Skip to content

Dictate into Documents

Goal

Create a profile optimized for drafting in note apps, docs, and editors where you want quick capture, clean structure, and predictable formatting.

Why This Pattern Exists

OpenScreech processing is intentionally layered:

  1. ASR captures what you said.
  2. Word replacement applies deterministic cleanup.
  3. Optional LLM applies style/structure transformations.
  4. Paste behavior controls how the receiving app interprets content.

For writing, this separation matters because deterministic replacements keep domain terms stable, while optional LLM can improve readability without losing consistency.

Preset: Writer Default (Balanced)

Use this when you want fast drafting with light cleanup and no over-automation.

Setting Where Value Why
Profile name Profiles -> Profile Writer - Balanced Keep workflow-specific naming so hotkeys remain obvious.
Assigned hotkey Manage Hotkeys -> Profile Writer - Balanced Makes this profile the starting behavior for that hotkey.
Auto paste into active window Profiles -> Profile On Minimizes friction during drafting.
Add transcriptions to temporary history Profiles -> Profile On Lets you recover recent snippets quickly.
Word Replacement Enabled Profiles -> Word Replacements On Applies reliable pre-LLM cleanup.
Word Replacement Library Profiles -> Word Replacements Writer Terms Keeps writing vocabulary stable.
LLM Enabled Profiles -> LLM On (optional) Enables style pass when you want cleaner prose.
LLM processor stage Profiles -> LLM A dedicated writer processor References reusable provider and prompt settings.
LLM API Provider LLM Post-Processing Library -> processor Ollama or OpenAI Compatible Match your local/private or hosted stack.
Timeout (s) LLM Post-Processing Library -> processor 45 Better tolerance for longer post-processing prompts.
Paste input format Profiles -> Paste Markdown (if LLM outputs markdown), else Plain Text Controls how conversion to destination format is interpreted.
Paste format Profiles -> Paste Auto (Multi-format) Good compatibility across modern editors.
Saved Outputs Enabled Profiles -> Saved Outputs Off (default) Keep writer flow lightweight unless audit trail is needed.

Preset: Writer Fast (No LLM)

Use this for low latency and maximum predictability.

Setting Where Value Why
Profile name Profiles -> Profile Writer - Fast Explicitly marks no-LLM profile.
LLM Enabled Profiles -> LLM Off Reduces delay and external dependency risk.
Word Replacement Enabled Profiles -> Word Replacements On Maintains terminology correction without model calls.
Paste input format Profiles -> Paste Plain Text LLM-off path is plain-text by design.
Paste format Profiles -> Paste Rich Text or Plain Text Choose based on destination app behavior.

Word Replacement Starter Library (Writer Terms)

Create a library named Writer Terms and add rules like:

Source Replacement Case Sensitive Notes
open screech OpenScreech Off Brand normalization.
api API Off Acronym normalization.
llm LLM Off Acronym normalization.
new paragraph \n\n Off Spoken structure helper.
bullet point - Off Rapid list drafting.

If LLM is enabled, keep prompt constraints explicit:

  • Preserve meaning.
  • Improve readability.
  • Keep technical terms unchanged.
  • Avoid adding facts.

Example prompt text:

You are editing dictated draft text.
Preserve intent and all technical terms exactly.
Improve punctuation, paragraphing, and clarity.
Do not add new facts or remove important details.
Return only the edited text.

Setup Sequence (5-10 Minutes)

  1. Open Profiles -> Add -> rename profile to Writer - Balanced.
  2. Enable Word Replacements, assign Writer Terms library.
  3. Decide whether LLM is on or off for this profile.
  4. Set paste options (Paste input format, Paste format).
  5. Ensure Auto paste into active window is enabled.
  6. Map an enabled hotkey to Writer - Balanced in Manage Hotkeys....
  7. Dictate a short paragraph and verify paste result in your target app.

Validation Checklist

  • Dictation starts/stops reliably via assigned hotkey.
  • Terms like OpenScreech and LLM are normalized correctly.
  • Output arrives in target app with expected formatting.
  • Temporary history stores recent snippets for recovery.

Tuning Guidance

  • If output is too slow: disable LLM or lower prompt complexity.
  • If formatting is inconsistent across apps: prefer Paste format = Plain Text first, then try richer formats deliberately.
  • If terms keep drifting: move more normalization into word replacement rules.

Quick Tips

  • Map separate hotkeys to Writer - Fast and Writer - Balanced. You can begin with either and quick-switch while listening if you realize the draft does or does not need an LLM pass.
  • Keep terminology corrections in word replacements and prose shaping in the LLM. This prevents a creative model from becoming the only place a product name is spelled correctly.

See Also

Visual Notes

  • TODO: add screenshot of writer profile settings.
  • TODO: add screenshot of writer word replacement library.
  • TODO: add GIF dictating into a notes app with immediate formatted paste.