Parakeet engine with Apple Neural Engine enables fast and accurate transcription: ~300 ms per sentence, ~700 ms per paragraph, and around 20 seconds for 1 hour of audio.

Transcribe audio files in MP3, WAV, M4A, FLAC, and OGG formats via drag and drop.

Full transcription history with searchable text.

Recording overlay, start-minimized option, and hold-to-talk mode.

This issue has been reproduced and fixed in version 1.0.44, now available on the website. Note that when you update the app, the models will be recompiled, so the app may briefly hang during recording. Each new app version → new hash → AI (MLX) code recompilation.

Thank you for the feedback. I will improve the onboarding and setup in the next app release.

Sorry about that — this feature is no longer available due to its cloud-based nature, but I plan to bring it back. The video was recorded before open-sourcing, so I’ll remove it to avoid confusion.

Update: Replaced the site video with a demo showing the actual app behavior.

Thank you for the feedback. I’ll work on reproducing the issue and will do my best to fix it.

I used it quite a bit today and happy to say that it worked flawlessly. Thanks for making this app!

Thank you — any feedback is appreciated. I want to be useful, not rich, so I made it open and free for everyone

Try https://whisperclip.com —it delivers low-latency, real-time voice-to-text streaming to any macOS app.

No, I use an in-house model based on some open analogs.

No, they already have, but their mechanism have a lot of limitations because it's not a primary business for them. More details you may read in our blog article - https://cydanix.com/blog/Evaluating-the-Limitations-of-Nativ... .

Sorry, correct link for article is https://cydanix.com/blog/evaluating-the-limitations-of-nativ...

No, but may be in future