#Hardware for Local Voice
1 messages · Page 1 of 1 (latest)
Later probably will be better but for best results you want to use a dedicated GPU as accelerator
I suppose either would work fine if you use the smaller models or the turbo ones
Tho keep in mind that you likely don't want to buy the smallest possible device in size like the ikoolcore as they tend to be quite noisy
Hi @faint jasper,
For local voice, I assume you want to run an LLM with Ollama or something. Whatever platform it needs a big GPU to run the LLM, so make sure you start there, then find a box it can live in.
If it's just for whisper and piper a smallish gpu will also do. The models are fairly lightweight
You are buying hardware, so now's the time to make sure the platform will work, even if you don't get the GPU yet. It is where the OP will likely want to go. And there is Whisper versions that are GPU enabled.
best results for voice? make sure you're using proper storage like ssd. voice use alexa/google. Voice PE isn't really ready yet.
It doesn't use either if configured accordingly
If that's what you want, buy it. Nothing we say here will change your mind. Remember this, though...
@faint jasper The XY problem is asking about your attempted solution rather than your actual problem.
I don’t think this can run Core (yet), but it can run the voice stack: https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/nano-super-developer-kit/
I would love to run this along side my NUC that is running HA.
If it wouldn't be unobtainable atm ^^
It may run core through docker, it shouldnt be too special of a hardware
you can get the larger nx or agx ones :smug: but at that price real gpus have more bang