Fix a word it gets wrong
The handful of words every recogniser mangles, corrected once.
Every recogniser mangles the same few words for a given person, and no amount of speaking clearly fixes it. Usually they are names: your project, your colleagues, the library nobody outside your team has heard of.
- 1
Open the word list from the menu bar
One rule per line, with an example already in it so the shape is obvious before you type anything.
- 2
Write what it heard, and what you meant
Two columns. The left is matched against the transcript without regard to case; the right is what replaces it, exactly as you typed it.
- 3
Press Done
It saves and closes. The next thing you dictate already has the rule applied.
Corrections are applied after recognition rather than inside it. That is the whole reason a rule that makes things worse is something you can see happening and undo, rather than a model that has quietly learned something wrong.
When it goes wrong
A rule made things worse
A short left-hand side matches inside longer words.
Corrections happen after recognition rather than inside it, so a bad rule is visible on screen and can be deleted. Prefer the longest phrase that is actually wrong over the shortest one that would match it.
Two windows disagreed about my rules
Both had the list open and one saved over the other.
A window with unsaved work keeps it. One without picks up whatever the other saved the next time you look at it.
A rule with a hash in the replacement got truncated
This was a real bug, and it is fixed. A hash inside a replacement is not a comment.
Update to the current build. C sharp survives being written as C sharp now.