I’m curious what sort of functions I might be able to access from a mycroft device that isn’t connected to the internet; for example, while traveling, camping or when the power is out. It would be nice if I could still use mycroft to control smart devices on the same network, do simple calculations or communicate to other mycroft devices on the same network.
In this example, I presume the mycroft units would still be alive by virtue of UPSs or because they’re running as software within laptops that still have charge.
EB
MyCroft will be a connected device.
The speech analysis will be done in the cloud (it needs a lot of processing power) so, no internet, no MyCroft.
1 Like
The SmartThings hub works off the grid.
Some kind of fall-back STT could be useful maybe? Probably impossible on a RPi and would probably take a long time to process on a more powerful machine - for now - but it could also a stepping stone towards bringing STT functionality from the cloud to the local device in the loooooong term.
2 Likes
Well, Sphinx can be used for that (maybe still used for hotword detection), but as you say, it’s veeeeeeeery slow and not effective at all at recognizing speech according to my tests (maybe better in english, but in french, it was very bad)…
But maybe MyCroft will come with something better…
2 Likes
Queue big STT annoucement @ryanleesipes (it is the 14th…)
1 Like
Sphinx has at best approx. 60% recognition rate locally (which is terrible in the scheme of things). We are working towards solutions that move this direction, but it is going to be a marathon - not a sprint.
As @Autonomouse alluded to, today is a good day to watch openstt.org - some neat stuff is gonna pop up there.
1 Like
i’m eager to see that. I hope I will be able to play with that later tonight =)
something i would like to see, is the ability to install your own MyCroft server hosted off a home ubuntu server. the daemon for the service could also be called “WitchCroft” lol 
what would be cool, is if you could have a daemon service on a local server that has databases of wikipedia, freedb, and other open source knowledge bases. that way - if there is a zombie apocalypse, you could still maintain the A.I. locally hosted off your own machine. or have it set up, to where it could use your own machine for the primary server, and the MyCroft servers for secondary. that way it could become trained during normal usage. or possibly a split function to where it sends the data to both servers. that way there would always be a backup resource.
2 Likes
You know that someone, somewhere will actually do that… probably more than one actually
you know something that would be really cool, is after the project is complete, have a user group setup a p2p type server network that could host and share knowledge repositories. that way a community of users could come together and pool their servers as knowledge nodes, and see how suped up the AI could become. that would be really cool kind of a linux community AI project. that way, the community could create their own quasi cloud env with out having to spring for the money to buy a who datacenter cloud env. just have it set up to transfer data simular to bittorrent, or over a tunneled ssh line somehow.
3 Likes
Honestly, I don’t see this as too tough - from the YouTube presentation of Mycroft architecture, all pieces talk over a Web Bus on a specific port. Spinning up a service to run on a central machine shouldn’t be too hard - then just redirect the “clients” (Mark 1 / Pi / Desktop) to send their waveforms to the local Server instead of the Cloud…
The hardest part could be programming the redirection & setting up the Config.
Maybe I need to dust off my programming skills…
what would be cool, is if you could have a daemon service on a local server that has databases of wikipedia, freedb, and other open source knowledge bases. that way - if there is a zombie apocalypse, you could still maintain the A.I. locally hosted off your own machine. or have it set up, to where it could use your own machine for the primary server, and the MyCroft servers for secondary. that way it could become trained during normal usage. or possibly a split function to where it sends the data to both servers. that way there would always be a backup resource.