What is the current best hardware for room-level voice assistance?

[gentle gyro]

I'm not looking for something like an m5 atom echo that you have to carry around or go digging to find- I want a replacement for an echo dot.

[sleek cradle]

Like right now - you can take https://github.com/formatBCE/Respeaker-Lite-ESPHome-integration

Tomorrow NabuCasa will present their device Voice PE - if you find it acceptable, you can try buying it (if there will be stock, which was promised).
Also there's pre-order for FutureProofHomes Satellite1, which will be modular and capable. Reportedly, it should be ready somewhere end of the month.

I personally use Respeaker Lite throughout the home.

[gentle gyro]

Does it actually work? The Seeed studio 2mic pi hat powered by the same stuff was absolutely awful...

And by that I mean it basically didn't work past 5 feet away

[grave wadi]

I don't believe that the Seeed studio 2mic uses the XMOS analog processer like the Respeaker-lite. Its has been my experience that the Respeaker lite with the ESP32-S3 works better than the Wyoming Satellite with the Respeaker 2 mic and Raspberry PI Zero 2W.

[sleek cradle]

Exactly. Respeaker Lite is the first Seeed dev kit with XMOS.

[sleek cradle]

It works.

[gloomy peak]

I'm also using Respeaker Lite (thanks in part to formatBCE) and it's been pretty solid.

[wind edge]

Guys, it is December 19th, do we have any new updates on voice kit? if so where will the dev announce it?

[gritty cedar]

https://www.youtube.com/live/ZgoaoTpIhm8?si=359yVRSaKDPCCKI5

[void knoll]

@sleek cradle that new Koala Satellite of yours looks great too! will have to follow that project as I see lot of potential 👍

[sleek cradle]

Thanks! I will detalize the instructions on assembling it. It really looks like off-the-shelf product after assembly, and is basically ready to be shipped as-is to those who want just connect it to HA same way it works with Nabu Casa PE device. 🙂

[gentle gyro]

so it sounds like the advice is either Respeaker Lite, Koala Sattelite, or HA Voice Preview?

[simple trout]

I will probably end up with all 3 in my home at some point. Very exciting seeing so much progress in this space!

[gentle gyro]

Sounds like it all started improving all at once, last I checked it was basically "Carry an atom echo everywhere for a mid experience, or use something else and either be unable to buy the hardware anymore or get hardware and have a terrible experience"

[sleek cradle]

Koala is made of Respeaker Lite.
Also there's FutureProofHomes Satellite1, another great choice with same XMOS chip.

[sleek cradle]

Yeah everyone was working in parallel with PE guys, and shown their work right after 🙂
To be frank, Respeaker Lite is on market for some time already - but it's still based on PE firmware, so was considered pre-release before PE released. 🙂 Same for Koala.

[gentle gyro]

I don't think I've had a chance to look at this suff sincemarch or so

maybe earlier lol

I just know that I tried a bunch of stuff and they were so bad I started reading https://pysdr.org/index.html and https://lwn.net/Kernel/LDD3/ in hopes of maybe improving them, e.g. the 2mic hat was basically doing zero signal processing at all and the driver was buggy

ended up having IRL stuff come up before I could finish reading them tho -_-

Glad to see there's a lot of movement in the space though 😄

[forest sleet]

Im tired of my echo too

its dumb and it got a attitude sometimes

like it isnt hard to turn off the light and turn on fan in one command

[sleek cradle]

This is currently possible only with LLM on HA. 🙂
Jetson Orin Nano Super or GGPT-4o-mini will do.

[forest sleet]

my home assistant green comes in tomorrow so ill set everything up then

[cloud tiger]

For local llm gotta be careful. I haven't really found one that works as well as gpt 4o mini. I tried qwen2.5 14b and simply asked if any lights are on, and it proceeded to turn on every single light in HA 🤣

[sleek cradle]

That's why I use local LLM only for common knowledge. All device-specific things are done with default agent. 🙂

[cloud tiger]

Lol yeah, it's nice to use it for chaining command and such, but I've only seen 4o mini do well with that

Or for stuff i don't want to code out that an llm is good at, like creating a weather forecast from weather data or summarizing calendar events/querying data

I could do it, but it's easier to have the llm do it 😁