hidden hatch
I'm using Voice PE with the local option and whatever stuff it installs by default (wyoming, piper, whisper). The results are unusable - more than half of the time I say something to it, it mishears my words as other words.
What are the next steps if I don't want to pay for Nabu Casa subscription? My Home Assistant CPU is only an i5-8500T and is running HAS in docker, so I'm unsure what language models I'm capable of running.
grave frigate
Use a larger Whisper model, with larger beam size. You may need a GPU (with CUDA cores) to make it work well
STT is an intensive operation and will cost you either time, hardware or subscription fees for someone else's hardware
hidden hatch
Use a larger Whisper model, with larger beam size. You may need a GPU (with CUDA...
I found a list here; https://alphacephei.com/vosk/models, is it as simple as dropping this file in my HAS disk and changing a line somewhere?
This poor performance seems deliberately deceptive of novices from official channels, as I don't remember ever seeing this limitation stressed in any of the release party videos, but perhaps I'm wrong on that. Kind of annoying.
mellow sedge
Many of the videos repeatedly stress you need powerful hardware and "need to know what you are doing", and throughout the discord and probably on the forums as well you can see where people repeatedly say you generally need a gpu and your own docker server running the components to achieve cloud level performance close to what nabu cloud has.
grave frigate
...especially in languages other than English