Mundwerk 1.2 — hands-free dictation and automatic paragraphs

When I shipped Mundwerk 1.0, two things were explicitly on the roadmap and not in the box. Both are in now. Here is what changed across 1.1 and 1.2, and the small mistake I made shipping it.

Hold-to-talk was the wrong default for long text

Mundwerk 1.0 was push-to-talk: hold the hotkey, speak, release. That is perfect for a quick sentence and tiring for a paragraph. So 1.2 adds a toggle mode — press the hotkey once to start recording, press it again to stop. Your hands stay on the keyboard, or off it entirely, for as long as you want to talk. Hold-to-talk is still the default; toggle lives one switch away in Settings → Hotkey.

Pauses became punctuation

Once you can dictate for minutes at a time, the output needs structure. In toggle mode Mundwerk now reads your pauses: a short pause inserts a space, a longer one starts a new paragraph. You set the pause length and whether a paragraph is a line break or a blank line. It turns a wall of text into something you can actually read back, without touching the keyboard mid-thought.

Getting this right was mostly a silence-detection problem. The first cut measured pauses unreliably, so the paragraph breaks landed in the wrong places — that is fixed now, and word onsets right after a pause no longer get clipped.

Pick the dictation language

1.1 added an explicit language picker — German, English or Automatic. German stays the default and is tuned for German dictation with English technical terms sprinkled in; English is there for fully English dictation, like writing prompts for AI tools; Automatic detects per dictation. A useful detail: with German pinned, the smaller models will sometimes translate an English dictation into German — Large v3 Turbo is far more robust if you switch languages a lot.

Shipping fast enough to break something

Mundwerk updates through EdDSA-signed Sparkle deltas, not the App Store, precisely so I can move quickly. The honest downside showed up here: 1.2.0 quietly dropped the last-dictation text and the “Correct” button in toggle mode, because the recording session ended before transcription finished. 1.2.1 a day later put both back. Fast updates cut both ways — the fix shipped the next morning, which is the upside of owning the channel.

Where to go from here

The app and documentation live at mundwerkapp.de, with a short overview on the Mundwerk page here. Still on the list: an English UI and streaming partial transcripts while you speak.

Mundwerk is a one-person project. Feedback goes a long way.