Caption anything
Your microphone, everything playing on the PC, or only the apps you tick (Chrome, Teams, VLC, …), mixed together.
Free and open source · Windows 10/11
Live captions for everything you hear.
AppLiveCaption turns speech from your microphone, your speakers or just the apps you pick into captions, right on your PC. No cloud, no account, and no internet once a model is downloaded.
Prefer not to install? Get the portable zip
Everything you need to follow along
Made for videos, meetings, games and lectures, in English, Mandarin and the languages of Taiwan.
Live as people talk
Captions update while someone speaks and settle into a final line when they pause.
Floating overlay
An always-on-top caption bar for videos, games and calls. Drag it anywhere, resize it, or let clicks pass through it.
Private by design
Speech is recognized on your PC and never uploaded. No account, no cloud.
Pick your model
Whisper from tiny to large-v3 and turbo, plus models made for Taiwan: Breeze ASR 25 and 26, and Taiwan Tongues. Add your own from Hugging Face or a folder, and remove the ones you don't need.
GPU boost
Got an NVIDIA GPU? One click downloads GPU support and captions get many times faster. No GPU? It runs on the CPU too.
Traditional Chinese
Simplified Chinese output is turned into Traditional (OpenCC s2twp), on by default.
History
Every session is saved. Copy it, or export it as TXT or SRT subtitles.
Install it, or carry it
Install it in a click, no admin rights needed. Or take the portable zip: settings, history and models stay in its folder, even on a USB stick.
Floating overlay
Captions on top of everything
Turn on the overlay and the latest lines float above whatever you're watching or playing.
- Stays on top of videos, games and full-screen calls
- Drag it to move it, drag its edges to resize it
- Click-through, so clicks reach the window below
- Make it taller for more lines, in the size, colors and opacity you like
Feels right at home on Windows
Fluent Design with Mica, light and dark mode, and your accent color.
Pick a model, choose what to listen to and press Start. The status line shows the latency and the device in use.
Download models with one click, add your own, see how much space they take, and turn on GPU support.
Every session is saved with timestamps. Copy it, or export TXT or SRT subtitles.
Up and running in a minute
No Python and no account. Install it, or unzip the portable version and go.
-
1
Install or unzip
Run the installer, or put the portable folder anywhere you like: Documents, another drive, a USB stick.
-
2
Pick a model
A short first-time setup suggests a model that's fast enough for your PC, downloads it, and offers GPU support if your NVIDIA GPU needs it.
-
3
Choose what to listen to
Microphone, system audio, or just the apps you tick. Ticked apps are picked up whenever they start.
-
4
Press Start
Captions appear as people talk. Turn on the overlay to keep them on top.
Which model should I use?
Every model runs on your PC. The app warns you when one will probably be too slow for real time.
Whisper large-v3-turbo
GPUThe pick for an NVIDIA GPU with 6 GB or more
Close to large-v3 accuracy and much faster: partial captions in about 0.3 s on an RTX 4070 Laptop. 100 languages.
Breeze ASR 25
GPUTaiwanese Mandarin, or Mandarin mixed with English
Fine-tuned by MediaTek Research for switching between Mandarin and English mid-sentence.
Breeze ASR 26
GPUTaiwanese Hokkien (Taigi)
Trained on about 10,000 hours of Taigi speech. Writes it as Mandarin Chinese characters.
Taiwan Tongues ASR CE
GPUHakka, Hokkien, Mandarin and English
Whisper large-v2 fine-tuned for the languages of Taiwan.
Whisper base or small
CPUPCs without an NVIDIA GPU
Light enough for real time on the CPU, with good accuracy for clear speech.
Privacy
Your audio stays on your PC
Speech is recognized locally and never uploaded. The app goes online only when you ask it to:
- Downloading a model from Hugging Face
- Downloading GPU support from PyPI
- Checking GitHub for a new version
Everything else, from starting the app to captioning, works without internet.
Questions and answers
Which versions of Windows does it work on?
Windows 10 and 11, 64-bit. Captioning individual apps needs Windows build 20348 or later (in practice Windows 11); on older versions the microphone and system audio still work.
Do I need a graphics card?
No. Without an NVIDIA GPU, models run on the CPU; pick Whisper base or small for real time. With an NVIDIA GPU and a recent driver, download GPU support (first-time setup offers it, or use the Models page) and use the bigger models.
Does it need the internet?
Only to download models and GPU support, and when you check for updates. After that it works completely offline.
Where does it keep my settings and transcripts?
Installed, it uses your Windows profile: settings and transcripts in %APPDATA%\AppLiveCaption, and logs, models and GPU support in %LOCALAPPDATA%\AppLiveCaption. Uninstalling lets you choose which of them to delete. The portable version keeps everything next to AppLiveCaption.exe, so moving or deleting its folder moves or removes everything.
Can it caption just one app, like a browser or Discord?
Yes. Choose Apps as the source and tick the apps you want; only their audio is used, optionally mixed with your microphone.
Does it translate?
No. It writes down what is said, in the language it's said in. Translation and speaker identification are out of scope.
How much does it cost?
Nothing. AppLiveCaption is free and open source under the GPL-3.0 license. The speech recognition models are made by their authors and come with their own licenses, shown on the Models page.
Is it made with AI?
Yes, it's vibe-coded! I built it with Claude Code, Anthropic's AI coding assistant, running Claude Opus 5.5. Without this amazing tool, I wouldn't have been able to build the app I wanted in such a short time.
See what's being said
Free for Windows 10 and 11. Install it and you're ready.
Prefer not to install? Get the portable zip