Saywind

How it works

Hold the Fn (Globe) key and talk. When you let go, your words appear wherever your cursor is, usually in a fraction of a second.

Speech recognition uses NVIDIA Parakeet (the Parakeet Ultra model), running on your Mac’s Neural Engine. After the first-run model download, dictation works offline.

Features

• Hands-free mode: double-tap Fn to start, then tap it again to stop.
• Learns names you correct: if Saywind mishears a name and you fix it in the text it just pasted, it learns it.
• Instant start: the microphone stays ready, muted, between dictations, so your first word isn’t cut off. You can turn it off in Settings.
• Personal dictionary: add words, with “sounds like” spellings.
• Replacements: “when I say X, write Y”.
• British spelling: optional.
• History: searchable, and kept on your Mac.
• Your clipboard is put back after Saywind pastes.
• Esc cancels a dictation.
• Fn shortcuts still work, such as Fn+arrows.

Setting up

1. Download the disk image and drag Saywind to Applications.
2. Open Saywind.
3. Allow Microphone and Accessibility access when asked.
4. In System Settings › Keyboard, set “Press 🌐 key to” to “Do Nothing”, so the key is free for Saywind.
5. Wait for the one-time speech model download (about 630 MB). Then hold Fn and talk.

Requirements

• A Mac with Apple silicon (M1 or later).
• macOS 14 Sonoma or later.
• The Fn (Globe) key needs an Apple keyboard. You can use Right Option instead.

Privacy

Your audio and text never leave your Mac. There’s no analytics, no tracking and no account. Saywind only uses the network to download speech models the first time each is needed, and to check for a newer version at most once a day.

Acknowledgements

Saywind uses these models and software, with thanks:

• NVIDIA Parakeet (CC BY 4.0)
• Parakeet Ultra by Moondream (CC BY 4.0)
• Parakeet CTC 110M by NVIDIA (CC BY 4.0)
• FluidAudio by FluidInference (Apache 2.0)