Tries to fetch the label list from the mlmodel file itself and due to newer converted mlmodels having a different output (float32) it also numpy converts the model output to floats.
* remove unused dependencies
* expose configs to enable/disable two-way as experimental
* remove Few logging and clean up settings
* initial imap settings scaffolding and 90-min stream refresh
* fix sse refresh and make it configurable
* fix double printing in provider
* clean up logger names
* safer imap re-login and proper SSE exits
* some more logging with ids
* fix MQTT refresh and cleanly cancel tasks on restart
* document IMAP requirements
* bump 0.6.3 for release
* fix new device load error
* bump 0.6.4 for release
* parse endpoint and ffmpeg args
* aiortc logging, detect ffmpeg input source format
* log intercom source
* bump
* write sdp to file
* bump
* stop loop correctly
* run ffmpeg to convert inputs to known format
* options don't seem to do anything
* bump
* refactoring to use webrtc for video
* pass through ffmpeg to clean up stream for PyAV
* relay talkback audio
* cachebust
* mute talkback with silence
* cleanly exit orphaned subprocesses
* rename class
* bump minor version
* fetch webrtc stream from prebuffer
* prebuffer still not working
* Revert "prebuffer still not working"
This reverts commit b30ccebb17.
* Revert "fetch webrtc stream from prebuffer"
This reverts commit 8dbd308c59.
* try to fix buffer overrun
* expose option to purge pip cache and reinstall deps
* some more logging
* use v1 pypi
* bump 0.5.2
* per-camera settings to toggle webrtc implementation
* change setting name
* wire up existing push to talk for Intercom
* documentation
* bump 0.5.3
* 0.5.4 with new python dependency versioning
* bump av build
* 0.5.5
* optimize event waiting by reducing polling
* more minor optimizations
* tweak performance and requirements.txt
* reimplement with pion webrtc
* bump 0.6.0
* fix interfaces for doorbells
