Pipecat-ai

Added

- Added `DailyTransport` event `on_participant_left`.

- Added support for receiving `DailyTransportMessage`.

Fixed

- Images are now resized to the size of the output camera. This was causing
images not being displayed.

- Fixed an issue in `DailyTransport` that would not allow the input processor to
shutdown if no participant ever joined the room.

- Fixed base transports start and stop. In some situation processors would halt
or not shutdown properly.

Changed

- `MoondreamService` argument `model_id` is now `model`.

- `VADAnalyzer` arguments have been renamed for more clarity.

Fixed

- Fixed an issue with `DailyInputTransport` and `DailyOutputTransport` that
could cause some threads to not start properly.

- Fixed `STTService`. Add `max_silence_secs` and `max_buffer_secs` to handle
better what's being passed to the STT service. Also add exponential smoothing
to the RMS.

- Fixed `WhisperSTTService`. Add `no_speech_prob` to avoid garbage output text.

Added

- Added `DailyTranscriptionSettings` to be able to specify transcription
settings much easier (e.g. language).

Other

- Updated `simple-chatbot` with Spanish.

- Add missing dependencies in some of the examples.

Added

- Allow stopping pipeline tasks with new `StopTaskFrame`.

Changed

- TTS, STT and image generation service now use `AsyncGenerator`.

Fixed

- `DailyTransport`: allow registering for participant transcriptions even if
input transport is not initialized yet.

Other

- Updated `storytelling-chatbot`.

Added

- Added Intel GPU support to `MoondreamService`.

- Added support for sending transport messages (e.g. to communicate with an app
at the other end of the transport).

- Added `FrameProcessor.push_error()` to easily send an `ErrorFrame` upstream.

Fixed

- Fixed Azure services (TTS and image generation).

Other

- Updated `simple-chatbot`, `moondream-chatbot` and `translation-chatbot`
examples.

Changed

Many things have changed in this version. Many of the main ideas such as frames,
processors, services and transports are still there but some things have changed
a bit.

- `Frame`s describe the basic units for processing. For example, text, image or
audio frames. Or control frames to indicate a user has started or stopped
speaking.

- `FrameProcessor`s process frames (e.g. they convert a `TextFrame` to an
`ImageRawFrame`) and push new frames downstream or upstream to their linked
peers.

- `FrameProcessor`s can be linked together. The easiest wait is to use the
`Pipeline` which is a container for processors. Linking processors allow
frames to travel upstream or downstream easily.

- `Transport`s are a way to send or receive frames. There can be local
transports (e.g. local audio or native apps), network transports
(e.g. websocket) or service transports (e.g. https://daily.co).

- `Pipeline`s are just a processor container for other processors.

- A `PipelineTask` know how to run a pipeline.

- A `PipelineRunner` can run one or more tasks and it is also used, for example,
to capture Ctrl-C from the user.