How many custom vocabulary entries can Amical handle before it slows down?

Amical handles up to several thousand vocabulary entries without measurable latency impact because the recognition layer treats vocabulary as a weighted candidate set, not a full search. Most domain experts plateau around five hundred to twelve hundred entries — the long tail past that point covers terms you use less than once a month, which contribute little to overall accuracy. If you find yourself adding twenty new entries a week and never pruning, your list is accumulating noise. A quarterly review where you delete anything you have not dictated in ninety days keeps the list at a healthy size and improves recognition for the terms that remain.

Can Amical learn vocabulary automatically from my writing?

Amical does not crawl your documents to auto-extract vocabulary because that would require sending document content to a training pipeline, and Amical is local-first by design. The trade-off is that vocabulary curation stays a manual step — you decide which terms to add, with pronunciation hints, and you control the file. The upside is full transparency: nothing about your medical notes, legal work, or proprietary codebase ever leaves your machine in service of improving recognition. For most users, the manual curation step takes about an hour to seed and ten minutes a week to maintain, in exchange for keeping every domain artefact private.

Will my pronunciation hints work if I have a non-US accent?

Yes. Pronunciation hints are best written in the phonetic shape of how you actually say the word, not a textbook pronunciation. An Indian English speaker saying 'doctor' with a clearer 'd' should write the hint to match their accent, not the American 'dakter' shape. A French speaker dictating English medical terms should write hints that reflect their natural pronunciation. The hint is a private mapping for your microphone and your voice; copying someone else's hint set rarely transfers cleanly because their accent shapes are not yours. Test each hint by dictating the term ten times after adding it.

How to teach Amical your technical vocabulary (medical, legal, code)

A workflow for adding drug names, case citations, framework names, and other domain terms to Amical so your dictation stops mangling the words you use most.

Generic dictation engines were not trained on your domain. They handle 'meeting' and 'deadline' fine, then turn 'tamoxifen' into 'tamafone' or 'useEffect' into 'use effect'. The fix is custom vocabulary: a list of domain terms you teach Amical once, after which they transcribe correctly for the rest of time. This guide shows the exact workflow for opening the vocabulary file, adding terms with optional pronunciation hints, grouping them by domain, and sharing a polished vocabulary list with your team so everyone benefits from the work. The medical, legal, and software engineering examples below apply equally to any specialist field — finance, biology, architecture, music — where standard transcription falls apart.

Step by step

1
Open Amical Preferences and find the Custom Vocabulary panel
Launch Amical, click the menu bar icon, and open Preferences. Go to the Custom Vocabulary panel. You will see a list interface where each entry has a term and optionally a pronunciation hint. The vocabulary file lives on your local disk and is editable as plain text in a code editor — useful when pasting in a large list. Before adding anything, dictate a few sentences from your domain into a scratch note to baseline which words actually fail. Do not add words you do not need; a smaller, accurate list outperforms a bloated one.
2
Add domain terms with pronunciation hints for anything spelled oddly
For each failing term, add an entry: the term as you want it transcribed, plus a pronunciation hint that matches how you actually say it. 'Tamoxifen' with hint 'ta-MOX-i-fen'. 'Eloquent' with hint 'EL-o-kwent' if you are referencing the Laravel framework component and want to disambiguate from the adjective. 'useEffect' with hint 'use-eh-fect' so React hooks do not get spaced apart. Pronunciation hints matter more than spelling for low-frequency words because Amical matches audio shape first, then validates against the vocabulary list. A term without a hint still helps, but a hint cuts errors faster.
3
Group entries by domain so language switches stay clean
If you write in more than one domain — medical notes during the day, code comments at night, legal filings on weekends — keep separate vocabulary lists rather than one giant mix. Amical lets you activate a vocabulary list per app or per project. Bind your medical vocabulary to your EHR app, your code vocabulary to Cursor and VS Code, your legal vocabulary to Word and Westlaw. This prevents a drug name from auto-correcting your function name and vice versa. The per-domain split also makes it possible to share each list independently with the right colleagues.
4
Source vocabulary from real artefacts, not memory or generic word lists
Do not try to remember every term you use. Open the last twenty documents you wrote in your domain — patient notes, legal briefs, codebase files, contracts — and pull every uncommon proper noun, acronym, drug, framework, or case name into a text file. Five hundred entries sourced from real recent work will outperform two thousand entries from a generic specialist glossary. Generic glossaries include archaic or unused terms that pollute the recognition space. After two weeks of dictating with the seeded list, prune anything that did not get used; keep the list lean.
5
Export your vocabulary and share it with teammates in the same domain
Amical's custom vocabulary file is a plain text or JSON file on disk — the exact format is shown in Preferences. Export your cleaned list, share it in your team's wiki or Notion, and ask colleagues to import it into their own Amical install. Doctors on the same service, lawyers in the same practice group, engineers on the same codebase all benefit from a shared seed list because they write about the same drugs, cases, and modules every day. Treat the vocabulary file like a team dotfile: version it, review additions in pull requests, prune it every quarter.

Experience the future of dictation

Local and Cloud Models

Choose from powerful cloud-based AI models or run everything locally for maximum privacy and control. Switch seamlessly between providers to find the perfect balance of speed, accuracy, and security.

Custom Vocabulary

Customize the AI to recognize your specific terminology, jargon, and proper nouns for industry-specific accuracy and personalized transcription results.

Project PhoenixMcConaughey

CI/CD pipelinek8s

SOC2PostgreSQL

JiraROI metrics

⌘Space

⌥⌘A

CtrlAltSpace