Better way of generating text

[ppetru]

Right now it's just a random selection of words.

What many other Morse trainers use is canned QSO-style templates with placeholders for name / rig / etc. This works but quickly gets repetitive.

We'd ideally have something like a Markov chain (or one of the newfangled char-rnn type text generators) trained on a bunch of both natural text and QSOs. Such a model would be relatively compact yet generate endless amounts of novel text to practice with. The main issue would be getting enough real QSO text for training.