMAssistantAggregator` now exposes the following events:
     - `on_assistant_turn_started`: triggered when the assistant turn starts
     - `on_assistant_turn_stopped`: triggered when the assistant turn ends
     - `on_assistant_thought`: triggered when there's an assistant thought
  available
  (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))
 - Added `KrispVivaTurn` analyzer for end of turn detection using the Krisp VIVA
  SDK (requires `krisp_audio`).
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))
 - Added support for setting up a pipeline task from external files. You can now
  register custom pipeline task setup files by setting the
  `PIPECAT_SETUP_FILES` environment variable. This variable should contain a
  colon-separated list of Python files (e.g. `export
  PIPECAT_SETUP_FILES="setup1.py:setup.py:..."`). Each file must define a
  function with the following signature:
    ```python
    async def setup_pipeline_task(task: PipelineTask):
        ...
    ```
  (PR [#3397](https://github.com/pipecat-ai/pipecat/pull/3397))
 - Added a keepalive task for `InworldTTSService` to keep the service connected
  in the event of no generations for longer periods of time.
  (PR [#3403](https://github.com/pipecat-ai/pipecat/pull/3403))
 - Added `enable_vad` to `Params` for use in the `GladiaSTTService`. When
  enabled, `GladiaSTTService` acts as the turn controller, emitting
  `UserStartedSpeakingFrame`, `UserStoppedSpeakingFrame`, and optionally
  `InterruptionFrame`.
  (PR [#3404](https://github.com/pipecat-ai/pipecat/pull/3404))
 - Added `should_interrupt` property to `GladiaSTTService` to configure whether
  the bot should be interrupted when the external service detects user speech.
  (PR [#3404](https://github.com/pipecat-ai/pipecat/pull/3404))
 - Added `VonageFrameSerializer` for the Vonage Video API Audio Connector
  WebSocket protocol.
  (PR [#3410](https://github.com/pipecat-ai/pipecat/pull/3410))
 - Added `append_trailing_space` parameter to `TTSService` to automatically
  append a trailing space to text before sending to TTS, helping prevent some
  services from vocalizing trailing punctuation.
  (PR [#3424](https://github.com/pipecat-ai/pipecat/pull/3424))
 ### Changed
 - Updated `ElevenLabsRealtimeSTTService` to accept the
  `include_language_detection` parameter to detect language.
    ```python
      stt = ElevenLabsRealtimeSTTService(
          api_key=os.getenv("ELEVENLABS_API_KEY"),
          include_language_detection=True
      )
    ```
  (PR [#3216](https://github.com/pipecat-ai/pipecat/pull/3216))
 - Updated `SpeechmaticsSTTService` to use new Python Voice SDK with improved
  VAD, Smart Turn capabilities, and brings dramatic improvements to latency
  without any impact on accuracy. Use the `turn_detection_mode` parameter to control
  the endpointing of speech, with `TurnDetectionMode.EXTERNAL` (default),
  `TurnDetectionMode.ADAPTIVE`, or `TurnDetectionMode.SMART_TURN`.
    ```python
      stt = SpeechmaticsSTTService(
          api_key=os.getenv("SPEECHMATICS_API_KEY"),
          params=SpeechmaticsSTTService.InputParams(
              language=Language.EN,
 turn_detection_mode=SpeechmaticsSTTService.TurnDetectionMode.ADAPTIVE,
              speaker_active_format="<{speaker_id}>{text}</{speaker_id}>",
          ),
      )
    ```
  (PR [#3225](https://github.com/pipecat-ai/pipecat/pull/3225))
 - `daily-python` updated to 0.23.0.
  (PR [#3257](https://github.com/pipecat-ai/pipecat/pull/3257))
 - `TranscriptionFrame` and `InterimTranscriptionFrame` produced by
  `DailyTransport` now include the transport source (i.e., the originating
  audio track).
  (PR [#3257](https://github.com/pipecat-ai/pipecat/pull/3257))
 - Updates to Inworld TTS services:
    - Improved `InworldTTSService`'s websocket implementation to better flush
      and close context to better handle long inputs.
    - Improved docstrings for `InworldTTSService` and `InworldHttpTTSService`.
  (PR [#3288](https://github.com/pipecat-ai/pipecat/pull/3288))
 - Updated `DeepgramSTTService` to push user started/stopped speaking and
  interruption frames when `vad_enabled` is set to true. This centralizes the
  frames into the service, removing the need to have your application code
  handle Deepgram's events and push these frames.
  (PR [#3314](https://github.com/pipecat-ai/pipecat/pull/3314))
 - Added encoding validation to `DeepgramTTSService` to prevent unsupported
  encodings from reaching the API. The service now raises `ValueError` at
  initialization with a clear error message.
  (PR [#3329](https://github.com/pipecat-ai/pipecat/pull/3329))
 - Updated `read_audio_frame` & `read_video_frame` methods in
  `SmallWebRTCClient` to check if the track is enabled before logging a
  warning.
  (PR [#3336](https://github.com/pipecat-ai/pipecat/pull/3336))
 - Updated `CartesiaTTSService` to support setting `language=None`, resulting in
  Cartesia auto-detecting the language of the conversation.
  (PR [#3366](https://github.com/pipecat-ai/pipecat/pull/3366))
 - The bundled Smart Turn weights are now updated to v3.2, which has better
  handling of short utterances, and is more robust against background noise.
  (PR [#3367](https://github.com/pipecat-ai/pipecat/pull/3367))
 - Updated `SpeechmaticsSTTService` dependency to `speechmatics-voice[smart]>=0.2.6`
  (PR [#3371](https://github.com/pipecat-ai/pipecat/pull/3371))
 - Smart Turn now takes into account `vad_start_seconds` when buffering audio,
  meaning that the start of the turn audio is not cut off. This improves
  accuracy for short utterances.
    - The default value of `pre_speech_ms` is now set to 500ms for Smart Turn.
  (PR [#3377](https://github.com/pipecat-ai/pipecat/pull/3377))
 - Improved Krisp SDK management to allow `KrispVivaTurn` and `KrispVivaFilter`
  to share a single SDK instance within the same process.
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))
 - Updated default model for `GroqTTSService` to `canopylabs/orpheus-v1-english`
  and voice ID to `autumn`.
  (PR [#3399](https://github.com/pipecat-ai/pipecat/pull/3399))
 - Enhanced `FastAPIWebsocketTransport` with optional protocol-level audio
  packetization via the `fixed_audio_packet_size` parameter to support media
  endpoints requiring strict framing and real-time pacing.
  (PR [#3410](https://github.com/pipecat-ai/pipecat/pull/3410))
 - `DeepgramTTSService` and `RimeTTSService` now set `append_trailing_space` to
  `True` to prevent punctuation (e.g., “dot”) from being pronounced.
  (PR [#3424](https://github.com/pipecat-ai/pipecat/pull/3424))
 - Updated `GeminiLiveLLMService` to push `LLMThoughtStartFrame`,
  `LLMThoughtTextFrame`, and `LLMThoughtEndFrame` when the model returns
  thought content.
  (PR [#3431](https://github.com/pipecat-ai/pipecat/pull/3431))
 ### Deprecated
 - `pipecat.audio.interruptions.MinWordsInterruptionStrategy` is deprecated. Use
  `pipecat.turns.user_start.MinWordsUserTurnStartStrategy` with
  `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - `FrameProcessor.interruption_strategies` is deprecated, use
  `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - The `LLMUserAggregatorParams` and `LLMAssistantAggregatorParams` classes in
  `pipecat.processors.aggregators.llm_response` are now deprecated. Use the new
  universal `LLMContext` and `LLMContextAggregatorPair` instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - Deprecated the `emulated` field in the `UserStartedSpeakingFrame` and
  `UserStoppedSpeakingFrame` frames.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - `EmulateUserStartedSpeakingFrame` and `EmulateUserStoppedSpeakingFrame`
  frames are deprecated.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - ⚠️ `TransportParams.turn_analyzer` is deprecated and might result in
  unexpected behavior, use `LLMUserAggregator`'s new `user_turn_strategies`
  parameter instead.
  (PR [#3045](https://github.com/pipecat-ai/pipecat/pull/3045))
 - For `SpeechmaticsSTTService`, the `end_of_utterance_mode` parameter is
  deprecated. Use the new `turn_detection_mode` parameter instead, with
  `TurnDetectionMode.EXTERNAL`,`TurnDetectionMode.ADAPTIVE`, or
  `TurnDetectionMode.SMART_TURN`. The `enable_vad` parameter is also
  deprecated and is inferred from the `turn_detection_mode`.
  (PR [#3225](https://github.com/pipecat-ai/pipecat/pull/3225))
 - `OpenAILLMContext` and its associated things (context aggregators, etc.) are
  now deprecated in favor of the universal `LLMContext` and its associated
  things.
  From the developer's point of view, switching to using `LLMContext`
  machinery will usually be a matter of going from this:
    ```python
    context = OpenAILLMContext(messages, tools)
    context_aggregator = llm.create_context_aggregator(context)
    ```
    To this:
    ```
    context = LLMContext(messages, tools)
    context_aggregator = LLMContextAggregatorPair(context)
    ```
  (PR [#3263](https://github.com/pipecat-ai/pipecat/pull/3263))
 - `STTMuteFilter` is deprecated and will be removed in a future version. Use
  `LLMUserAggregator`'s new `user_mute_strategies` instead.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))
 - `FrameProcessor.interruptions_allowed` is now deprecated, use
  `LLMUserAggregator`'s new parameter `user_mute_strategies` instead.
  (PR [#3297](https://github.com/pipecat-ai/pipecat/pull/3297))
 - `PipelineParams.allow_interruptions` is now deprecated, use
  `LLMUserAggregator`'s new parameter `user_turn_strategies` instead. For
  example, to disable interruptions but still get user turns you can do:
    ```python
    context_aggregator = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(
            user_turn_strategies=UserTurnStrategies(
 start=[TranscriptionUserTurnStartStrategy(enable_interruptions=False)],
            ),
        ),
    )
    ```
  (PR [#3297](https://github.com/pipecat-ai/pipecat/pull/3297))
 - `TranscriptProcessor` and related data classes and frames
  (`TranscriptionMessage`, `ThoughtTranscriptionMessage`,
  `TranscriptionUpdateFrame`) are deprecated. Use `LLMUserAggregator`'s and
  `LLMAssistantAggregator`'s new events (`on_user_turn_stopped` and
  `on_assistant_turn_stopped`) instead.
  (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))
 - Deprecated support for the `vad_events` `LiveOptions` in
  `DeepgramSTTService`. Instead, use a local Silero VAD for VAD events.
  Additionally, deprecated `should_interrupt` which will be removed along with
  `vad_events` support in a future release.
  (PR [#3386](https://github.com/pipecat-ai/pipecat/pull/3386))
 - Loading external observers from files is deprecated, use the new pipeline
  task setup files and `PIPECAT_SETUP_FILES` environment variable instead.
  (PR [#3397](https://github.com/pipecat-ai/pipecat/pull/3397))
 ### Fixed
 - Improved error handling in `ElevenLabsRealtimeSTTService`
  - Fixed an issue in `ElevenLabsRealtimeSTTService` causing an infinite loop
  that blocks the process if the websocket disconnects due to an error
  (PR [#3233](https://github.com/pipecat-ai/pipecat/pull/3233))
 - Fixed a bug in `STTMuteFilter` where the user was not always muted during
  function calls, especially when there were multiple simultaneous calls.
  (PR [#3292](https://github.com/pipecat-ai/pipecat/pull/3292))
 - Fixed a `RNNoiseFilter` issue that would cause a "[Errno 12] Cannot allocate
  memory" error when processing silence audio frames.
  (PR [#3322](https://github.com/pipecat-ai/pipecat/pull/3322))
 - Updated `SpeechmaticsSTTService` for version `0.0.99+`:
    - Fixed `SpeechmaticsSTTService` to listen for `VADUserStoppedSpeakingFrame`
      in order to finalize transcription.
    - Default to `TurnDetectionMode.FIXED` for Pipecat-controlled end of turn
      detection.
    - Only emit VAD + interruption frames if VAD is enabled within the plugin
      (modes other than `TurnDetectionMode.FIXED` or `TurnDetectionMode.EXTERNAL`).
  (PR [#3328](https://github.com/pipecat-ai/pipecat/pull/3328))
 - Fixed an issue with function calling where a handler failing to invoke its
  result callback could leave the context stuck in IN_PROGRESS, causing LLM
  inference for subsequent function call results to block while waiting on the
  unresolved call.
  (PR [#3343](https://github.com/pipecat-ai/pipecat/pull/3343))
 - Fixed an issue with DeepgramTTSService where the model would output "Dot"
  instead of a period in some circumstances.
  (PR [#3345](https://github.com/pipecat-ai/pipecat/pull/3345))
 - Fixed an issue in `traced_stt` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3351](https://github.com/pipecat-ai/pipecat/pull/3351))
 - Fixed an issue in GeminiLiveLLMService where TranscriptionFrames were
  occasionally not pushed.
  (PR [#3356](https://github.com/pipecat-ai/pipecat/pull/3356))
 - Fixed potential memory leaks and initialization issues in `KrispVivaFilter`
  by improving SDK lifecycle management.
  (PR [#3391](https://github.com/pipecat-ai/pipecat/pull/3391))
 - Fixed timing issue in `BaseOutputTransport` where the bot speaking flag was
  set after awaiting, allowing the event loop to re-enter the method before the
  guard was set.
  (PR [#3400](https://github.com/pipecat-ai/pipecat/pull/3400))
 - Fixed an issue in `traced_llm` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3422](https://github.com/pipecat-ai/pipecat/pull/3422))
 - Fixed an issue in `traced_tts`, `traced_gemini_live`, and
  `traced_openai_realtime` where `model_name` in OpenTelemetry appears as
  `unknown`.
  (PR [#3428](https://github.com/pipecat-ai/pipecat/pull/3428))
 - Fixed `request_image_frame` (for backwards compatibility) and restored
  function-call–related fields in `UserImageRequestFrame` and
  `UserImageRawFrame`, preventing a case where adding a non-LLM message to the
  context could trigger duplicate LLM inferences (on image arrival and on
  function-call result), potentially causing an infinite inference loop.
  (PR [#3430](https://github.com/pipecat-ai/pipecat/pull/3430))
 - Fixed `LLMContext.create_audio_message()` by correcting an internal helper
  that was incorrectly declared async while being run in `asyncio.to_thread()`.
  (PR [#3435](https://github.com/pipecat-ai/pipecat/pull/3435))
 ### Other
 - Added `52-live-transcription.py` foundational example demonstrating live
  transcription and translation from English to Spanish. In this example, the
  bot is not interruptible: as the user continues speaking, English
  transcriptions are queued, and the bot continuously translates and speaks
  each queued sentence in Spanish without being interrupted by new user speech.
  (PR [#3316](https://github.com/pipecat-ai/pipecat/pull/3316))
 - Added a new foundational example `53-concurrent-llm-evaluation.py` that shows
  how to use `UserTurnProcessor`.
  (PR [#3372](https://github.com/pipecat-ai/pipecat/pull/3372))
 - Added a new foundational example `28-user-assistant-turns.py` that shows how
  to use the new `LLMUserAggregator` and `LLMAssistantAggregator` events to
  gather a conversation transcript.
  (PR [#3385](https://github.com/pipecat-ai/pipecat/pull/3385))
 ## [0.0.98] - 2025-12-17
 ### Added
--- a/changelog/3045.added.md
+++ b/changelog/3045.added.md
@@ -1,42 +0,0 @@
 - Introducing user turn strategies. User turn strategies indicate when the user turn starts or stops. In conversational agents, these are often referred to as start/stop speaking or turn-taking plans or policies.
  User turn start strategies indicate when the user starts speaking (e.g. using VAD events or when a user says one or more words).
  User turn stop strategies indicate when the user stops speaking (e.g. using an end-of-turn detection model or by observing incoming transcriptions).
  A list of strategies can be specified for both strategies; strategies are evaluated in order until one evaluates to true.
  Available user turn start strategies:
    - VADUserTurnStartStrategy
    - TranscriptionUserTurnStartStrategy
    - MinWordsUserTurnStartStrategy
    - ExternalUserTurnStartStrategy
  Available user turn stop strategies:
    - TranscriptionUserTurnStopStrategy
    - TurnAnalyzerUserTurnStopStrategy
    - ExternalUserTurnStopStrategy
  The default strategies are:
    - start: [VADUserTurnStartStrategy, TranscriptionUserTurnStartStrategy]
    - stop: [TranscriptionUserTurnStopStrategy]
  Turn strategies are configured when setting up `LLMContextAggregatorPair`. For example:
  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_turn_strategies=UserTurnStrategies(
              stop=[
                  TurnAnalyzerUserTurnStopStrategy(
                      turn_analyzer=LocalSmartTurnAnalyzerV3(params=SmartTurnParams())
                  )
              ],
          )
      ),
  )
  ```
  In order to use the user turn strategies you must update to the new universal `LLMContext` and `LLMContextAggregatorPair`.
--- a/changelog/3045.deprecated.2.md
+++ b/changelog/3045.deprecated.2.md
@@ -1 +0,0 @@
 - ⚠️ `TransportParams.turn_analyzer` is deprecated and might result in unexpected behavior, use `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
--- a/changelog/3045.deprecated.3.md
+++ b/changelog/3045.deprecated.3.md
@@ -1 +0,0 @@
 - `FrameProcessor.interruption_strategies` is deprecated, use `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
--- a/changelog/3045.deprecated.4.md
+++ b/changelog/3045.deprecated.4.md
@@ -1 +0,0 @@
 - `EmulateUserStartedSpeakingFrame` and `EmulateUserStoppedSpeakingFrame` frames are deprecated.
--- a/changelog/3045.deprecated.5.md
+++ b/changelog/3045.deprecated.5.md
@@ -1 +0,0 @@
 - Deprecated the `emulated` field in the `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` frames.
--- a/changelog/3045.deprecated.6.md
+++ b/changelog/3045.deprecated.6.md
@@ -1 +0,0 @@
 - The `LLMUserAggregatorParams` and `LLMAssistantAggregatorParams` classes in `pipecat.processors.aggregators.llm_response` are now deprecated. Use the new universal `LLMContext` and `LLMContextAggregatorPair` instead.
--- a/changelog/3045.deprecated.md
+++ b/changelog/3045.deprecated.md
@@ -1 +0,0 @@
 - `pipecat.audio.interruptions.MinWordsInterruptionStrategy` is deprecated. Use `pipecat.turns.user_start.MinWordsUserTurnStartStrategy` with `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
--- a/changelog/3205.added.md
+++ b/changelog/3205.added.md
@@ -1 +0,0 @@
 - Added `RNNoiseFilter` for real-time noise suppression using RNNoise neural network via pyrnnoise library.
--- a/changelog/3216.changed.md
+++ b/changelog/3216.changed.md
@@ -1,7 +0,0 @@
 - Updated `ElevenLabsRealtimeSTTService` to accept the `include_language_detection` parameter to detect language.
  ```python
    stt = ElevenLabsRealtimeSTTService(
        api_key=os.getenv("ELEVENLABS_API_KEY"),
        include_language_detection=True
    )
  ```
--- a/changelog/3225.changed.md
+++ b/changelog/3225.changed.md
@@ -1,15 +0,0 @@
 - Updated `SpeechmaticsSTTService` to use new Python Voice SDK with improved VAD,
  Smart Turn capabilities, and brings dramatic improvements to latency without
  any impact on accuracy. Use the `turn_detection_mode` parameter to control the
  endpointing of speech, with `TurnDetectionMode.EXTERNAL` (default),
  `TurnDetectionMode.ADAPTIVE`, or `TurnDetectionMode.SMART_TURN`.
  ```python
    stt = SpeechmaticsSTTService(
        api_key=os.getenv("SPEECHMATICS_API_KEY"),
        params=SpeechmaticsSTTService.InputParams(
            language=Language.EN,
            turn_detection_mode=SpeechmaticsSTTService.TurnDetectionMode.ADAPTIVE,
            speaker_active_format="<{speaker_id}>{text}</{speaker_id}>",
        ),
    )
  ```
--- a/changelog/3225.deprecated.md
+++ b/changelog/3225.deprecated.md
@@ -1,4 +0,0 @@
 - For `SpeechmaticsSTTService`, the `end_of_utterance_mode` parameter is deprecated.
  Use the new `turn_detection_mode` parameter instead, with `TurnDetectionMode.EXTERNAL`,
  `TurnDetectionMode.ADAPTIVE`, or `TurnDetectionMode.SMART_TURN`. The `enable_vad`
  parameter is also deprecated and is inferred from the `turn_detection_mode`.
--- a/changelog/3233.fixed.md
+++ b/changelog/3233.fixed.md
@@ -1,2 +0,0 @@
 - Improved error handling in `ElevenLabsRealtimeSTTService`
 - Fixed an issue in `ElevenLabsRealtimeSTTService` causing an infinite loop that blocks the process if the websocket disconnects due to an error
--- a/changelog/3257.changed.2.md
+++ b/changelog/3257.changed.2.md
@@ -1 +0,0 @@
 - `TranscriptionFrame` and `InterimTranscriptionFrame` produced by `DailyTransport` now include the transport source (i.e., the originating audio track).
--- a/changelog/3257.changed.md
+++ b/changelog/3257.changed.md
@@ -1 +0,0 @@
 - `daily-python` updated to 0.23.0.
--- a/changelog/3263.deprecated.md
+++ b/changelog/3263.deprecated.md
@@ -1,15 +0,0 @@
 - `OpenAILLMContext` and its associated things (context aggregators, etc.) are now deprecated in favor of the universal `LLMContext` and its associated things.
  From the developer's point of view, switching to using `LLMContext` machinery will usually be a matter of going from this:
  ```python
  context = OpenAILLMContext(messages, tools)
  context_aggregator = llm.create_context_aggregator(context)
  ```
  To this:
  ```
  context = LLMContext(messages, tools)
  context_aggregator = LLMContextAggregatorPair(context)
  ```
--- a/changelog/3267.added.md
+++ b/changelog/3267.added.md
@@ -1,8 +0,0 @@
 - Added `GrokRealtimeLLMService` for xAI's Grok Voice Agent API with real-time voice conversations:
  - Support for real-time audio streaming with WebSocket connection
  - Built-in server-side VAD (Voice Activity Detection)
  - Multiple voice options: Ara, Rex, Sal, Eve, Leo
  - Built-in tools support: web_search, x_search, file_search
  - Custom function calling with standard Pipecat tools schema
  - Configurable audio formats (PCM at 8kHz-48kHz)
--- a/changelog/3268.added.md
+++ b/changelog/3268.added.md
@@ -1 +0,0 @@
 - Added an approximation of TTFB for Ultravox.
--- a/changelog/3288.changed.md
+++ b/changelog/3288.changed.md
@@ -1,5 +0,0 @@
 - Updates to Inworld TTS services:
  - Improved `InworldTTSService`'s websocket implementation to better flush and
    close context to better handle long inputs.
  - Improved docstrings for `InworldTTSService` and `InworldHttpTTSService`.
--- a/changelog/3289.added.md
+++ b/changelog/3289.added.md
@@ -1 +0,0 @@
 - Added a new `AudioContextTTSService` to the TTS service base classes. The `AudioContextWordTTSService` now inherits from `AudioContextTTSService` and `WebsocketWordTTSService`.
--- a/changelog/3291.added.md
+++ b/changelog/3291.added.md
@@ -1,4 +0,0 @@
 - `LLMUserAggregator` now exposes the following events:
   - `on_user_turn_started`: triggered when a user turn starts
   - `on_user_turn_stopped`: triggered when a user turn ends
   - `on_user_turn_stop_timeout`: triggered when a user turn does not stop and times out
--- a/changelog/3292.added.md
+++ b/changelog/3292.added.md
@@ -1,29 +0,0 @@
 - Introducing user mute strategies. User mute strategies indicate when user input should be muted based on the current system state.
  In conversational agents, user mute strategies are used to prevent user input from interrupting bot speech, tool execution, or other critical system operations.
  A list of strategies can be specified; all strategies are evaluated for every frame so that each strategy can maintain its internal state. A user frame is muted if any of the configured strategies indicates it should be muted.
  Available user mute strategies:
  * `FirstSpeechUserMuteStrategy`
  * `MuteUntilFirstBotCompleteUserMuteStrategy`
  * `AlwaysUserMuteStrategy`
  * `FunctionCallUserMuteStrategy`
  User mute strategies replace the legacy `STTMuteFilter` and provide a more flexible and composable approach to muting user input.
  User mute strategies are configured when setting up the `LLMContextAggregatorPair`. For example:
  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_mute_strategies=[
              FirstSpeechUserMuteStrategy(),
          ]
      ),
  )
  ```
  In order to use user mute strategies you should update to the new universal `LLMContext` and `LLMContextAggregatorPair`.
--- a/changelog/3292.deprecated.md
+++ b/changelog/3292.deprecated.md
@@ -1 +0,0 @@
 - `STTMuteFilter` is deprecated and will be removed in a future version. Use `LLMUserAggregator`'s new `user_mute_strategies` instead.
--- a/changelog/3292.fixed.md
+++ b/changelog/3292.fixed.md
@@ -1 +0,0 @@
 - Fixed a bug in `STTMuteFilter` where the user was not always muted during function calls, especially when there were multiple simultaneous calls.
--- a/changelog/3297.deprecated.2.md
+++ b/changelog/3297.deprecated.2.md
@@ -1 +0,0 @@
 - `FrameProcessor.interruptions_allowed` is now deprecated, use `LLMUserAggregator`'s new parameter `user_mute_strategies` instead.
--- a/changelog/3297.deprecated.md
+++ b/changelog/3297.deprecated.md
@@ -1,12 +0,0 @@
 - `PipelineParams.allow_interruptions` is now deprecated, use `LLMUserAggregator`'s new parameter `user_turn_strategies` instead. For example, to disable interruptions but still get user turns you can do:
  ```python
  context_aggregator = LLMContextAggregatorPair(
      context,
      user_params=LLMUserAggregatorParams(
          user_turn_strategies=UserTurnStrategies(
              start=[TranscriptionUserTurnStartStrategy(enable_interruptions=False)],
          ),
      ),
  )
  ```
--- a/changelog/3300.added.md
+++ b/changelog/3300.added.md
@@ -1 +0,0 @@
 - Added `use_ssl` parameter to `NvidiaSTTService`, `NvidiaSegmentedSTTService` and `NvidiaTTSService`.
--- a/changelog/3314.changed.md
+++ b/changelog/3314.changed.md
@@ -1 +0,0 @@
 - Updated `DeepgramSTTService` to push user started/stopped speaking and interruption frames when `vad_enabled` is set to true. This centralizes the frames into the service, removing the need to have your application code handle Deepgram's events and push these frames.
--- a/changelog/3316.added.md
+++ b/changelog/3316.added.md
@@ -1 +0,0 @@
 - Added `enable_interruptions` constructor argument to all user turn strategies. This tells the `LLMUserAggregator` to push or not push an `InterruptionFrame`.
--- a/changelog/3316.other.md
+++ b/changelog/3316.other.md
@@ -1 +0,0 @@
 - Added `52-live-transcription.py` foundational example demonstrating live transcription and translation from English to Spanish. In this example, the bot is not interruptible: as the user continues speaking, English transcriptions are queued, and the bot continuously translates and speaks each queued sentence in Spanish without being interrupted by new user speech.
--- a/changelog/3322.fixed.md
+++ b/changelog/3322.fixed.md
@@ -1 +0,0 @@
 - Fixed a `RNNoiseFilter` issue that would cause a "[Errno 12] Cannot allocate memory" error when processing silence audio frames.
--- a/changelog/3328.added.md
+++ b/changelog/3328.added.md
@@ -1 +0,0 @@
 - Added `split_sentences` parameter to `SpeechmaticsSTTService` to control sentence splitting behavior for finals on sentence boundaries.
--- a/changelog/3328.fixed.md
+++ b/changelog/3328.fixed.md
@@ -1,4 +0,0 @@
 - Updated `SpeechmaticsSTTService` for version `0.0.99+`:
  - Fixed `SpeechmaticsSTTService` to listen for `VADUserStoppedSpeakingFrame` in order to finalize transcription.
  - Default to `TurnDetectionMode.FIXED` for Pipecat-controlled end of turn detection.
  - Only emit VAD + interruption frames if VAD is enabled within the plugin (modes other than `TurnDetectionMode.FIXED` or `TurnDetectionMode.EXTERNAL`).
--- a/changelog/3329.changed.md
+++ b/changelog/3329.changed.md
@@ -1 +0,0 @@
 - Added encoding validation to `DeepgramTTSService` to prevent unsupported encodings from reaching the API. The service now raises `ValueError` at initialization with a clear error message.
--- a/changelog/3334.added.md
+++ b/changelog/3334.added.md
@@ -1,2 +0,0 @@
 - Added word-level timestamp support to `AzureTTSService` for accurate text-to-audio synchronization.
--- a/changelog/3336.changed.md
+++ b/changelog/3336.changed.md
@@ -1 +0,0 @@
 - Updated `read_audio_frame` & `read_video_frame` methods in `SmallWebRTCClient` to check if the track is enabled before logging a warning.
--- a/changelog/3343.fixed.md
+++ b/changelog/3343.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue with function calling where a handler failing to invoke its result callback could leave the context stuck in IN_PROGRESS, causing LLM inference for subsequent function call results to block while waiting on the unresolved call.
--- a/changelog/3345.fixed.md
+++ b/changelog/3345.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue with DeepgramTTSService where the model would output "Dot" instead of a period in some circumstances.
--- a/changelog/3346.added.md
+++ b/changelog/3346.added.md
@@ -1 +0,0 @@
 - Added `pronunciation_dict_id` parameter to `CartesiaTTSService.InputParams` and `CartesiaHttpTTSService.InputParams` to support Cartesia's pronunciation dictionary feature for custom pronunciations.
--- a/changelog/3351.fixed.md
+++ b/changelog/3351.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue in `traced_stt` where `model_name` in OpenTelemetry appears as `unknown`.
--- a/changelog/3356.fixed.md
+++ b/changelog/3356.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue in GeminiLiveLLMService where TranscriptionFrames were occasionally not pushed.
--- a/changelog/3357.added.md
+++ b/changelog/3357.added.md
@@ -1 +0,0 @@
 - Added support for using the HeyGen LiveAvatar API with the `HeyGenTransport` (see https://www.liveavatar.com/).
--- a/changelog/3360.added.md
+++ b/changelog/3360.added.md
@@ -1,8 +0,0 @@
 - Added image support to `OpenAIRealtimeLLMService` via `InputImageRawFrame`:
  - New `start_video_paused` parameter to control initial video input state
  - New `video_frame_detail` parameter to set image processing quality ("auto",
    "low", or "high"). This corresponds to OpenAI Realtime's `image_detail`
    parameter.
  - `set_video_input_paused()` method to pause/resume video input at runtime
  - `set_video_frame_detail()` method to adjust video frame quality dynamically
  - Automatic rate limiting (1 frame per second) to prevent API overload
--- a/changelog/3366.changed.md
+++ b/changelog/3366.changed.md
@@ -1 +0,0 @@
 - Updated `CartesiaTTSService` to support setting `language=None`, resulting in Cartesia auto-detecting the language of the conversation.
--- a/changelog/3367.changed.md
+++ b/changelog/3367.changed.md
@@ -1,3 +0,0 @@
 - The bundled Smart Turn weights are now updated to v3.2, which has better
  handling of short utterances, and is more robust against background
  noise.
--- a/changelog/3371.changed.md
+++ b/changelog/3371.changed.md
@@ -1 +0,0 @@
 - Updated `SpeechmaticsSTTService` dependency to `speechmatics-voice[smart]>=0.2.6`
--- a/changelog/3372.added.2.md
+++ b/changelog/3372.added.2.md
@@ -1 +0,0 @@
 - Added `UserTurnProcessor`, a frame processor built on `UserTurnController` that pushes `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` frames and interruptions based on the controller's user turn strategies.
--- a/changelog/3372.added.md
+++ b/changelog/3372.added.md
@@ -1 +0,0 @@
 - Added `UserTurnController` to manage user turns. It emits `on_user_turn_started`, `on_user_turn_stopped`, and `on_user_turn_stop_timeout` events, and can be integrated into processors to detect and handle user turns. `LLMUserAggregator` and `UserTurnProcessor` are implemented using this controller.
--- a/changelog/3372.other.md
+++ b/changelog/3372.other.md
@@ -1 +0,0 @@
 - Added a new foundational example `53-concurrent-llm-evaluation.py` that shows how to use `UserTurnProcessor`.
--- a/changelog/3374.added.md
+++ b/changelog/3374.added.md
@@ -1 +0,0 @@
 - Added `should_interrupt` property to `DeepgramFluxSTTService`, `DeepgramSTTService`, and `SpeechmaticsSTTService` to configure whether the bot should be interrupted when the external service detects user speech.
--- a/changelog/3377.changed.md
+++ b/changelog/3377.changed.md
@@ -1,5 +0,0 @@
 - Smart Turn now takes into account `vad_start_seconds` when buffering audio,
  meaning that the start of the turn audio is not cut off. This improves
  accuracy for short utterances.
 - The default value of `pre_speech_ms` is now set to 500ms for Smart Turn.
--- a/changelog/3385.added.md
+++ b/changelog/3385.added.md
@@ -1,4 +0,0 @@
 - `LLMAssistantAggregator` now exposes the following events:
   - `on_assistant_turn_started`: triggered when the assistant turn starts
   - `on_assistant_turn_stopped`: triggered when the assistant turn ends
   - `on_assistant_thought`: triggered when there's an assistant thought available
--- a/changelog/3385.deprecated.md
+++ b/changelog/3385.deprecated.md
@@ -1 +0,0 @@
 - `TranscriptProcessor` and related data classes and frames (`TranscriptionMessage`, `ThoughtTranscriptionMessage`, `TranscriptionUpdateFrame`) are deprecated. Use `LLMUserAggregator`'s and `LLMAssistantAggregator`'s new events (`on_user_turn_stopped` and `on_assistant_turn_stopped`) instead.
--- a/changelog/3385.other.md
+++ b/changelog/3385.other.md
@@ -1 +0,0 @@
 - Added a new foundational example `28-user-assistant-turns.py` that shows how to use the new `LLMUserAggregator` and `LLMAssistantAggregator` events to gather a conversation transcript.
--- a/changelog/3386.deprecated.md
+++ b/changelog/3386.deprecated.md
@@ -1 +0,0 @@
 - Deprecated support for the `vad_events` `LiveOptions` in `DeepgramSTTService`. Instead, use a local Silero VAD for VAD events. Additionally, deprecated `should_interrupt` which will be removed along with `vad_events` support in a future release.
--- a/changelog/3391.added.md
+++ b/changelog/3391.added.md
@@ -1 +0,0 @@
 - Added `KrispVivaTurn` analyzer for end of turn detection using the Krisp VIVA SDK (requires `krisp_audio`).
--- a/changelog/3391.changed.md
+++ b/changelog/3391.changed.md
@@ -1 +0,0 @@
 - Improved Krisp SDK management to allow `KrispVivaTurn` and `KrispVivaFilter` to share a single SDK instance within the same process.
--- a/changelog/3391.fixed.md
+++ b/changelog/3391.fixed.md
@@ -1 +0,0 @@
 - Fixed potential memory leaks and initialization issues in `KrispVivaFilter` by improving SDK lifecycle management.
--- a/changelog/3397.added.md
+++ b/changelog/3397.added.md
@@ -1,6 +0,0 @@
 - Added support for setting up a pipeline task from external files. You can now register custom pipeline task setup files by setting the `PIPECAT_SETUP_FILES` environment variable. This variable should contain a colon-separated list of Python files (e.g. `export PIPECAT_SETUP_FILES="setup1.py:setup.py:..."`). Each file must define a function with the following signature:
  ```python
  async def setup_pipeline_task(task: PipelineTask):
      ...
  ```
--- a/changelog/3397.deprecated.md
+++ b/changelog/3397.deprecated.md
@@ -1 +0,0 @@
 - Loading external observers from files is deprecated, use the new pipeline task setup files and `PIPECAT_SETUP_FILES` environment variable instead.
--- a/changelog/3399.changed.md
+++ b/changelog/3399.changed.md
@@ -1 +0,0 @@
 - Updated default model for `GroqTTSService` to `canopylabs/orpheus-v1-english` and voice ID to `autumn`.
--- a/changelog/3400.fixed.md
+++ b/changelog/3400.fixed.md
@@ -1 +0,0 @@
 - Fixed timing issue in `BaseOutputTransport` where the bot speaking flag was set after awaiting, allowing the event loop to re-enter the method before the guard was set.
--- a/changelog/3403.added.md
+++ b/changelog/3403.added.md
@@ -1 +0,0 @@
 - Added a keepalive task for `InworldTTSService` to keep the service connected in the event of no generations for longer periods of time.
--- a/changelog/3404.added.2.md
+++ b/changelog/3404.added.2.md
@@ -1 +0,0 @@
 - Added `enable_vad` to `Params` for use in the `GladiaSTTService`. When enabled, `GladiaSTTService` acts as the turn controller, emitting `UserStartedSpeakingFrame`, `UserStoppedSpeakingFrame`, and optionally `InterruptionFrame`.
--- a/changelog/3404.added.md
+++ b/changelog/3404.added.md
@@ -1 +0,0 @@
 - Added `should_interrupt` property to `GladiaSTTService` to configure whether the bot should be interrupted when the external service detects user speech.
--- a/changelog/3410.added.md
+++ b/changelog/3410.added.md
@@ -1 +0,0 @@
 - Added `VonageFrameSerializer` for the Vonage Video API Audio Connector WebSocket protocol.
--- a/changelog/3410.changed.md
+++ b/changelog/3410.changed.md
@@ -1 +0,0 @@
 - Enhanced `FastAPIWebsocketTransport` with optional protocol-level audio packetization via the `fixed_audio_packet_size` parameter to support media endpoints requiring strict framing and real-time pacing.
--- a/changelog/3422.fixed.md
+++ b/changelog/3422.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue in `traced_llm` where `model_name` in OpenTelemetry appears as `unknown`.
--- a/changelog/3424.added.md
+++ b/changelog/3424.added.md
@@ -1 +0,0 @@
 - Added `append_trailing_space` parameter to `TTSService` to automatically append a trailing space to text before sending to TTS, helping prevent some services from vocalizing trailing punctuation.
--- a/changelog/3424.changed.md
+++ b/changelog/3424.changed.md
@@ -1 +0,0 @@
 - `DeepgramTTSService` and `RimeTTSService` now set `append_trailing_space` to `True` to prevent punctuation (e.g., “dot”) from being pronounced.
--- a/changelog/3428.fixed.md
+++ b/changelog/3428.fixed.md
@@ -1 +0,0 @@
 - Fixed an issue in `traced_tts`, `traced_gemini_live`, and `traced_openai_realtime` where `model_name` in OpenTelemetry appears as `unknown`.
--- a/changelog/3430.fixed.md
+++ b/changelog/3430.fixed.md
@@ -1 +0,0 @@
 - Fixed `request_image_frame` (for backwards compatibility) and restored function-call–related fields in `UserImageRequestFrame` and `UserImageRawFrame`, preventing a case where adding a non-LLM message to the context could trigger duplicate LLM inferences (on image arrival and on function-call result), potentially causing an infinite inference loop.
--- a/changelog/3431.changed.md
+++ b/changelog/3431.changed.md
@@ -1 +0,0 @@
 - Updated `GeminiLiveLLMService` to push `LLMThoughtStartFrame`, `LLMThoughtTextFrame`, and `LLMThoughtEndFrame` when the model returns thought content.
--- a/changelog/3435.fixed.md
+++ b/changelog/3435.fixed.md
@@ -1 +0,0 @@
 - Fixed `LLMContext.create_audio_message()` by correcting an internal helper that was incorrectly declared async while being run in `asyncio.to_thread()`.
		`@@ -1 +0,0 @@`
			- ⚠️ `TransportParams.turn_analyzer` is deprecated and might result in unexpected behavior, use `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
		`@@ -1 +0,0 @@`
			- `FrameProcessor.interruption_strategies` is deprecated, use `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
		`@@ -1 +0,0 @@`
			- `EmulateUserStartedSpeakingFrame` and `EmulateUserStoppedSpeakingFrame` frames are deprecated.
		`@@ -1 +0,0 @@`
			- Deprecated the `emulated` field in the `UserStartedSpeakingFrame` and `UserStoppedSpeakingFrame` frames.
		`@@ -1 +0,0 @@`
			- The `LLMUserAggregatorParams` and `LLMAssistantAggregatorParams` classes in `pipecat.processors.aggregators.llm_response` are now deprecated. Use the new universal `LLMContext` and `LLMContextAggregatorPair` instead.
		`@@ -1 +0,0 @@`
			- `pipecat.audio.interruptions.MinWordsInterruptionStrategy` is deprecated. Use `pipecat.turns.user_start.MinWordsUserTurnStartStrategy` with `LLMUserAggregator`'s new `user_turn_strategies` parameter instead.
		`@@ -1 +0,0 @@`
			- Added `RNNoiseFilter` for real-time noise suppression using RNNoise neural network via pyrnnoise library.
		`@@ -1,2 +0,0 @@`
			- Improved error handling in `ElevenLabsRealtimeSTTService`
			- Fixed an issue in `ElevenLabsRealtimeSTTService` causing an infinite loop that blocks the process if the websocket disconnects due to an error
		`@@ -1 +0,0 @@`
			- `TranscriptionFrame` and `InterimTranscriptionFrame` produced by `DailyTransport` now include the transport source (i.e., the originating audio track).
		`@@ -1 +0,0 @@`
			`- Added an approximation of TTFB for Ultravox.`