A personal dictionary fixes the words speech recognition keeps getting wrong: colleagues' names, product names, places and jargon. In Whisperer, open Dictionary in Workspace, type what the app hears and what it should write, and every future dictation and meeting transcript uses the right spelling. Whisperer also learns words from the corrections you make, so the list grows on its own.
Speech models are trained on general speech. They've never heard of your company's product or your new colleague's surname.
The problem#
Everyday English works fine. Specific names don't:
Typical mishearings without a dictionary:
- "Priya Shah" becomes "pre a shaw"
- "Maya Levin" becomes "my a leaven"
- "Supabase" becomes "super base"
- "Kubernetes" becomes "Cooper Netties"
- "Okonkwo" becomes "oh con co"
These add up. You spend more time fixing names than you saved by dictating. An entry takes a few seconds and fixes the word for good.
Setting up your dictionary#
Dictionary
4867 corrections from 12 packs
Open the Dictionary page
Open Whisperer's Workspace window and choose Dictionary in the sidebar.
Add an entry
Each entry has two parts: the incorrect form (what the app hears) and the correct form (what it should write). You can add a category and a note if you like.
Fix close variations
Leave Also fix close variations on. Then "pre a shaw", "pria shah" and similar mishearings all become "Priya Shah".
Test it
Hold Fn, say a sentence with the word, release and paste with ⌘V.
Don't try to add everything up front. Add words as you notice them going wrong. After a few days you'll have covered most of your working vocabulary.
It learns from your corrections#
When you fix a word, Whisperer can learn it. Learned entries appear in the dictionary marked Learned, next to the ones you added yourself. You can edit or delete them like any other entry.
What to add#
- People. Colleagues, clients, family. Unusual or non-English names benefit most.
- Your company and products. Product names, internal tool names, project code names.
- Places. Street names, small towns, venues.
- Industry terms. Medical, legal or financial vocabulary that general models rarely hear.
- Brands and services you mention often, like Supabase, Figma or HubSpot.
- Acronyms you say as words.
Prompt words#
The Dictionary page has a second tab, Prompt Words: a short list of terms handed to the speech model before it starts, so it expects them. Use prompt words for a handful of names you say constantly. Use dictionary entries for fixing specific mishearings.
Meetings use it too#
With Pro, the same dictionary is applied to meeting transcripts. If "Okonkwo" is in your dictionary, it's spelled right in Tuesday's call transcript and in the action items written from it.
How many entries you need#
Not many. Ten entries catch the worst offenders. Twenty-five or so cover the names and terms most people use every week. Start small and let the learned entries fill in the rest.
FAQ#
Does Whisperer learn new words on its own?#
Yes. When you correct a word, Whisperer can learn it and adds it to the dictionary marked as learned. You can edit or delete learned entries at any time.
Does the dictionary work for meeting transcripts?#
Yes. The same dictionary is applied to dictation and to meeting transcripts, so colleagues' names and product terms come out right in both.
Related: Personal dictionary feature, Getting started. See pricing and all dictation features.