Back to work

spitr

Hold a key, talk, and your words appear at the cursor, transcribed on-device and cleaned up by the model you choose.

0:04

The project

spitr is a voice tool for macOS and Windows. Hold the right Option key, talk, and let go: the audio is transcribed on the machine with NVIDIA Parakeet v3, a language model of your choice removes the fillers and fixes the punctuation, and the text is typed wherever the cursor is. Select text anywhere and press ⌃⌥S to hear it read aloud. German and English are detected on the fly, and the audio never leaves the computer.

Try it in any text field: hold the right Option key, say a sentence with all its ums, and watch the clean version land at your cursor. spitr is free and provided as is; the install guide, terms of use and privacy notes are on the releases page.

My part

Built solo, end to end: product, interaction design and engineering. The Mac app is native Swift and AppKit with an on-device CoreML speech model, and the recording island draws its waveform from the live mic level. The Windows version runs on a shared .NET core with a WPF shell. Signed builds and Sparkle updates ship from the public releases repo.