#AutoReader
finite aspen
The basics are working: I have created an automation with a sentence trigger (the last line before any character's line), and the STT and TTS quality (using cloud) are fast and good enough to be usable. I have a good, user-friendly input scheme as well.
The main issue I have is that I can't use a static wake word or button press to trigger it, because the former would mess up the flow (can't say "ok nabu" before every line) and the latter would prevent the use of props or hand motions without a prop.
Is there a way to do this? Is it easy enough to train dynamic wake words and have something like 10 new wake sentences per day? Can I constantly do sentence detection with cloud, or can I do it all locally if I get a ~1070 class GPU?
The other less major but still important piece to this is that when I open assist from the browser or Android app, it responds with "done" every time. Is there a way to turn that off? I only want it to respond with the specific sentences I've input.
Thanks in advance!
rain swan
That's not suitable for the default conversation engine, but you can code a custom conversation engine which can do just that: scripted dialog. You'd be a pioneer, though, so don't expect any tutorials
finite aspen
Got it, are there any good starting points for this e.g. engine implementations that have tunable parameters for things like follow-up listen time (or maybe you're saying there aren't any agents that are set up for actual conversations?)