Think aloud — write faster
Voice is 2-3× faster than typing. An article draft is born at the pace of thought, not fingers.
✍️ Use case
Voice typing for writers and journalists
Article drafts, transcription of your own thoughts, notes. Punctuation and capitalization automatic. Custom words you add by voice.
Sound familiar?
- Ideas slip away faster than you can type them
- Transcribing an interview takes an hour for 15 minutes of audio
- Win+H doesn't know rare words, names, terms
- A $15/mo Wispr Flow subscription is not an option
What changes with AuroraWhisp
Punctuation handled
Periods, commas, dashes — the app places them automatically based on intonation. Say "new paragraph" by voice.
Names and terms — trainable
Record a tricky word once by voice — the app remembers every variant of how you say it.
Dictation history
Every recognized text is saved locally. Search, export, copy. Easy to come back to a draft a week later.
One writer's story
I write novels. A 3,000-word scene by hand — two hours of focused work and aching wrists. By voice — 35-40 minutes and intact hands. At first I worried voice would not work for "creative". Turned out the opposite: you speak in scenes, not words. A room description, a dialogue between two characters, an internal monologue — all flow more naturally spoken than typed. You still edit by hand later — but the draft is there, and that is the hard part.
A writer's day with AuroraWhisp
Morning (45 min): dictate ideas and scene fragments into Obsidian / Scrivener — by the time the coffee brews you have 1,500 words of raw material. Daytime (1-2 hours in Word): main writing, alternating voice and hands — voice for long descriptions and dialogue, hands for quick edits and formatting. Evening (as needed): drag an mp3 of an interview or your own audio note onto transcribe.bat — a minute later you have a ready txt file. Transcribing an hour of audio takes two minutes.
Which models are best for writing
For an English draft when speed matters and editing comes later: Sherpa Zipformer English — ~150 ms per phrase, very strong on conversational English. For final polish or accented interviews: Whisper Large-v3 on an NVIDIA GPU — 100-300 ms, best handling of specialised vocabulary and unusual names. Whisper Medium / Distil-Large-v3 — a middle ground if you have a GPU but Large is too slow. Russian and other languages — see the security page for the full list.
Voice punctuation: what to say
The default set works out of the box and is intuitive: "comma" → ",", "period" / "full stop" → ".", "question mark" → "?", "exclamation" → "!", "colon" → ":", "semicolon" → ";", "dash" → "—", "new line" → line break, "new paragraph" → blank line + indent, "open paren" / "close paren" → "(" / ")", "open quotes" / "close quotes" → smart quotes. Capital letter at the start of a sentence — automatic. Custom rules in Settings → Voice Replacement.
Especially good for long interviews, daily journaling, book drafts.