So they're finally switching from Google STT to Mozilla Deepspeech, but that project includes no pre-learned models IIRC. So where is MyCroft getting its data from? Or is Mozilla releasing models in 2018?