Unfortunately I’m a government contractor of retirement age, and my employer is busy putting a couple million contractors like me out of work. The future for people of my age in the US is pretty bleak, that’s partially why I’ve been focused on smaller implementations since last summer. I’ve also had to reduce my living footprint and am now looking at moving abroad within the next few months (contract ends in June). So I’m down to only a couple Pi’s and a couple monitors (my grandson took most of my collection with him to college this year). I have to reduce my footprint further yet, soon.
To get where I was, I experimented with MS Cognitive Services and Google Voice (I do a lot of my work in STT and TTS) and Mycroft/Picroft. I also pruned my own versions of Pocketsphinx and its relatives. My problem was the time it takes to process and respond, so real interactive I/O ran through the Linux box, as did some of the larger image files. A lot of the speech I was recorded and I set triggers for playback. Generating 3D figures was fun but such a time suck! I ended up playing with one figure in multiple skins, repeated across the monitor array in slightly different poses. My Python scripts were great fun but I’m not a great Python programmer: a lot more could have been done using a cleaner code stream (and something like Tensorflow). I didn’t get there because of Joyce’s epiphany: I have long been looking for a device to create that could generate a long-term income stream for my impending retirement. I’m on the cusp of that.
Once set up in an affordable destination, I will probably turn back to what was most fun: creating and animating the 3D figures for the large interactive video array.
I live in the home town of Meow Wolf and I see the video array as a powerful competitor for them: their stuff is almost purely built-world and takes forever to construct and/or change anything. My system would be almost instantly portable: a series of monitors, Pi’s and peripherals controlled by a Linux box. Once an array is set up and operating, reloading the array with a whole new “show” would take a couple hours at most. And once a library of “shows” is built up…
The US is littered with declining retail malls, some of which are trying to pivot to become boutique malls with community centers and social centers. The owners of those malls are searching the world for entertainment options to, hopefully, increase their customer flows and bring some social value back to their dinosaurs. That’s only one profitable use case.
Imagine a group of people being able to join in a “conversation” with one (or more) of the characters in the array. Each of those characters being relatively independent of each other and yet able to refer queries to each other as appropriate. One that I was looking to develop was “Evening at the Improv,” complete with a rowdy audience and potential for human interaction. Another is the “Greeting Wall” in most children’s museums around the world. I installed a simple system in my granddaughter’s elementary school and the kids wiped my MS hobbyist account in a couple days, just triggering the simplest of things repeatedly to see the video response. That caused me to get into Mycroft/Picroft and play with that some. All that said, I’m not married to anyone or anything for any of it. So I may only have a couple Pi’s in my drawer these days but there are a couple dozen micro-SDs in there with them (don’t even think about the updates I’m behind - each has something of value from where I stopped that development stream).
All that said, I’m now only working with a proximity detector triggering a specific set of playback functions in a very small form device. Light-powered with battery backup, wifi connected, programmable (and custom message recordable) via phone app. Proof-of-concept runs on a single Pi with a proximity detector and audio speaker. Hopefully I’ll get finished with this before my next life on this world…