Speaking instead of typing when drafting
By Oliver Schmalzl · Written: · Archive discovery date:
13 July 2025 · Archive date
Some thoughts are easier to say out loud than to write down neatly straight away. Speech recognition can turn them into text. A usable workflow does not end with the recognised sentence, though: sorting and proofreading come next.
A short text of your own is enough to try it out. No real conversations, other people's voices or confidential content are needed.
Two different tasks
The first task is recognising the spoken words. The second is revising the resulting text. If both happen together, an unnoticed change in meaning can be harder to spot.
That is why a simple test would start literally: "The left tab is five millimetres wide; the right width is still open." The result must keep the open detail. A number added afterwards would be a change in content, not just an improvement in language.
MDN distinguishes between speech recognition and speech synthesis in the Web Speech API. Which processing happens locally depends on the specific setup. An interface that looks local is not proof of this.
Raw text becomes a brief
After dictating, you can add a heading, paragraphs and open questions. An AI can help make the structure visible. It should, however, mark missing details and not add details that merely seem to fit.
For an observation, you can compare a short and a longer sentence of your own. Technical terms, units and sentence boundaries are interesting. The benefit shows in whether the statement becomes clear faster and can be checked reliably. A smoothly recognised text is a good start, but not yet a finished description.
Original finds
These links were the starting point for this post. The commentary in the text is mine; the linked posts are not my own tests.
- Speaking instead of typing when programming
- Meeting notes as Markdown
- Transcribr describes local speech recognition with JSON output, batch processing and a FileMaker connection.
- Personal local voice assistant with a wake word and German answers; check the offline claim.
- Creating several content drafts such as a newsletter, a post and a reel script from one voice note.
About the date: This post was written on 11 October 2026. The archive date corresponds to the earliest related find in the collection, not to a publication at that time. Technical context and sources added later refer to the time of writing.