Edit: False alarm. Speed is back to being as erratic as ever :(
Hehe, had my hopes up there for a second.
Personally I definitely *have* enjoyed a steady increase in my graphs over the two weeks I have been using my program now. :) Getting 80+wpm average on random texts which was nigh impossible before, and I've even scored a 104wpm! Metamorphosis is the easiest text and my average has gone from ~77 to ~87wpm on it. Kinda trying to type Portrait of the Artist as a Young Man in order and over 350 typed excerpts (yes, I type a lot) my average has gone from ~72 to ~82.
Unrelated, I have been trying to plug Colemak in the wiki pages. Someone added the project to reddit the other day and I got 10k visitors and 3000 downloads, hopefully I can help raise Colemak awareness. :) The colemak.com page seems to have been edited recently though and a lot of links have been removed from the sidebar (and some of the ones left link to empty pages)...
Occasionally there are some non-printable characters in the text that show up as [square]. When such a character occurs, it is nigh impossible to proceed with that piece of text. It might be prudent to filter() the text prior to it getting to the user.
What kind of non-printables? Is the imported text in UTF-8? As in, are they control characters in the text or is it a font issue?
"Weeding out" stuff is harder than just filtering, because if someone decides to practice Navaho or Mongolian, that's their business and I can't really distinguish those characters from "noise". Of course that will probably break the sentence detection, but still... Likewise I try to be careful imposing too much formatting etc. on the texts as I can't really say for sure that the users don't want it there.
If you are not typing in-order you can filter them out by disabling/deleting them via the regex. See the wiki Questions page. Other than that I just recommend filtering and improving the source texts and re-importing.
In the future I might split text importing into two distinct parts: one for importing verbatim lessons, and one for importing more specifically "texts" a la Gutenberg with options to filter non-printable ascii and impose formatting removing double-spaces, etc... I dunno, it's just an idea at this point.