Postprocessing
Replacement rules
Whisper transcribes what you say. If you say “new line,” Whisper writes “new line.” That’s usually not what you wanted. Replacement rules turn spoken phrases into the symbols and characters you actually meant.
The defaults
TongueType ships with sensible defaults, all on by default:
- “new line” → line break
- “new paragraph” → two line breaks
- “open parenthesis” / “close parenthesis” →
(/) - “open bracket” / “close bracket” →
[/] - “open curly” / “close curly” →
{/} - “open quote” / “close quote” → curly “ and ”
- “dollar sign” →
$, “percent sign” →% - “at sign” / “at symbol” →
@ - “hash sign” / “pound sign” →
#
Toggle individual rules on or off with the checkbox, edit a rule’s phrase or replacement in place, or click the minus button to remove it. Reset to Defaults brings the original list back if you ever paint yourself into a corner.
Adding your own
Click Add Rule, type a phrase on the left and the replacement on the right. A few things people typically add:
- Their email signature: “sign off” →
—
Cory - Code symbols: “arrow function” →
=>, “triple equals” →=== - Common phrases: “dot com” →
.com
Special characters: \n and \t
Two escape sequences are recognized in the replacement field:
\n: a line break\t: a tab
Combine them however you want. The default “new paragraph” rule uses \n\n for a blank line between paragraphs, for example. After a line break, TongueType automatically capitalizes the next sentence for you.
How matching works
Rules are case-insensitive and match whole words. Longer phrases win, so “new paragraph” always beats “new line”. Surrounding spaces and stray punctuation get cleaned up automatically, so you don’t need to worry about Whisper occasionally adding a comma after “parenthesis.”
Cancel phrases
If you say one of these phrases at the end of a recording, TongueType throws the entire transcription out instead of inserting it:
- “scratch that”
- “cancel that”
- “discard that”
Edit the comma-separated list in Settings to add your own (“forget that,” “never mind,” whatever fits how you actually talk). Or turn the feature off entirely if it ever fires when you didn’t mean it to.
Filtering Whisper annotations
Whisper sometimes writes things like “[music]”, “(laughter)”, “*coughing*”, or “♪ upbeat music ♪” when it hears non-speech audio. Useful in some contexts, an irritation in most.
Turn on Filter non-speech annotations in Settings to strip them out before the text reaches your cursor. Off by default so we don’t accidentally eat anything you actually dictated.
Turning postprocessing off
The two big switches at the top of the tab let you disable everything if you’d rather get Whisper’s raw output:
- Replace spoken phrases with symbols: off means rules don’t run.
- Filter non-speech annotations: off means “[music]” and friends stay in.
- Cancel transcription with a keyword: off means “scratch that” just gets transcribed like any other phrase.
Where to next
- Dictation: the hold-to-record loop where these rules apply.
- Transcribing files: postprocessing runs on file transcripts too.