* Handle sentence triggers and registered intents in Assist LLM API
* Remove from LLM
* Check sentence triggers and local intents first
* Fix type
* Fix type again
* Use pipeline language
* Fix cloud test
* Clean up and fix translation key
* Refactor async_recognize
* Allow tts options of type list such as profiles in google_cloud
* Update tests/components/tts/test_media_source.py
* Don't mix engine specific options with other options
* Fix test
* Update assist_pipeline snapshots
---------
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
Co-authored-by: Paulus Schoutsen <paulus@home-assistant.io>
Co-authored-by: Erik Montnemery <erik@montnemery.com>
* Allow targetting conversation agent as pipeline
* Test that we can use a conversation entity as an assist pipeline
* Add test for WS get
---------
Co-authored-by: Michael Hansen <mike@rhasspy.org>
* Default agent as entity
* Migrate constant to point at new value
* Fix tests
* Fix more tests
* Move assist pipeline back to cloud after dependenceis
* Deconflict based on wake word
* Undo test
* Make wake up key a string, rename error
* Update snapshot
* Change to "wake word phrase" and normalize
* Move normalization into the wake provider
* Working on describe
* Use satellite info to resolve wake word phrase
* Add test for wake word phrase
* Match phrase with model name in wake word provider
* Check model id
* Use one constant wake word cooldown
* Update homeassistant/components/assist_pipeline/error.py
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
* Fix wake word tests
---------
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
* Add websocket command to capture audio from a device
* Update homeassistant/components/assist_pipeline/pipeline.py
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
* Add device capture test
* More tests
* Add logbook
* Remove unnecessary check
* Remove seconds and make logbook message past tense
---------
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
* Use webrtc-noise-gain instead of webrtcvad package
* Switching to ProcessedAudioChunk
* Refactor VAD and fix tests
* Add vad no chunking test
* Add test that runs audio enhancements
* Add pipeline option to save wake/stt audio to media
* Add debug_recording_dir to assist_pipeline YAML config
* Clean up and additional tests
* Remove I/O in event loop
* Organize saved audio by pipeline name and device id
* Record wake/stt debug audio in separate thread
* Fix after rebase
* Use timestamp instead of pipeline id for directory name
* Add WAV write error test
* Join thread in executor
* Make arguments for async_pipeline_from_audio_stream keyword-only after hass
* Use a bytearray ring buffer
* Move generator outside
* Move stt stream generator outside
* Clean up execute
* Refactor VAD to use bytearray
* More tests
* Refactor chunk_samples to be more correct and robust
* Change AudioBuffer to use append instead of setitem
* Cleanup
---------
Co-authored-by: Paulus Schoutsen <balloob@gmail.com>
* Add wake component
* Add wake support to Wyoming
* Add helper function to assist_pipeline (not complete)
* Rename wake to wake_word
* Fix platform
* Use send_event and clean up
* Merge wake word into pipeline
* Add wake option to async_pipeline_from_audio_stream
* Add start/end stages to async_pipeline_from_audio_stream
* Add wake timeout
* Remove layer in wake_output
* Use VAD for wake word timeout
* Include audio metadata in wake-start
* Remove unnecessary websocket command
* wake -> wake_word
* Incorporate feedback
* Clean up wake_word tests
* Add wyoming wake word tests
* Add pipeline wake word test
* Add last processed state
* Fix tests
* Add tests for wake word
* More tests for the codebot
* Assist pipeline to use configured values
* Include voice in TTS-START event
* Use correct tts language var
* More vars
* Apply suggestions from code review
* Update
---------
Co-authored-by: Bram Kragten <mail@bramkragten.nl>
Michael Hansen
tronikos
Erik Montnemery
Joost Lekkerkerker
Paulus Schoutsen
Thomas55555