olive harness
Hello Community,
Around two years ago, I set up two Raspberry Pi 4 devices, each equipped with a Seeed microphone HAT and running Home Assistant with Rhasspy. Voice commands were processed through Node-RED, and I housed the devices in Logitech speakers to achieve decent audio quality.
After moving to a larger space, however, the setup became unreliable. It frequently triggered false positives, and when it should have responded, it often failed to recognize commands—unless I spoke loudly or shouted. Since we didn’t have a strong need for a voice assistant at the time, I took the devices offline rather than investing time in fine-tuning them.
With the recent improvements to Home Assistant's voice assistant functionality over the past few releases, my interest has been reignited. I’m now debating whether to rebuild the Rhasspy-based setup or explore Nabu Casa’s official Voice Assistant (is "PE" the correct term for the dedicated hardware?).
Audio quality is important to me, and I would prefer to use a high-quality speaker for the assistant. While it might be possible to integrate the PE device into a speaker enclosure, I assume that would render the physical button unusable.
For context: I now run Home Assistant on a more capable machine—an Intel i3-12100 system hosted as a VM on Proxmox. I’ve also enabled GPU passthrough for another VM (using the integrated GPU of the i3-12100). This leads me to wonder:
Would it be feasible to run a local LLM on this hardware at an acceptable speed? And is it even possible to share the integrated GPU across multiple VMs?
I would greatly appreciate your insights. Would it make more sense to revisit the custom Rhasspy solution, or is the official Voice Assistant a more viable and future-proof direction given the current state of Home Assistant?
Thank you in advance!