I'm convinced that voice technology could be positively impactful for my life, but unwilling to use it when it's driven by companies who profit from the data collected. There should be an open source, secure, self-hosted option for voice assistance, how can that become a reality?
There's mycroft[0] that uses Mozilla's Common voice[1], but last time I checked the speech-to-text stuff was still being done in the cloud.
The Speech to Text was taking a general approach that isn't nearly as well optimized as FAANG-level STT. So it needed a GPU to run.
A few projects are available to do this. The most advanced one is probably Mycroft, which is usually run with an external account but can be completely self-hosted.
I remember a few like Almond and one posted to HN a few months back that could change the color of some lights right in the local client session, no internet connection needed.
Still only use them for two things - the weather and setting alarms/timers. Great stuff