Post by @lina@vt.social

Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

RE: https://vt.social/@lina/117579418539240256

I want to replace it with a local Home Assistant voice thing, but first I need to train an ethical language model for it (it is possible but existing ones are not).

Then either buy entirely new hardware, or replacement ESP32 PCBs, since nobody rooted the Home Mini yet...

(It is possible to do this without an LLM but it would be janky. Small LLMs are actually useful for building human/computer interfaces like this, and it is possible to do it with fully ethical training.)

Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

@lina@vt.social 6 hr. ago

Google Home just read me an ad for Gemini in response to a voice command. 💢💢💢

I almost threw it out the window...

I really need to replace this thing.

0 Boosts

0 Likes

Alex

@stsquad@mastodon.org.uk 17 min. ago

@lina I'm sure one of the FLOSS home automation projects has collected a bunch of training data for this under an open license. Possibly Mozilla? https://en.wikipedia.org/wiki/Common_Voice

Common Voice

voice dataset by Mozilla

Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

@lina@vt.social 10 min. ago

@stsquad TTS/STT is solved, the issue is a clean language model to go between them. I want to train (fine tune) one based on KL3M.

HA has a training dataset for it, just need to put the pieces together (and probably extend the dataset a bit)

This federated comment is not yet available here.

Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

@lina@vt.social 28 min. ago

@koakuma Some people call them SLMs but nobody agrees on the threshold!

aaron

@fireborn@dragonscave.space 32 min. ago

@lina I do a lot of this already, apart from the self training. I have no idea where to start with that, but local LLMs for home automation/management is a huge win, especially when all the devices are on a VLAN and have no connection to the internet at all.

Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

@lina@vt.social 29 min. ago

@fireborn Yeah I already have everything on HA (and all questionable IoT devices on a separate VLAN). Just need to sort out the voice frontend, both the software/model and the hardware.

I have a rough plan for the model but I need to do some research.

Discussion

Discussion