Inspired by the Alexa Guard functionality I would like to propose a project to build a similar functionality in a Mycroft style (open source and with privacy).
This will require to train a specific model for Mycroft Precise that covers sounds of alarms from smoke and carbon monoxide detectors and breaking glass. That model will probably run on a special Precise-instance (in parallel to the Wakeword listener instance) that can trigger a Mycroft “Watchdog skill” and passes a recording of the detected sound (e.g. glass breaking) to the skill.
Possibly actions of the Watchdog skill could be:
send a push message to a mobile phone including the sound recording
turn on light in the house in order to make a possible burglar go away
Anyone interested in such functionality and/or willing to participate working on it?
There’s another similar use case as well in terms of emergency situations in Australia. Australia is pretty, well, burny, because climate change, and our asshat government is in love with coal. I digress.
Long story short, we frequently get bushfires. We have excellent fire service / emergency procedures, primarily because bushfires are so frequent. Our public radio will broadcast the `“Standard Emergency Warning Signal” if there is an imminent threat. It sounds fairly scary, because it’s supposed to. It usually means there’s a bushfire nearby, or a cyclone on the way.
What if … Mycroft could “listen in” to the radio and determine if it hears the emergency alert? Then wake people up on the Device or start alerting?
There’s so many caveats with this of course; you need internet access to listen in to the radio, in an emergency there may be no internet, the risk of a false positive will make people not trust the system and so on.
But there has to be a way to harness voice for emergency alerting.