I am looking for a Intel I3 9th gen with 16Gb ram and 125Gb ssd + 256Gb SATA SSD mini Pc for running HAOS as VM on Proxmox. Is it good enough for running local voice.
Hardware I found in local markets include Dell OptiPlex 7070, HP MP 9 G4 HP G800 G4 etc. Which one of these is recommended for local Voice.
For ZIGBEE I am using SLZB-06M.
I have an Optiplex 7050 SFF with an i5 7500T CPU, running proxmox and HAOS. Mine is one of the low power, Small Form Factor models which are too small for any video card, "but HAOS runs headless so that's ideal" I thought. And then the voice processing wants more AI-style processing ... and then they added AI proper ...
I run Voice Assist with only one HA Voice PE, and I have opted for Home Assistant Cloud for the speech-to-text and text-to-speech - but that's because I didn't want to spend the time fiddling with all the local options. Mine rarely goes over 2% CPU utilisation, so I should revisit the local STT and TTS.
I suggest you get a PC with a case large enough for a full-size video card to give you the ability to upgrade to local AI in future.
The voice question is. What experience do you want.
Voice comes in multiple flavors.
Your box by itself will run HAOS fine.
If you want speech to phrase. It should work fine. Speech-to-Phrase brings voice home - Voice chapter 9 - Home Assistant
If you use cloud llm / cloud STT it will also be fine.
Local STT on CPU with Piper / ONNX is perfectly fine... (if you have enough Cpu cores / aren't overdriving your box)
BUT If you want locally driven LLM you. WILL (not optional) need an appropriate GPU or inference core.
So you think this puts Beelink out of the running as a good option? Just curious
Firstly, I defer to Nathan, especially when it comes to AI.
I'm not familair with Beelink - is it an intel micro PC or NUC, or an SBC (like Raspberry Pi) ? Doesn't matter.
Yousaf was talking about buying a mini PC and in hindsight my recommendation if he is thinking of AI in the future, is to go for one in a desktop or tower case that has space for a full-sized video card ... much easier to add a video card later. I am sure that Nathan wouldn't have mentioned the EGPU if it wasn't acceptable - but to me having it as an add-on looks awkward and more expensive.
Curious ... initially Raspberry Pi was the recommended hardware for Home Assistant (because it was cheap, low power and devices could connect direct to GPIO pins) ... then gradually HA evolved to become a server with devices connecting over LAN, still with low power as a feature ... and now the incorporation of AI with HA, we want a powerful GPU which means a high power PC.
Something has also happened to the price/performance value.
Initially RPi were superior in this, but RPi have become more expensive and miniPCs have become cheaper, so they are even or maybe MiniPCs are even a bit ahead today.
If it was me right now I would not be considering local AI, unless you want to be running na expensive and power hungry PC 24/7. Local Ai needs a lot of PC for not a lot of gain.
Actually, I just voice for replacing Alexa, not for AI. I just need to turn on the lights with a voice command.
So, something like this? Home Assistant Voice Preview Edition - Home Assistant
But voice processing will be happening on the HA server itself isn't it?
Sorry for going off-topic.
As Nathan mentioned in the third post above, there are several levels/options - of which most should run happily on any of the PCs you mentioned, and still give you plenty of room to upgrade the hardware later.
It comes down to how sophisticated you want your voice assistant to be. Note that all Open Source alternatives still fall short of the experience from the multinational players with exceedingly deep pockets ... and most "reviews" I have seen fail any device for not being a duplicate of the amazon/google/apple experience.
If you want to control your home assistant by predefined voice commands (eg "turn the kitchen light on") then yes, that is easy on the PCs you mentioned. The simplest option is to make use of Home Assistant Cloud to do the Speech-to-text (STT) and text to speech (STT) processing - this uses Nabu Casa external server processing to give a very smooth and quick response ... but at a small monthly cost.
Speech to phrase runs as just another integration on your HAOS server, which requires a bit more setting up and uses more of your server's CPU and resources ... but shouldn't be an issue until you have quite a few satellite devices (microphone and speaker units like the HA Voice Preview Edition) around your property and responding to commands at the same time. Depending on your PC, responses could take a little longer; but the advantage is that all processing is totally in your home.
However most people have got used to amazon/google/apple devices and expect them to do a lot more than just turn lights on. HA Voice Assist is able to link to other HA apps to tell you the weather, etc. ... but that is far short of the generalised Artificial Intelligence the big players provide. It seems to me that these days a lot of the HA Voice Assist effort seems to be going into AI and implementing the LLMs (Large language models) ... hence this thread drifting in that direction 
I suggest to give HAOS a try - there's a good post in the community on how to install HAOS on proxmox. You can add a voice satellite device or three, and try the various Voice Assistant options to see how well they work for you in your environment ... without much up-front cost.
I am currently using HAOS as VM in Proxmox on BeeLink EQ11(N100 16GB ram) but without voice. Want to set another node at another location for HAOS with local voice and 2 satellite voice nodes just to turn on and off the lighs etc no big AI stuff.
Agreed, I now have a mishmash of (well working) and RPI4 running Weewx and some daemonms that feed data into HA via REST API - the HA being on a RPI5 in a pironman 5 case running rasbpian (so the display on the front works) with QEMU/KVM VM running HAOS.... I will be switching to some more powerful hardware at some point... ! I am leaning towards beelink