Dictation glossary

Custom vocabulary

Custom vocabulary means a user-supplied list of words, names, and spellings that a speech-to-text system should prefer when it hears something similar. The recognizer still listens to audio; the list biases decoding toward those tokens instead of a common word that sounds alike. It is how you keep a coworker named Siobhan, a library called psycopg, or a product codename from being rewritten as something generic.

Get WhisperJot Try the browser demo
In more detail

What is Custom vocabulary?

Speech models are trained on broad language. They are good at 'meeting' and bad at your internal package name. Custom vocabulary is a correction layer: sometimes a bias during decoding, sometimes a replace-after-the-fact pass. Either way, you are teaching the product your lexicon, not retraining the whole model. Homophones remain hard. A short unique token works better than a common word you hope to steal from English. Teams often share one list so everyone's dictation spells the same customers and APIs.

Custom vocabulary is not a language pack and not a full grammar. It will not make an English model fluent in a language it never learned. It also is not magic against noise: if the audio is garbage, no hint list saves you. Used well, it drops the frustrating class of errors that make dictation feel unprofessional — the ones you would never type. Measure it with the errors you actually correct, not with a public word error rate on clean read speech.

For writers who speak

Why it matters for dictation

Dictation fails first on the words that make your job specific. If every proper noun needs a fix, speaking is slower than typing. A vocabulary list is the cheapest accuracy upgrade you can apply without changing engines. It is also why a generic voice-typing panel feels fine for groceries and poor for work.

In this product

How WhisperJot handles it

WhisperJot includes custom vocabulary so you can teach names, jargon, and identifiers; correction keeps those spellings intact after recognition. Processing still runs on-device by default with the local engines; Jot Cloud is opt-in and does not remove the vocabulary step.

Related terms

Keep reading in the glossary.

Text cleanup

Text cleanup means editing a raw speech-to-text transcript so it looks like written language: punctuation, capitalization, filler-word removal, and user-defined replacements.

Read Text cleanup →

Filler-word removal

Filler-word removal means deleting spoken hedges — um, uh, like, you know — from a transcript so the remaining text reads more like writing than like a raw recording.

Read Filler-word removal →

Word error rate (WER)

Word error rate (WER) is a score of how many mistakes a speech recognizer made compared with a reference transcript.

Read WER →

Voice dictation

Voice dictation is a way of writing in which you speak and software turns those words into editable text in an application.

Read Voice dictation →
Questions

Straight answers.

See also: Use cases, Pricing.

What is custom vocabulary in speech-to-text?

Custom vocabulary is a list of terms you add so the recognizer prefers your spellings of names, brands, and jargon. The model still hears audio; the list steers it away from a common word that sounds similar. It is a hint, not a new language. You use it when out-of-the-box dictation keeps mangling the words you would never mistype by hand.

Does custom vocabulary retrain the speech model?

Usually no. Most dictation apps apply a bias or a text replacement without updating model weights. That is why a new name can start working after you save it, with no hour-long training job. A few specialist systems do adaptation, but consumer dictation treats vocabulary as a lightweight lexicon. If a vendor claims otherwise, they should say what is actually trained and where that job runs.

Will custom vocabulary fix every name?

No. Very short words, names that collide with common English, and audio that is buried in noise still fail. Add the exact spelling you want, keep the list to terms you actually say, and pair it with a quiet microphone. Think of it as closing the gap on domain terms, not as a substitute for a better engine or a better recording.

One hotkey, any focused app.

Private local transcription by default, with an optional opt-in cloud engine.

Get WhisperJot — $99/year