Real-time speech-to-text subtitles from your Mac's system audio. Captures audio from any application, transcribes it on-device, and displays subtitles in a floating overlay — with optional machine translation.
Keywords: speech recognition, real-time transcription, system audio capture, on-device ASR, floating subtitles, ScreenCaptureKit, Apple Neural Engine, offline captions, language learning, machine translation, macOS accessibility, live captioning, SFSpeechRecognizer, TranslationSession, whisper.cpp
English | 简体中文
- Per-app system audio capture — select any running app (browser, video player, conferencing tool); captures its audio via ScreenCaptureKit without affecting other system sounds
- Real-time on-device transcription — low-latency speech-to-text using Apple Speech framework with Apple Neural Engine; partial results stream continuously as you hear them
- Dual recognition engine — Apple Speech (default, zero-setup, 50+ languages) or Whisper (custom GGML model import via whisper.cpp)
- Multi-language recognition — choose from 14 recognition languages in Settings, including English (US/UK), Chinese (Mandarin/Cantonese), Japanese, Korean, French, German, Spanish, Portuguese, Russian, Italian, Dutch, and Arabic. Models auto-download on first use
- On-device translation — Apple Translation framework runs entirely on the Neural Engine; no network required after model download
- 17 target languages — choose your translation target in Settings: Chinese (Simplified/Traditional), Japanese, Korean, French, German, Spanish, Portuguese, Russian, Italian, Dutch, Arabic, Thai, Vietnamese, Polish, Turkish, and Indonesian
- Toggle on/off — translation can be enabled/disabled from Settings with one click; translated text renders below the original
- Floating overlay — transparent, always-on-top NSPanel that stays visible over fullscreen video; draggable, auto-resizing
- History lines — recently spoken lines fade above the current subtitle; adjustable count (1–5 lines)
- Customizable appearance — adjustable font size (14–36pt), background opacity (10–90%), and window width
- 100% on-device — speech recognition and translation run entirely on Apple Silicon; no data leaves your Mac, works offline after model download
- UI language switching — English / 中文 interface; switch from Settings (requires restart)
- macOS 26.0 or later
- Apple Silicon Mac
- Speech recognition model for your target language — auto-downloaded by macOS on first use
- Translation models — manually downloaded from System Settings for each language pair (one-time setup)
- Open
SubtitleOverlay.xcodeprojin Xcode - Select the SubtitleOverlay scheme and your Mac as destination
- Press Cmd+R to build and run
- Grant Screen Recording and Speech Recognition permissions when prompted
- Play audio in any app (Netflix, YouTube, Zoom, Safari, Chrome, etc.)
- Select the target app from the dropdown in Subtitle Overlay
- Click Start to begin audio capture and transcription
- The subtitle overlay appears — drag it anywhere on screen
- Open Settings (Cmd+,) to:
- Select recognition language and translation target from dropdown pickers
- Toggle translation on/off
- Adjust font size, background opacity, and history lines
- Switch recognition engine or import a Whisper model
- Change UI language
On-device translation requires language models downloaded from System Settings:
- Open System Settings → General → Language & Region
- Scroll to the bottom → Translation Languages
- Download models for your source language and target language
- Click Refresh in Subtitle Overlay to verify
Translation models are separate from system language packs. Having a language as your system preferred language does not install its translation model.
The app separates speech recognition (audio → text) from machine translation (text → text). Each can be configured independently through the Settings UI.
| Aspect | Detail |
|---|---|
| Framework | Speech (SFSpeechRecognizer) |
| Engine options | Apple Speech (on-device) or Whisper (custom GGML) |
| How to change language | Settings → Model tab → Recognition Language dropdown — choose from 14 supported locales |
| Supported languages | Arabic, Chinese (Mandarin, Cantonese), Dutch, English (US, UK), French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, and more |
| Model download | Automatic — macOS downloads the on-device model for your locale on first use |
| Aspect | Detail |
|---|---|
| Framework | Translation (TranslationSession) |
| How to change target | Settings → Model tab → Translate To dropdown — choose from 17 target languages |
| Supported target languages | Arabic, Chinese (Simplified, Traditional), Dutch, French, German, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Thai, Turkish, Vietnamese |
| Model download | Manual — System Settings → General → Language & Region → Translation Languages |
Target App Audio → ScreenCaptureKit → SCStream (16kHz mono PCM)
↓ Speech Recognition Layer
┌─────────────────────┐
│ SFSpeechRecognizer │
│ whisper.cpp │
└──────────┬──────────┘
↓ Translation Layer(optional)
┌─────────────────────┐
│ TranslationSession │
└──────────┬──────────┘
↓
┌─ Translated Text ──┐
↓ ↓
ContentView Preview SubtitlePanelView
(live transcription) (floating overlay)
| Layer | Framework | Role |
|---|---|---|
| Audio capture | ScreenCaptureKit | Per-application audio at 16kHz mono, queue depth 1 for minimal latency |
| Speech recognition | Speech (SFSpeechRecognizer) | On-device ASR via Apple Neural Engine; .search task hint for low latency; supports 50+ locales |
| Translation | Translation (TranslationSession) | On-device NMT via Apple Neural Engine; supports 15+ language pairs |
| Overlay UI | SwiftUI + AppKit | NSPanel (.nonActivatingPanel, .floating level, borderless) hosting SwiftUI via NSHostingView |
SubtitleOverlay/
├── SubtitleOverlay.xcodeproj/
├── SubtitleOverlay/
│ ├── App/ # App entry point & delegate
│ ├── Services/ # AudioCaptureManager, SpeechRecognizer,
│ │ TranslationService, WhisperRecognizer
│ ├── UI/ # ContentView, SettingsView, SubtitlePanelView,
│ │ SubtitleWindowController
│ ├── Models/ # AppSettings, LanguageManager, LanguageOptions
│ ├── Resources/ # Localizable.xcstrings, entitlements
│ └── Assets.xcassets/ # App icon
├── Info.plist
├── favicon.png
├── LICENSE
├── README.md
└── README_CN.md
| Problem | Solution |
|---|---|
| No apps in dropdown | Grant Screen Recording permission: System Settings → Privacy & Security → Screen Recording, then relaunch |
| Speech recognition unavailable | Ensure the speech model for your selected language is downloaded: System Settings → Privacy & Security → Speech Recognition |
| Translation not appearing | Enable "Show Chinese Translation" in Settings; download translation models from System Settings → General → Language & Region → Translation Languages; click Refresh |
| Translation shows "Models not installed" | Translation models ≠ system language packs. Scroll to the bottom of Language & Region → Translation Languages section |
| No audio captured | Confirm the target app is playing audio and its window is visible on screen |
| Subtitles lag behind audio | Reduce queueDepth in AudioCaptureManager (default: 1); ensure .search task hint is set |
Copyright (c) 2026 SweelLong. See LICENSE for full terms.