Skip to main content
Every field has words a general speech model will not know: your coworkers’ names, your product’s name, the acronyms your team uses daily. The Dictionary is where you teach Rekody yours.

​
How it works

Add a term and Rekody corrects it on every dictation, deterministically and instantly, with no AI model in the loop:
  • Exact matches are recased to your spelling: “rekody” becomes “Rekody”, even mid-sentence and with punctuation attached (“rekody,” becomes “Rekody,”).
  • Near misses are fixed when they are within a letter or two of exactly one term: “recody” becomes “Rekody”.
  • It stays conservative. Everyday English words are never rewritten into your terms, even when the spelling is close (“change” stays “change”), and if a word sits close to two of your terms, Rekody leaves it alone rather than guess.
Multi-word terms are fine: “Core ML” is one dictionary entry and is matched as a whole phrase. When AI cleanup is enabled, your terms are also listed in the cleanup prompt so the model preserves them, but the deterministic pass runs either way and works with cleanup off. Good candidates:
  • People and place names that get misheard.
  • Product and company names, especially lowercase or invented ones.
  • Technical jargon and acronyms.

​
What it fixes today, and what it can’t yet

The corrector works on the final text of every engine: Rekody Streaming, local Whisper, Groq, and Deepgram all get the same pass before your text lands. Honest limits, so you know what to expect:
  • Near-misses, yes. Different words, not yet. “recody” becomes “Rekody”, but if the engine heard a different real word entirely (“recording” when you said “rekody”), the corrector leaves it alone on purpose: rewriting real words would corrupt ordinary sentences.
  • Streaming shows raw partials. With Rekody Streaming, the live words in the pill are the engine’s raw output; corrections apply to the text that actually lands at your cursor.
The permanent fix for the hard mishears is to bias the engines themselves with your terms so they hear you correctly in the first place. That work is in progress per engine: local Whisper (#88), Deepgram (#89), Groq (#90), and Rekody Streaming (#91), where it is hardest and will land last.

​
Dictionary or correction?

  • Fixing a one-off mistake? Correct the entry in the Library.
  • Fixing the same word for the third time? Put it in the Dictionary and stop fixing it.
The Dictionary lives on your Mac like everything else in Rekody. It is not uploaded anywhere.

Related topics

SnippetsArchitectureFAQRekodyLibrary