voice PE feature status

[ashen vault]

just wanted to ask some questions about the voice PE

dose it use the same integrations as other satellites use or dose it have its own?
can u use my own voices or wake words?
can i set times or reminders for certain things and also be reminded after when the times up?
how dose implementing a LLM work? is it just a fallback if the question cant be handled or is it LLM or no LLM?
can i run things in the cloud instead of my server, if i have a subscription to nabu casa?
is there anything to watch out for atm, crashes? bugs? etc.

[gleaming pecan]

if uses the esphome integration

it comes with some stock wake words and there are a few more around you can switch to if you take ownership and add them to the firmware config.
totally custom wake words are not currently easy to do. (but not impossible if you are so inclined but results my vary)

if you run piper locally for TTS then you can use various models for that including custom ones for example made with TextyMcSpeechy

you can indeed set timers. as for reminders on a calendar.... that can be done with automations

you can link it to a LLM in the voice pipeline. and you can set it to try and resolve locally but fall back to LLM

you can run STT and TTS in the cloud with NC cloud

even with llm's its perhaps not a perfect drop in replacement for other products from huge companies. however it does very well and its constantly improving

[ashen vault]

thank you so much :)

[zenith basin]

"it comes with some stock wake words and there are a few more around you can switch to if you take ownership and add them to the firmware config.
totally custom wake words are not currently easy to do. (but not impossible if you are so inclined but results my vary)" do you have any guidance documents on how this could be done? Uploading the custom wake word files into the share folder location still have them show up for the ESP home section like they do for they wyoming satellite section under devices. Wakeword is available for selection in voice assistant under wake word section

[gleaming pecan]

I think you are talking about wake words for "openWakeWord"
These are not the same as wake words for "microWakeWord"
to be clear:
the process of just uploading files to the share "WILL NOT WORK" with a device configured to use MWW
the models you make for OWW "WILL NOT WORK" with MWW

so you will either:

need to convert your VPE firmware to use OWW then on the voice pipeline you can add a "streaming wake word" which will allow you to use OWW models. there are some examples of this conversion around but they are mostly not up to date and not supported. however if your so inclined it can be done if your familiar with esphome config.

the other option is to train your own MWW model there is some documentation on the github repo for MWW however there is no "friendly" process. but if you do manage to train a model then you will then need to edit the firmware config on the VPE to point to a custom repo storing your model

[zenith basin]

makes sense, assuming at some point custom wake word will be better implemented at some point, will just hold off for now