Speech recognition models don't know your tech stack. They've never heard of your internal tools. Say "Kubernetes" and you get "Cooper Netties." Say your project name and you get... something else entirely.
A personal dictionary fixes this. Add the word once, and it gets recognized correctly from then on.
Related: Personal Dictionary Feature, Code Mode, Getting Started Guide. See pricing and all features.
Ready to try voice dictation on your Mac?
Free download. No account required. 100% offline.
Download on the Mac App StoreThe Problem#
Whisper models are trained on general speech. Everyday English works fine. Tech vocabulary does not:
Common misrecognitions without a personal dictionary:
- "Kubernetes" becomes "Cooper Netties" or "Kuber nettys"
- "useState" becomes "use state" or "you state"
- "PostgreSQL" becomes "post gress Q L" or "post rescue"
- "Tailwind" becomes "tail wind" or "tail wined"
- "Vercel" becomes "versaille" or "verse cell"
- "Supabase" becomes "super base" or "soup a base"
These errors add up. You'll spend more time fixing misrecognitions than you saved by dictating. A dictionary entry takes five seconds and solves the problem permanently.
Setting Up Your Dictionary#
Here's what the Dictionary interface looks like in Whisperer:
Dictionary
4867 corrections from 12 packs
Open Personal Dictionary
Open Whisperer from the menu bar and go to Settings > Personal Dictionary. This is where you add and manage custom terms.
Add your first term
Click Add and type the term exactly as you want it to appear when recognized. The spelling you enter is the spelling that gets inserted.
Test it
Hold Fn, speak a sentence using the term, and verify it's recognized correctly. If not, check the spelling in your dictionary and try again.
Don't try to add everything upfront. Just add terms when you notice they're being misrecognized. After a few days of use, you'll have covered most of your working vocabulary.
What to Add#
These categories cause the most problems:
Library & Framework Names
React, Next.js, Tailwind, Prisma, Drizzle, Zustand, Zod, tRPC, Vite, Bun, Deno, FastAPI, Django, Flask, Express, NestJS, Svelte, Nuxt, Remix, Astro.
Infrastructure & Services
Kubernetes, Docker, Terraform, Vercel, Supabase, PlanetScale, Neon, Cloudflare, Hetzner, Fly.io, Railway, AWS, GCP, Azure, PostgreSQL, Redis, Kafka.
Project-Specific Terms
Your product name, internal tool names, microservice names, API endpoint names, database table names — anything unique to your codebase that a general model won't know.
Team Member Names
Names of colleagues, clients, and stakeholders you frequently mention in messages, emails, and documentation. Unusual or non-English names especially benefit from dictionary entries.
Acronyms & Abbreviations
CI/CD, RBAC, gRPC, OAuth, JWT, SSR, SSG, ISR, CSR, ORM, SDK, CLI, WASM, HIPAA — any acronym you regularly dictate.
Domain Vocabulary
Industry-specific terms: medical terminology if you're in healthtech, financial terms for fintech, legal terms for legaltech. These are rarely in general speech models.
Before and After#
The difference is obvious:
“Deploy the next JS app to Vercel with Supabase”
Deploy the next JS app to versaille with super base“Deploy the Next.js app to Vercel with Supabase”
Deploy the Next.js app to Vercel with Supabase“Set up Kubernetes with Terraform on Hetzner”
Set up Cooper Netties with terra form on Hetzner“Set up Kubernetes with Terraform on Hetzner”
Set up Kubernetes with Terraform on HetznerWith dictionary entries, technical terms work like regular words.
Pro Tips#
Compound terms: For multi-word terms like "Next.js" or "Fly.io", add the complete term including punctuation. Whisperer will match the spoken form to your written form.
Casing matters in Code Mode. If you add "useState" to your dictionary, it will be recognized with that exact casing when Code Mode is active. For natural language mode, standard sentence casing rules still apply.
Starter List for Web Developers#
Here's a quick-start dictionary you can build from if you work in the JavaScript/TypeScript ecosystem:
React
Next.js
TypeScript
Tailwind
Prisma
Zustand
Zod
tRPC
Vercel
Supabase
PostgreSQL
Redis
Vite
Bun
Deno
ESLint
Prettier
Vitest
Playwright
Storybook
Add these as a starting point and expand as you encounter new misrecognitions in your daily workflow.
How Many Terms Do You Need?#
Not many:
- 10 terms catches the worst offenders
- 25 terms covers your main tech stack and project names
- 50+ terms handles pretty much everything
Most people hit 95% accuracy with 25-30 entries. That's maybe 10 minutes of work total, spread over your first few days of use.
Related: Personal Dictionary Feature, Code Mode, Getting Started Guide. See pricing and all features.
Ready to try voice dictation on your Mac?
Free download. No account required. 100% offline.
Download on the Mac App Store