Custom vocabulary

What personal vocabulary fixes

General transcription is very good at ordinary language and predictably bad at the words that are specific to your life: your colleagues' names, your company's internal acronyms, the library you have been debugging all week.

Personal vocabulary is a list of the terms you care about, written exactly the way you want them spelled. VoiceSnap Pro gives the list to the transcription model with every dictation and voice note, so when it hears something that sounds like one of your terms, it prefers your spelling. You add a word once instead of correcting it every day.

Adding terms

Open Settings → Personal vocabulary, write one word or phrase per line with your preferred spelling and capitalisation — PostgreSQL, not postgresql — and press Save vocabulary.

• Up to 100 words or phrases.
• Up to 80 characters each, and 2,000 characters for the whole list.
• Shared across every computer signed in to your account. An edit made offline is kept on that computer and saved to your account when it is back online.

Names

Names are the highest-value entries, because getting one wrong in an email is more embarrassing than any other transcription error. Add the people you write to and about: colleagues, clients, managers, the team you keep tagging.

A name spelled differently from how it sounds — Siobhán, Xiuying, Ng — is exactly what the list is for: the model has your spelling in front of it when it hears the sound. It is a strong hint rather than a rule, so an unusual name can still come out phonetically now and then.

Acronyms and initialisms

Write acronyms the way you want them to appear: GDPR, SaaS, OAuth, iOS, gRPC, PagerDuty. Mixed capitalisation is exactly what a model cannot guess, and the list is where it learns yours.

If your house style writes U.S. rather than US, add the dotted form.

Jargon, product names and code identifiers

Everything you say a hundred times a week and always have to fix belongs here: Kubernetes, Tailwind, Postgres, webhook, idempotent, the internal codename nobody outside your company has ever heard.

For code, add the identifiers you say often that are two English words with no space — useEffect, getUserById — and spell them the way your codebase does.

Building the list without sitting down to write it

Do not try to fill the list in one session. Add a term the moment it annoys you, and once a week skim your dictation history for the same mistake appearing twice — those are the entries worth adding. With a hundred places, spend them on the words you actually say.

How and when it is applied

Your vocabulary is one of your account's settings, so every computer you sign in on has it. With each dictation or voice note it goes, with the audio, through our server to OpenAI as spelling context. It is not used to train models, and it is never treated as an instruction to rewrite what you said: it only helps the model choose between spellings of what it heard.

It applies in every language you dictate in, which is what you want for names and product names: Kubernetes is Kubernetes in a German sentence too.