How to properly take control of a pre-flashed Atom Echo in ESPHome?

[toxic cradle]

I was trying to take control and reflash, but am getting this error

INFO ESPHome 2025.11.0
INFO Reading configuration /config/esphome/m5stack-atom-echo-084b54.yaml...
Failed config

packages: [source /config/esphome/m5stack-atom-echo-084b54.yaml:5]
  
  Failed to load packages. voice-assistant/m5stack-atom-echo.yaml does not exist in repository.
  m5stack.atom-echo-voice-assistant: github://esphome/firmware/voice-assistant/m5stack-atom-echo.yaml@main

Guess the url is pretty outdated in this version. This is the new configuration

https://github.com/esphome/wake-word-voice-assistants/blob/main/m5stack-atom-echo/m5stack-atom-echo.minimal.factory.yaml

[distant glade]

you can flash to latest version by USB using THIS page then take control.

[toxic cradle]

I would like to do this remotely, if it's possible

The yaml configurations seem quite complex though. Which one of the three is used under the hood?

Also not sure how to get it into download mode. It doesn't show up at /dev/ttyUSB0

[distant glade]

you could get the binary file and send it via esphome ota i beleive

[toxic cradle]

Let me try a kernel update. That used to fix problems with ch341 module in other cases for me

[toxic cradle]

Worked after updating and rebooting

[distant glade]

awesome, good luck with it.

[toxic cradle]

I am already running into pipeline errors, probably because the RAM is full when too large contexts are sent

😄

[distant glade]

the AE is a good device for messing around and basic testing but its pretty limited when compared to something like the Voice-PE.

[toxic cradle]

Yeah, but the Voice PE is too expensive. It's just not worth 70€ with that crappy speaker. I think I will give the ESP Muse Luxe a try https://raspiaudio.com/product/esp-muse-luxe/

That one I could also use as a media player for e.g. the bathroom or kitchen

[wanton ore]

You're missing the point. The speaker doesn't matter, if it can't hear you well.
Currently there's no better HA satellites than based on XMOS chips. It's Voice PE, Respeaker Lite, FutureProofHomes Satellite1 and (lately) Respeaker XVF3800.

The cheapest is Respeaker Lite, and I have 5 of them working good across my home. Playing with replacing one of them with XVF3800 in the living room, since that one has the most advanced (in theory anyways) beam forming, AEC and NC capabilities.

Regarding the speaker - you anyways better off with using external speaker via 3.5mm jack.

[distant glade]

I use a soundcore mini on one of my VPE's for a better speaker and it works nicely - details HERE

[toxic cradle]

You just need a dual mic array. I have very good results with a WM8960 Pi hat, which costs 20€ and not 70€ like the XMOS chip. I am not paying 70€ for a satellite with a bad speaker or just a PCB. For that money I could get a second-hand HomePod mini and could just use Siri.

[toxic cradle]

I don't want any cable mess laying around. I am not buying a 70€ satellite to then hook a 20€ speaker to it. It just doesn't make sense. I am going to try out other options. If the ESP Muse Luxe doesn't work well, I am going to build my own speakers.

In any case, the Atom Echo is the worst option so far. I am definitely going to return it.

or maybe use it in the toilet without a response

[light cedar]

Hey @toxic cradle I just developed some fixes to make my Atom Echo work - maybe it’ll help you out? https://community.home-assistant.io/t/atom-echo-works-for-a-while-after-restart-but-then-stops-responding/657761/14

[toxic cradle]

Thanks, doubling the ring buffer was already enough for my case it seems 🙂

[wanton ore]

Ok I see you have your opinion. Good luck. I tried like dozen of different approaches in last 2 years, including several simple arrays. But I won't be convincing you. 🙂
P.S. Respeaker Lite with pre-soldered ESP32S3 is around 30 bucks.

[toxic cradle]

I am German and whisper won't understand me correctly anyway, even if I use a high-qualit microphone. Without an LLM in the background instructed to correct wrong words, it wouldn't work for me at all. This approach works quite well btw.

Imagine you tell it to "turn of the kitchen light" and whispher understands "churn off the hitch and might". The LLMs are able to figure that out.

[wanton ore]

Well that's another problem - which doesn't cancel requirement for good mics. Especially for wake word, that can't be corrected by AI.
But if you have LLM, you probably have good enough hardware to use better models. Unless you use cloud AI subscription - but in this case why not use HA Cloud STT which is great?

[toxic cradle]

gpt-oss:120b is for free in Ollama

I doubt HA Cloud can compete with it

[wanton ore]

Exactly, so use good model for Whisper too. What you have now?

[wanton ore]

It can, but that's another topic.

[toxic cradle]

Even the large whisper model sucks at German. I am using the turbo model

Unless German gets its AI shit together, there won't be anthing better

[wanton ore]

Probably better mics would mitigate that partially..
But yeah I understand your trouble. Well, nothing to do with VA hardware. 🙂

[toxic cradle]

It doesn't work with an iPhone even

Whisper is mostly trained on English data sets. It is what it is

[wanton ore]

Well, then HA voice (or for that matter any voice) just not ready for you.

[toxic cradle]

That's the point, but it'll get there

[wanton ore]

But still it doesn't discard the points about hardware. And the fact that STT is bad in German so far doesn't justify cheapest hardware.

[distant glade]

thats cool, making your own is definitely an option. but if your using it as a satelite you should looking at using the XMOS chip or something similar to help clean up voice and filter its speaker out of the microphone and stuff. for some examples you can see how stuff is setup on on the VPE on its schmatic diagram HERE and the xmos firmware used is opensource HERE also the guys over on the esphome discord might be able to help.

[dire phoenix]

May I ask if Resppeaker XVF3800 is better than Voice PE? Is there anything special about him? I currently have two Voice PEs, and if it's better, I would like to replace one of them.