There’s 2 approaches:
a) use an STT assistant that works 100% locally such as mycroft.ai. Speech recognition won’t be as good as Alexa is but it may turn out to be sufficient for your needs, particularly if you only use it for the purpose of commanding your smart home (i.e. a very limited set of key words and logic).
b) setup any cloud based assistant outside your LAN and connect into your openHAB instance via myopenhab cloud. Alexa and GA will not talk to any of your devices, and all commands are gated (hence filtered) through openHAB.
Even this cloud feels not good for me. (Or may there be a full working switch for working one day with that cloud and another independent of it?)
Lets go in details for a)
Which “source” and “sink” are possible for this approach?
Any small devices (prefer PoE) or do I need to run a full OS? If so which Software will best provide a interface mor microphones and speakers to OH3?
which one, myopenhab ? It’s operated by the openHAB foundation and it’s just a gateway
that gives you 100% control of what happens on any input from the assistant.
(you could also setup your own instance but that’s barely worth the effort)
I’m not getting that point.
Don’t overdo your cloud avoidance attempts. Make yourself aware first what aspects of cloud you really do not accept and which ones might be okay, and make yourself aware that the more local the more efforts you’ll have to put in, and that’s by no means a linear relationship !
b) is a pretty good and privacy-valueing compromise. I’m telling as someone who has extensively explored and tried those options.
Distinguish STT from TTS. You don’t have to use the same device or service for both.
Many options for TTS and no relevant point not to use Google TTS or an Echo with the amazonechocontrol binding for that.
There’s also I think MaryTTS and others that work locally on your openHAB server.
Combining arbitrary TTS with arbitrary sinks is usually a hassle to get it working in all desired combinations.though. You might need extra software such as pulseaudio on your server.
The challenge is STT.
You’ll need one that an openHAB integration exists for so AFAIK you’re limited to those three assistants I listed.
Mycroft is OSS and available as a product like an Echo is, too. Check it out yourself and search the forum for more information.
Or you can install the SW yourself on a Raspi w/ a proper microphone but note then you will need some sort of micro array with echo cancellation etc such as the Seeed Respeaker for this to work well.
The Amazon Echo Binding, Web Socket error keeps spamming my log with error messages, and generally has ongoing issues compared with all my other bindings. I have tried “updated bindings” for over a year now. I give up. You would think a trillion dollar company like Amazon could help, but no.
Honestly I would think it’s pretty unfair to expect Amazon to help everyone write separate app integrations. They publish the API, and as long as their API works and doesn’t constantly change or have unexpected breaking behaviors, they’ve done what they can do.
Now, does the API work perfectly? I don’t know. I’m not sure if the web socket error stems from incorrect API behavior or a bug in the binding. But it sure does seem to be consistent. And there’s a fix available, and it’s even easier than ever now that we have the ability to install third party bindings from the UI.
We’re using free, open source software developed by unpaid volunteers. Yeah, there may be issues now and then, but I for one an grateful that it exists at all.