My Words · included with Pro
Say it once. Dicta gets it right from then on.
Names, product terms, acronyms, the words spell-check has never heard of: add them to My Words and Dicta corrects sound-alike misrecognitions toward your list, on every dictation, without you repeating yourself.
My Words is part of Pro, and it ships at launch alongside the rest of the app, not later.
Before and after
One word on your list, fixed every time
The main speech model hears a sound, not a spelling. Left to itself it writes the everyday word that sounds closest. Put your term on the list and Dicta corrects that guess back to the word you actually meant.
"Can you send the air loom invoice to accounting."
"Can you send the Aerlume invoice to accounting."
Your vocabulary screen
A plain list, and a single dial
My Words is exactly as simple as it looks: a list of up to 512 terms you type or paste in, and one correction-strength dial. There are no per-word settings, no pronunciation spellings to guess at, and no phonetic hints to author. You give Dicta the correct spelling; it handles the sounds.
- Paste a whole glossary at once; Dicta dedupes it case-insensitively and reports what it kept.
- Three-character minimum keeps stray one and two letter fragments out of the booster.
- The strength dial is the only knob, defaulted to balanced so most people never touch it.
How it works
A second pass, tuned to your list only
The main model transcribes first. Then a second, smaller pass rescores what it heard against your list, correcting only the words close enough to clear the gate you set. That is the whole mechanism, and it runs on-device in the same breath as the transcription.
Built for precision, not just recall
A second, smaller model rescoring pass checks the transcript against your list after the main model is done. A tunable similarity gate keeps it precise: real near-misses get fixed, and words that merely rhyme with one of your terms are left alone.
Your list, your rules
Up to 512 terms, three characters minimum. Add them one at a time, or paste a whole list at once and Dicta dedupes case-insensitively, drops anything too short, and tells you exactly how many it kept.
Say it once, and it sticks
No retraining, no overnight rebuild, no model to redownload. Add a term and it applies starting with your very next dictation, in every app where you dictate, that same second.
Tune how aggressive it is
One slider runs from off to eager, six positions in all, balanced in the middle. Cautious only fixes the closest sound-alikes; eager catches more but risks the occasional wrong swap. You decide where that trade-off sits, and nothing else needs configuring.
Where it runs
First in line, before anything else touches your words
My Words runs right after Parakeet decodes your audio, before filler-word removal and before your replacement rules, so the rest of the pipeline is already working from the corrected term. Nothing about this needs a network connection, and nothing about it depends on which writing style, if any, ends up applied afterward.
- Step 1
Transcribe
Parakeet decodes your audio to raw text on-device.
- Step 2
My Words
The booster rescores that text toward your list first.
- Step 3
Filler removal
Ums, uhs and false starts drop out next.
- Step 4
Your rules & style
Replacement rules and the writing style land last.
Local, always
Your list stays on your Mac
My Words lives on the device you typed it into, like everything else Dicta does. It is never uploaded to build a profile of you, and the boosting pass that uses it runs on the same on-device model as the rest of the pipeline.
Join the waitlist
Dicta is in preview. Leave your e-mail to join the waitlist. We only ever e-mail you about the launch, and you can leave anytime.
We store only your e-mail address and the opt-in timestamp. No marketing automation, no third parties.