English | 日本語

Nochi (ノッチ)

Real-time live translation of system audio displayed as subtitles in the macOS notch area.

A macOS menu bar app that captures system audio (not the microphone) via Core Audio Process Tap, performs real-time speech recognition, translates to a target language using Apple's on-device Translation framework, and displays live subtitles in a notch-shaped overlay pinned to the top of the screen.

Watch a YouTube video in Japanese, join a meeting in Spanish, or listen to a podcast in Korean — Nochi shows you live translated subtitles without touching your mic or sending anything to the cloud.

No DRM black screens — unlike ScreenCaptureKit, Process Tap uses kTCCServiceAudioCapture instead of screen recording, so DRM-protected apps (Netflix, Unext) won't black out their video.

• System audio only — captures all sound output via Core Audio Process Tap, no microphone needed
• No DRM interference — shows a purple dot instead of the screen recording indicator, DRM apps keep playing normally
• Zero-config — taps all system audio as stereo, works with any app

• Apple Speech — SFSpeechRecognizer with real-time partial results, on-device when available
• WhisperKit (pluggable) — CoreML-accelerated Whisper for higher accuracy, 3-second chunked transcription
• Auto-restart — seamlessly restarts on Apple Speech's ~60s timeout
• Watchdog — detects silent failures and force-restarts recognition
• Pause detection — commits sentences after 1.5s speech pause, even without punctuation

• Apple Translation — fully on-device, private, no API keys
• 20+ language pairs — Japanese, Korean, Chinese, Spanish, French, German, and more
• Real-time — translates partial results as speech is recognized, not just final sentences
• Prioritized partials — live partial translations take priority over committed sentences for responsive display

• Notch-shaped UI — custom AppleNotchShape with rounded lower corners, blends with the hardware notch
• Two display modes — "Translation Only" (clean subtitles) or "Original + Translation" (language learning)
• In-overlay language picker — switch source/target language without opening settings
• Control buttons — start/stop, display mode toggle, font size, quit
• Always on top — NSPanel at .screenSaver level, joins all Spaces, non-activating

• Carbon Events API — system-wide shortcuts that work even when the app is in the background

Nochi detects when on-device speech or translation models aren't installed for the selected language pair and guides users to download them — so they're never stuck wondering why transcription/translation isn't working.

A subtle dot appears on the mic icon when the current source → target pair needs setup.

Clicking the mic button pops up a friendly dialog instead of silently failing. The user can jump straight to System Settings or start anyway (server-based).

The app opens the exact System Settings pane where the language pack can be downloaded.

Common pairs (JA↔EN, EN↔ES, EN↔ZH…) are pre-installed by macOS and never trigger the flow. It only appears for pairs that actually need a download — like JA → TH above.

Switch engines in Settings. WhisperKit requires adding the WhisperKit SPM package.

Source and target languages can be mixed freely. Common pairs:

Any pair supported by Apple Translation works. Language packs download on first use via System Settings.

• macOS 15.0+ (Sequoia — required for Apple Translation framework)
• MacBook with notch (works without notch too, overlay pins to top of screen)
• Xcode 16+
• Audio Recording permission (for Core Audio Process Tap system audio capture)
• Speech Recognition permission (for SFSpeechRecognizer)
• Translation language packs (downloaded via System Settings > General > Language & Region > Translation Languages)

git clone https://github.com/jonpol01/Nochi.git
cd Nochi

open Nochi.xcodeproj
# Product -> Run (or Cmd+R)

No CocoaPods, no SPM dependencies for the base build. Pure Apple frameworks.

On first launch:

1. Audio Recording — macOS will prompt for audio capture permission
2. Speech Recognition — grant when prompted, or via System Settings > Privacy & Security > Speech Recognition

Go to System Settings > General > Language & Region > Translation Languages and download the language pair you need (e.g., Japanese + English).

Press Opt+Cmd+L or click the waveform icon in the menu bar > "Start Listening". Play any audio and watch subtitles appear in the notch.

For higher accuracy speech recognition via local Whisper models:

1. In Xcode, go to File > Add Package Dependencies
2. Enter: https://github.com/argmaxinc/WhisperKit
3. Replace the stub in SpeechRecognizer.swift with actual WhisperKit calls
4. The model (~150 MB) downloads on first use

MIT