I've accomplished this with the Atom Echo and they work... fine?

The speaker is essentially inaudible, but the mic works well enough for me to just yell at HomeAssistant to do things.

And hey, can't beat the size/price/power footprint and the deployment with ESPHome takes like, 30 seconds.

I only have one at the moment. I bought a cheap Dell Wyse thin client from eBay, installed a light Linux distro on it and set it up as a Wyoming Satellite and as a snapcast client. That is connected to an old all in one mini HiFi for speakers. For the mic I just have a small omnidirectional USB mic plugged into the front. Biggest hassle setting it up was ALSA since snapcast and Wyoming were fighting over the devices

I have a couple of Pi Zeros around the house I use as media players. They were running piCorePlayer. I replaced just the software with a vanilla Pi OS and installed Squeezelite and then Wyoming Satellite. I added a microphone and an automation to silence the media player as soon as a wake word is detected.

Voice recognition is adequate but I wish it was smarter.

(I should finish that blog post...)

I'm still dabbling with this, but so far I have a Pi Zero2W with a reSpeaker 2-Mics Pi HAT and an old speaker (no idea where from) that (luckily?) has the correct connector on.

It's ok as a geeky POC, but it needs work to gain sufficient WAF

I host Wyoming on a server with an Nvidia 4060ti, and point home assistant, on a vm on another server, to it. It works great.

I am making an esphome system for my old Yamaha AV receiver. RCA out via a cheap i2s chip, mems I2S mic in the enclosure that can hear well enough in the room, and an IR blaster output to change the AV receiver channel to the esp when voice is activated and change it back after.

Cool side effect is that it can also be used as an HA media player for streaming Spotify and local music to the speaker system.