I did see something recently about local LLMs and voice input layers. The post made it seem very Jarvis like, think it may have been the voice used or the name.
Knowing nothing about tech other than I want my privacy I am hoping it is feasible for the common man
There’s a mode for voice control that is even friendly to a Raspi 4 or 5, but it’s very simplistic in control, basically a super lightweight speech to text trained only on device names and aliases. Think the speech to text in late 2000s through early 2010s non-smart phones.
Small models for faster-whisper will run on even my little Dell Micro i5-6500T that I have Home Assistant running on, it’s just a little bit slow, but it absolutely works and is usable speed! I run a larger model currently offloaded to my server, which has an RTX 2070 Super in it, but that’s to make it perform more like how Google used to a long time ago, and it’s unused power most of the time.
They’re trying to make it as accessible as possible for sure. There’s even options to use cloud STT and TTS (they even include it in the Home Assistant Cloud optional feature), but it’s definitely cool as hell to be able to talk to an open-source-design speaker and get a reply and control any switches or lights or even my thermostat and robo vacuum without needing the Internet to work. As long as my Wi-Fi and HA box are up, I’ve got options!
It has several modes. The most basic is speech to text, pattern match, then implement. It also has text to speak for feedback. No actual AI in the loop.
It’s also capable of tying to AI models in various ways. It’s mainly intended for question answering. Either general, or about your data.
I personally don’t trust a non-deterministic AI having direct control of my house, so the split is useful.
I did see something recently about local LLMs and voice input layers. The post made it seem very Jarvis like, think it may have been the voice used or the name.
Knowing nothing about tech other than I want my privacy I am hoping it is feasible for the common man
There’s a mode for voice control that is even friendly to a Raspi 4 or 5, but it’s very simplistic in control, basically a super lightweight speech to text trained only on device names and aliases. Think the speech to text in late 2000s through early 2010s non-smart phones.
Small models for faster-whisper will run on even my little Dell Micro i5-6500T that I have Home Assistant running on, it’s just a little bit slow, but it absolutely works and is usable speed! I run a larger model currently offloaded to my server, which has an RTX 2070 Super in it, but that’s to make it perform more like how Google used to a long time ago, and it’s unused power most of the time.
They’re trying to make it as accessible as possible for sure. There’s even options to use cloud STT and TTS (they even include it in the Home Assistant Cloud optional feature), but it’s definitely cool as hell to be able to talk to an open-source-design speaker and get a reply and control any switches or lights or even my thermostat and robo vacuum without needing the Internet to work. As long as my Wi-Fi and HA box are up, I’ve got options!
It has several modes. The most basic is speech to text, pattern match, then implement. It also has text to speak for feedback. No actual AI in the loop.
It’s also capable of tying to AI models in various ways. It’s mainly intended for question answering. Either general, or about your data.
I personally don’t trust a non-deterministic AI having direct control of my house, so the split is useful.