Skip to main content
The Custom Dictionary feature helps OpenWhispr correctly transcribe words that are commonly misrecognized, such as uncommon names, technical jargon, brand names, and domain-specific terminology.

​
What Is the Custom Dictionary?

The Custom Dictionary is a list of words and phrases that you provide to the speech recognition model as context hints. When you add words to your dictionary:
  1. Words are sent as context to the transcription model (Whisper or cloud API)
  2. Model is more likely to recognize these specific words correctly
  3. Accuracy improves for specialized vocabulary in your dictations
  4. Works with all transcription modes (Cloud, BYOK, and Local)
Your agent name is automatically added to the dictionary and cannot be removed to ensure it’s always recognized during AI commands.

​
Why Use a Custom Dictionary?

Speech recognition models are trained on general language but may struggle with:

Uncommon Names

People: “Xanthe”, “Sergey”, “Niamh” Places: “Reykjavik”, “Bruges”, “Kyoto”

Technical Jargon

Software: “Kubernetes”, “PostgreSQL”, “OAuth”, “gRPC” Medical: “hypertension”, “electrocardiogram”, “acetaminophen”

Brand Names

Companies: “OpenWhispr”, “Anthropic”, “Supabase” Products: “ChatGPT”, “Cloudflare”, “Vercel”

Domain Terms

Legal: “amortization”, “indemnification”, “litigant” Scientific: “polymerase”, “mitochondrial”, “chromatography”

​
How It Works

​
Technical Implementation

When you dictate with a custom dictionary:
1

Words Combined

All dictionary words are joined into a single string
2

Sent as Prompt

This string is passed as the prompt parameter to the Whisper API or whisper.cpp
3

Context Provided

The model uses these words as context hints during transcription
4

Better Recognition

Words in the prompt are more likely to be recognized correctly in your speech
The Whisper prompt parameter provides context hints to the model, making it bias toward recognizing these specific words when they appear in your audio.

​
Adding Words to Your Dictionary

​
From Settings

1

Open Settings

Navigate to Settings → Dictionary in the Control Panel
2

Enter Word or Phrase

Type a word in the input field (e.g., “Kubernetes”)
3

Add to Dictionary

Press Enter or click the arrow button to add
4

Add Multiple at Once

Separate words with commas: “OpenWhispr, PostgreSQL, Vercel”

​
Auto-Learn Feature

OpenWhispr can automatically learn corrections you make:
When enabled (default), OpenWhispr detects when you manually correct transcribed text and automatically adds the corrected words to your dictionary.
How it works:
  1. OpenWhispr transcribes “cube or net ease” (wrong)
  2. You manually type “Kubernetes” (correct)
  3. OpenWhispr detects the correction
  4. “Kubernetes” is added to your dictionary
  5. A toast notification appears with Undo option
If OpenWhispr learns a word you didn’t intend, click Undo in the toast notification that appears after auto-learning.

​
Managing Your Dictionary

​
Viewing All Words

In Settings → Dictionary, you’ll see:
  • All words in your dictionary displayed as tags
  • Word count in the header
  • Your agent name (marked as auto-managed, cannot be removed)

​
Removing Words

1

Find the Word

Scroll through your dictionary to find the word you want to remove
2

Hover to Reveal X

Hover over the word tag to reveal the remove button
3

Click X to Remove

Click the small X icon to remove the word
You cannot remove your agent name — it’s automatically managed to ensure AI commands work correctly.

​
Clearing All Words

To start fresh:
  1. Click Clear All in the top-right of the dictionary view
  2. Confirm the action in the dialog
  3. All words except your agent name will be removed

​
Examples of Words to Add

PostgreSQL
Kubernetes
OAuth
gRPC
TypeScript
Supabase
Vercel
Cloudflare
JWT
WebSocket
Redis
Elasticsearch
MongoDB
GraphQL

​
Best Practices

Start Small

Add 10-20 words you use frequently rather than hundreds of obscure terms

Use Exact Spelling

Enter words exactly as you want them transcribed (including capitalization)

Add as Needed

When you notice a word being misrecognized, add it immediately

Review Periodically

Remove words you no longer use to keep the dictionary focused

​
What to Add

✅ Do add:
  • Words you use frequently in dictation
  • Words that are consistently misrecognized
  • Proper nouns (names, places, brands)
  • Technical terms in your field
  • Acronyms and abbreviations
❌ Don’t add:
  • Common English words (the model already knows these)
  • Phrases longer than 3-4 words (use individual words instead)
  • Hundreds of words you rarely use (clutters the prompt)
  • Misspellings or alternate spellings (use the correct form)

​
How Dictionary Works with Different Modes

Dictionary words are sent with your audio to the cloud transcription API as prompt context.Limitations: None — works seamlessly
If you add too many words (200+), some providers may truncate the prompt. Keep your dictionary focused on frequently used terms.

​
Troubleshooting

Possible causes:
  1. Word pronunciation is too similar to another word
  2. Background noise interfering with recognition
  3. Word is very uncommon even with dictionary hint
Solutions:
  • Speak more clearly and slowly
  • Reduce background noise
  • Try adding related words (e.g., add “Kubernetes cluster” not just “Kubernetes”)
  • Consider editing text manually for very rare terms
NVIDIA Parakeet (sherpa-onnx) does not support the prompt/context feature that Whisper uses. Dictionary words are not used with Parakeet.Solution: Switch to Whisper models for dictionary support
This is by design — your agent name must remain in the dictionary to ensure AI commands are recognized correctly.Solution: To change your agent name, go to Settings → Intelligence → Agent Name
If auto-learn detects corrections you didn’t intend:
  1. Click Undo in the toast notification immediately
  2. Manually remove the word from Settings → Dictionary
  3. Consider disabling auto-learn in Settings → Dictionary (if this setting exists)

​
Privacy & Storage

Your custom dictionary is stored locally in your browser’s localStorage and never sent to OpenWhispr servers.
  • Storage location: Browser localStorage under key customDictionary
  • Sync: Not synced across devices (local to each installation)
  • Backup: Export settings to back up your dictionary (future feature)
  • Privacy: Words are only sent to transcription APIs as context, not logged or stored by providers

​
Advanced: Dictionary Storage Format

For developers and advanced users, the dictionary is stored as a JSON array: 
["OpenWhispr", "Kubernetes", "PostgreSQL", "Anthropic", "YourAgentName"]
You can manually edit this in your browser’s developer console: 
localStorage.getItem('customDictionary')
localStorage.setItem('customDictionary', '["word1", "word2"]')
Manual editing is for advanced users only. Always back up your settings first.

​
Limits and Considerations

AspectLimitNotes
Max words~200 recommendedAPI prompt limits vary by provider
Max characters per wordNo hard limitKeep words/phrases concise
StoragelocalStorage~5MB total browser storage
SyncNot availableLocal to each device
Auto-learnEnabled by defaultCan be toggled in settings

​
Next Steps

AI Processing

Use your agent name in commands for AI-enhanced dictation

Transcription Modes

Choose the best mode for your dictionary use case

Notes System

Apply dictionary improvements to your voice notes

Basic Dictation

Master the fundamentals of OpenWhispr dictation

Build docs developers (and LLMs) love

Get started for freeTalk to us