Dictionary Guide

Conjure's dictionary system improves transcription accuracy and automates text corrections. It has three components: glossary entries, replacement rules, and auto-learn suggestions.

Glossary Entries

Glossary entries are vocabulary hints fed directly to whisper's initial_prompt parameter. When whisper sees these words in the prompt, it's significantly more likely to recognize them correctly in your speech.

When to Use Glossary Entries

Add words that whisper frequently misrecognizes:

• Technical jargon -- "Kubernetes", "PostgreSQL", "Tailwind", "Tauri"
• Product names -- "Conjure", "SubFrame", "GitHub", "VS Code"
• People's names -- "Josiah", "Satya", "Sundar"
• Acronyms -- "CI/CD", "BYOK", "LLM", "TTS"
• Domain-specific terms -- medical, legal, scientific vocabulary

How to Add Glossary Entries

1. Go to Settings > Dictionary (or click Dictionary in the sidebar)
2. Click "Add Entry"
3. Set the type to Glossary
4. Enter the word or phrase exactly as you want it recognized
5. Save

Glossary entries are concatenated and passed to whisper as a prompt prefix, capped at 600 characters to stay within whisper's prompt token limit. The most recently added entries take priority.

Replacement Rules

Replacement rules are find-and-replace patterns applied after transcription but before post-processing. They let you auto-correct recurring mistakes or expand abbreviations.

When to Use Replacement Rules

• Spelling corrections -- whisper consistently writes "conjur" instead of "Conjure"
• Name corrections -- "john" should always be "John" in your context
• Abbreviation expansion -- "btw" becomes "by the way"
• Technical formatting -- "react" should be "React", "javascript" should be "JavaScript"

How to Add Replacement Rules

1. Go to Settings > Dictionary
2. Click "Add Entry"
3. Set the type to Replacement
4. Enter the "From" text (what whisper produces) and the "To" text (what you want)
5. Save

Case sensitivity

Replacement rules use word-boundary matching. While matching is case-insensitive by default, the replacement output preserves the exact casing you specify. "conjure" won't match "Conjure" if case-sensitive mode is enabled.

Replacements are case-insensitive by default and applied using word boundaries to avoid partial matches.

Auto-Learn

Auto-learn is Conjure's intelligent correction tracking system. It observes when you manually correct transcription errors and suggests new dictionary entries when it detects recurring patterns.

How auto-learn works

Conjure tracks each correction pattern (original word -> corrected word) and counts occurrences. When the same correction crosses a configurable threshold, a dictionary entry suggestion appears. You can approve or dismiss each suggestion.

How Auto-Learn Works

1. You dictate text and it's transcribed
2. You notice an error and correct it using the inline word correction feature (click a word in the transcript, type the correction)
3. Conjure records the correction pattern (original word -> corrected word)
4. When the same correction happens multiple times (crossing a configurable threshold), Conjure suggests a new dictionary entry
5. You approve or dismiss the suggestion

Viewing Suggestions

Auto-learn suggestions appear in two places:

• Dictionary page -- a banner at the top shows pending suggestions with approve/dismiss buttons
• Sidebar badge -- a count badge appears on the Dictionary nav item when new suggestions are available

The Feedback Loop

Dictionary entries create a virtuous cycle:

Dictate -> Whisper misrecognizes "Conjure" as "conjur"
  -> You correct it manually
  -> Auto-learn records the pattern
  -> After N corrections, a suggestion appears
  -> You approve it as a Replacement rule
  -> The rule also feeds back as a Glossary hint to whisper
  -> Next dictation: whisper recognizes "Conjure" correctly

This means the more you use Conjure, the more accurate it becomes for your specific vocabulary.

Transcription feedback

Dictionary entries are fed to whisper via the initial_prompt parameter, improving future recognition of your vocabulary. The prompt is capped at 600 characters, so the most recently added entries take priority.

Categories

Organize your dictionary entries into categories (folders) for easier management:

1. Click the folder icon in the Dictionary view
2. Create a new category (e.g., "Medical Terms", "Project Names", "Slang")
3. Assign entries to categories by clicking the category selector on each entry
4. Collapse/expand categories in the grouped view

Categories are stored locally and are purely organizational -- they don't affect how entries are applied.

Dictionary Search and Filtering

The dictionary page provides:

• Search bar -- filter entries by text
• Type filter -- show All, Glossary only, or Replacement rules only
• Grouped view -- see entries organized by category
• Delete confirmation -- prevents accidental deletion of entries

Tips

• Start with glossary entries for your most-used technical terms -- the accuracy improvement is immediate
• Don't over-fill the glossary. The 600-character limit for whisper's initial prompt means only the most relevant entries should be included
• Use the inline correction feature (click words in transcript history) as your primary way to train auto-learn
• Review auto-learn suggestions regularly -- some may be one-off mistakes rather than recurring patterns
• Replacement rules are applied after voice commands, so you can use them to override voice command behavior if needed