LoopbackTranscriber
Windows desktop application · Open source (MIT)
Record, transcribe and summarize your meetings without anything leaving your computer.
It records the audio of one specific application (Teams, Zoom, Meet in the browser…) together with your microphone. It then transcribes it with Whisper, tells what you say apart from what the others say, and, when you ask for it, writes a summary with a language model. While you record, it can also show the meeting translated in a floating window.
How to use it
Optional: open live translation to follow the meeting translated while you record.
-
Pick the application
Teams is built in. Zoom, the browser or others are added in Applications.
See the guide -
Record
Press Start recording (or use the tray icon). The application, the default audio output and your microphone are recorded at the same time.
See the guide -
Get the transcript
When you press Stop recording, Whisper transcribes it and, if your microphone was recorded, separates Participants from You. It is saved in the history.
See the guide -
Summarize
In the AI summary tab, press Summarize: title, key points, goals and next steps.
See the guide
Features
Each feature links to its user guide and its technical explanation.
Per-application recording
Only the audio of the chosen program, plus your microphone on a separate track. If that track turns out silent, it uses the default audio output.
Any application
Teams is built in. Add Zoom, Discord or the browser (for Meet) from the processes that are playing audio.
Transcription with Whisper
Whisper large-v3-turbo, on a Vulkan GPU when there is one. Fixed language or automatic detection. Filters hallucinations, repetitions and echo.
Who is speaking
Your microphone is transcribed separately and aligned with the meeting. Each paragraph says Participants or You, with your name in front if you set one.
History and search
Search the title, application, date, summary and full text, ignoring case and accents. Filter by date and duration, and transcribe again with the current settings.
AI summary
Qwen3-4B writes a title, purpose, key points, topics, goals and next steps, in the application language.
Live translation
Floating window with what the participants say and its translation into Spanish, English, Portuguese, German or Russian.
Privacy and data
What uses the internet and where your files end up.
Everything stays on your computer
Audio, transcripts and summaries never leave it. The application only goes online to download the two models, from a fixed Hugging Face revision and checking their SHA256. The installer may also download Microsoft Visual C++ Redistributable if it is missing.
Only record with the consent of the people taking part.
Where everything is stored
In %LocalAppData%\LoopbackTranscriber: settings, models, transcripts, unprocessed recordings and log. The models and transcripts folders can be changed in Settings.
Each recording is a folder with the text in Markdown (.md) and the audio (.wav), plus, when there are any, your microphone (_mic.wav) and the summary (.summary.json).