See what's new in Spoke

The original Spoke bars react more clearly to your speech and move smoothly into the Bounce processing animation.

Long dictation joins sections at natural pauses, helping avoid repeated words where sections meet.

Spoke finishes recording when you release the shortcut and prepares your text with less waiting. Settings and history also open more quickly.

Fixed recent regressions affecting paste delivery and the extra Spoke Helper icon in the Dock.

With Nemotron, words appear while you speak. The text grows line by line, with a caret that follows your words without the view jumping around.

Parakeet and Nemotron turn spoken times, percentages, and units into written values. Parakeet also removes hesitation words such as “um” and “uh” and corrects sentence capitalization.

Model choices show when they are ready. Changing your selection does not switch the model in the middle of a recording.

Spoke does less work while recording and loads older history as needed. Settings and model lists load when you open them.

Fixed cancelled model starts that could block the next recording. If speech detection fails while you speak, Spoke keeps the audio and transcribes it when you finish.

Save names and specialized words in Dictionary. Spoke corrects close matches, including multi-word phrases, across local speech models.

NVIDIA Parakeet is available as a local speech model and becomes the recommended choice for English dictation.

Spoke remembers your microphone choice and makes your speaking level easier to see.

Record for up to five minutes. Short dictations stay together, while longer recordings are processed in smaller sections.

Notifications appear after dictation finishes, so they do not interrupt your recording.

Spoke adjusts dictation to the text around your cursor and keeps surrounding punctuation intact.

Fixed an issue that prevented saved transcripts from being copied correctly.

Downloading an update no longer restarts Spoke automatically. The update waits until you choose to install it.

Turn on Auto-Space in settings to leave a space after inserted text.

Saved vocabulary gives the local speech model context for the names and specialized terms you use.

Fixed downloads that failed to start or retry. Download progress appears in settings and the menu bar, and failures show a notification. If you still use 0.1.11 or 0.1.12, install a current version once from this website to restore in-app updates; your settings, history, and models are kept.

Install Whisper or Cohere Transcribe during setup, then switch models in settings. Model cards show download progress, let you cancel a download, and mark which model is installed and active.

Spoke tells you when a speech model is missing or dictation repeatedly fails, instead of leaving you waiting for text.

Grant the required permissions, restart Spoke, and continue where you left off. The settings panel also fits its contents more consistently.

Fixed problems that prevented Spoke from finding or installing updates on Apple Silicon Macs.

Turn your voice into text entirely on your Mac. Nothing is uploaded, and it works offline.

Spoke types straight into your active cursor, in any app, with no copy and paste.

The entire app is open for anyone to read, audit, and build on.