Live subtitles,
right from your notch.
NotchSub turns any audio your Mac plays — videos, calls, podcasts, streams — into live subtitles and translations. Free, open source, and completely on-device.
NotchSub turns any audio your Mac plays — videos, calls, podcasts, streams — into live subtitles and translations. Free, open source, and completely on-device.
Click the notch, and NotchSub starts listening to your Mac's audio, never your microphone.
Speech recognition and translation run entirely on your Mac. No account, no cloud, no audio ever leaves your computer.
Powered by OpenAI Whisper. Translate into 20 languages with Apple's on-device translation, or detect the spoken language automatically.
Words appear about a second after they're spoken, accelerated by the Apple Neural Engine.
MIT licensed, native Swift. No subscriptions or ads, and nothing is tracked. Read the code, fork it, improve it.
Three steps, about a minute to set up.
A panel drops down from the top of your screen and starts listening to system audio.
A YouTube video, a Zoom call, a lecture or a podcast. If your Mac plays it, NotchSub can subtitle it.
The translation appears in large text with the original line below it. Click the notch again to close the panel.
Pick whichever you prefer. Both install the same app.
Open Terminal and paste this command. It downloads the latest version into Applications and opens it.
On first launch NotchSub downloads the Whisper speech model (~630 MB) and optimizes it for your Mac's Neural Engine, which can take several minutes. Later launches take a few seconds.
No. NotchSub uses Core Audio's system audio tap and only hears what your Mac plays. Your own voice is never captured.
Only two one-time downloads: the speech model from Hugging Face, and Apple's language packs. Transcription and translation happen offline after that.
Any Apple Silicon Mac (M1 or newer) running macOS 15 Sequoia or later. Notch-less Macs and external displays get a small virtual notch at the top of the screen.
NotchSub is a free community project and isn't notarized with a paid Apple Developer ID yet. The one-line installer avoids the warning, or you can approve the app once in System Settings → Privacy & Security.
Speech can be recognized in 99 languages. Translation uses Apple's on-device engine, which covers about 20 major languages. The app's own interface follows your Mac's system language.