Hi everyone,
I’m currently running Home Assistant on a Raspberry Pi 4 and using Nabu Casa for my Voice Assistant pipeline. I want to move to a fully local STT/TTS setup to improve privacy and reduce dependency on the cloud.
My main goal is to achieve a response time of less than 2 seconds without a massive increase in power consumption. Currently, my Pi 4 struggles with local Whisper (taking 8-10s), which makes it unusable for daily tasks.
I happen to have an Intel Neural Compute Stick 2 (NCS2) from a previous project. I’ve seen some older threads about using it with OpenVINO, but I’m not sure if it’s well-supported in the current Wyoming/Whisper ecosystem in 2025.
My questions:
- Is anyone successfully using the Intel NCS2 for Whisper STT inference in 2025? If so, how is the performance compared to a modern Mini PC?
- For those with sub-2s response times, what hardware are you using that keeps power consumption low? I’ve heard a lot about the Intel N100, but would it still need the NCS2 or an iGPU passthrough to be that fast?
- Should I stick with the Pi 4 + NCS2 (if possible) or is it time to upgrade to an x86 Mini PC?
Looking forward to hearing about your local voice setups! Thanks