Custom Dictionary & Vocabulary
Teach Speakeasy your industry jargon, rare acronyms, teammate names, and voice snippets. Includes 1-click migration tools for Wispr Flow and Superwhisper switchers.
How Vocabulary Biasing Works
General speech recognition models struggle with niche developer frameworks, medical terminology, and startup names (e.g. "Postgres" transcribed as "post grass", or "Kubernetes" transcribed as "cooper nineties").
Speakeasy uses a lightweight on-device token prior logit bias matrix. When you register a custom word or phrase, the Core ML decoder boosts the probability of matching that exact spelling in under 2 milliseconds without requiring any model retraining or cloud synchronization.
Adding Custom Words & Replacements
Open Preferences → Dictionary or press Cmd + , and select the Dictionary tab.
| Spoken Sound / Alias | Target Output | Category |
|---|---|---|
post gres |
PostgreSQL |
Database / Tech |
cube cuddle |
kubectl |
DevOps CLI |
next js |
Next.js |
Framework |
c sharp |
C# |
Language |
insert my calendar link |
https://cal.com/nicholas/30min |
Voice Snippet |
1-Click Migration from Competitors
If you are switching from a subscription dictation tool, you do not need to rebuild your vocabulary from scratch. Speakeasy imports existing configurations seamlessly:
Migrating from Wispr Flow
In Wispr Flow settings, export your dictionary as wispr-dictionary.json. In Speakeasy, click Import Dictionary → Wispr Flow. All custom words, replacements, and tone preferences are imported instantly.
Migrating from Superwhisper
Speakeasy detects existing Superwhisper configuration files at ~/Library/Application Support/superwhisper/ and offers a 1-click import prompt on initial setup.
Dictionary File Format (JSON / CSV)
Teams and power users can maintain their dictionary as code in a version-controlled repository using standard JSON:
You can symlink ~/.config/dictation-app/dictionary.json to your personal dotfiles repository to keep all your Macs synchronized without cloud accounts or third-party servers.