Skip to content

Formatting Styles

shh can format your transcriptions using AI to match different contexts. Choose a style based on how you plan to use the transcription.

Available Styles

Neutral

Best for: Accuracy, raw data, archival

Neutral style returns the transcription exactly as Whisper outputs it, with no AI formatting applied. This is the fastest option and preserves the original phrasing.

When to use:

  • You need verbatim transcriptions
  • You'll edit the text yourself
  • You want to minimize API costs (no GPT call)
  • Accuracy matters more than readability

Example:

shh --style neutral

Input (spoken):

Um, so like, I think we should probably update the database schema, you know, to improve performance and stuff.

Output:

Um, so like, I think we should probably update the database schema, you know, to improve performance and stuff.

Casual

Best for: Quick notes, personal use, conversational context

Casual style removes filler words and smooths out the transcription while maintaining a conversational tone. It's like how you'd text a colleague.

When to use:

  • Taking quick notes or voice memos
  • Internal team communication
  • Personal documentation
  • You want readability without formality

Example:

shh --style casual

Input (spoken):

Um, so like, I think we should probably update the database schema, you know, to improve performance and stuff.

Output:

I think we should update the database schema to improve performance.

What it does:

  • Removes filler words (um, uh, like, you know)
  • Fixes minor grammatical issues
  • Maintains conversational tone
  • Keeps it concise but friendly

Business

Best for: Professional documentation, formal communication, presentations

Business style transforms your transcription into polished, professional text suitable for formal contexts. It adds structure and clarity.

When to use:

  • Meeting minutes or reports
  • Documentation for stakeholders
  • Professional emails or messages
  • Presentations or proposals

Example:

shh --style business

Input (spoken):

Um, so like, I think we should probably update the database schema, you know, to improve performance and stuff.

Output:

I recommend updating the database schema to improve performance and data integrity. This will provide measurable benefits to system efficiency.

What it does:

  • Removes all filler words
  • Restructures sentences for clarity
  • Uses professional vocabulary
  • Adds context and structure
  • May expand on ideas for completeness

Setting a Default Style

If you always use the same style, set it as your default:

# Set casual as default
shh config set default_style casual

# Now 'shh' automatically uses casual formatting
shh

Override the default anytime with the --style flag:

# Even with casual as default, use business for this recording
shh --style business

Style Comparison

Feature Neutral Casual Business
Filler words Kept Removed Removed
Tone Verbatim Conversational Professional
Structure Original Light cleanup Polished
Speed Fastest Fast Slower
API calls 1 (Whisper) 2 (Whisper + GPT) 2 (Whisper + GPT)
Cost Lowest Medium Medium

Combined with Translation

Styles work seamlessly with translation:

# Casual French transcription
shh --style casual --translate French

# Business English transcription from French audio
shh --style business --translate English

The formatting is applied after translation, ensuring natural phrasing in the target language.

Technical Details

How It Works

  1. Neutral: Whisper API only → Direct output
  2. Casual/Business: Whisper API → PydanticAI agent → Formatted output

AI Model

Formatting uses OpenAI's gpt-4o-mini model for cost-effective, high-quality results. This is separate from the Whisper API call.

Prompts

The formatting agent receives context about the desired style and applies transformations accordingly. Prompts are optimized for:

  • Preserving meaning and intent
  • Removing only true filler words (not meaningful pauses)
  • Maintaining speaker's voice in casual mode
  • Professional polish in business mode

Privacy

Both Whisper and GPT API calls are processed by OpenAI. If privacy is a concern:

Best Practices

For Voice Memos

Use casual style - it cleans up your thoughts without over-formalizing them.

shh config set default_style casual

For Meetings

Use business style - creates professional documentation ready to share.

shh --style business

For Transcription Archives

Use neutral style - preserves exact phrasing for future reference.

shh --style neutral

Cost Optimization

If you're watching API costs:

  • Use neutral for drafts, then manually edit
  • Use casual/business only for final versions
  • Set neutral as default, override when needed

Next Steps