Ive browsed the form, read what I can find online - nothing suggests mycroft cam work in a mobile net-isolated system.
That is, no matter if installed on my own hardware or the One, it REQUIRES net connection to function.
This seems to render the whole thing of only limited use on mobile platforms such as RVs. ( I travel extensively in my own vehicle and have some ideas regarding voice activated control of vehicle systems and accessories. But if mycroft is non-functional because my customer is in a cell-phone void, it rather renders the thing pointless)
Am I missing something?
Hi @Erik_Kloeppel, thanks for your feedback.
At the moment there are some dependencies which require internet connectivity. Let me walk you through them in more detail, using the approach of our voice stack
-
Wake Word - the Precise Wake Word engine does run on Device, it has no internet dependencies once it’s loaded, or if it needs to load a new model.
-
Speech to text - this is the hardest part of the voice stack to get to work offline. We’re currently partnering with Mozilla to implement DeepSpeech as an STT option, but currently it has to work cloud-based due to the high compute requirements. DeepSpeech on-device is the ultimate goal but is likely a little while away.
-
Intent Parsing - both the Adapt and Padatious Intent Parsers run locally on-Device.
-
Text to Speech - the Mimic TTS engine works on-Device
-
home.mycroft.ai - currently each Device requires a connection to home.mycroft.ai because this layer provides a lot of abstraction to other services, like Wikipedia, cloud-based STT, location services and so on. In our roadmaps you will see that we plan to provide an offline version of Mycroft that is not dependent on this layer.
-
Skills - Skills themselves often have online dependencies, particularly if they are gathering information from an API and presenting it via voice. So unless the Skill itself is completely contained on-Device then there will always be an online dependency at the Skill layer.
Kind regards,
Kathy
2 Likes