Dictated numbers now arrive as numbers. "four point eight" becomes 4.8,
"twenty percent" becomes 20%, "five dollars and fifty cents" becomes $5.50,
"the twenty first" becomes the 21st, and "five thirty pm" becomes 5:30 PM.
Spoken years and digit strings a model did not format are deliberately left
alone rather than guessed at.
Text now comes out identical to every other Lven app. The engines disagreed on
non-ASCII input — a CRLF line break, an accented letter written two different
ways, a non-breaking space — so the same sentence could differ depending on
where you dictated it.
An accented word preceded by its own first letter no longer loses that letter.
Cleanup of a finished transcript is roughly three times faster on ordinary
sentences.