Help

Start a transcript

1. Wait for Moonshine Base to finish downloading, or select a downloaded model.
2. Enter a transcript name. The suggested name is Call followed by the local date and time.
3. Enable the microphone and/or share a screen. Start transcription becomes available when a recognition model is ready.
4. Choose Start transcription. Pause keeps the session open; Stop saves it locally and releases capture tracks.

Moonshine Base is downloaded automatically on its first use. Other speech models are optional and download only when chosen.

Microphone and screen audio

The microphone switch asks the browser for permission. Screen sharing uses the browser’s share picker. Whether computer audio is available depends on the browser, operating system and the item chosen in that picker. A screen can still be captured when the browser does not provide an audio track.

Audio processing can reduce rumble and even speech volume. Turn it off if it makes a particular microphone less clear. The settings button opens the advanced controls; the default settings are normally the best starting point.

Screenshots and OCR

After sharing a screen and starting a transcript, choose Manual, Every interval, or When screen changes. Intervals can be set from one second to five minutes. Intervals below one minute can use significant browser storage.

Enable local OCR in Setup if you want readable screenshot text in the transcript. Enabling it downloads PaddleOCR and its runtime into browser-managed storage. OCR runs in the browser, not on a server. Each screenshot image uses its OCR result as alternative text. OCR can make mistakes; check the image or its source when wording matters.

Saved sessions and exports

Sessions, transcript segments and screenshots stay in this browser until you delete them. Open a saved session to return to it, or download TXT, JSON, SRT or VTT. Use the source filters to show microphone, screen audio, screenshot or OCR entries without splitting the chronological transcript.

Optional automatic speaker separation downloads a local browser model and labels anonymous speakers as a call progresses. Name a detected speaker once to apply that saved person’s name to their later blocks in that call. It does not recognise people in another call, and you can correct or add overlapping speakers manually.

Clearing browser site data can remove models and saved sessions. Export important transcripts before clearing site data, changing browsers, or using private browsing.

Accessibility

All primary controls have visible labels and keyboard access. Download status uses progress elements and status messages. Source filters are standard checkboxes. Screenshot alternative text is the local OCR result once processing is complete. Theme choices include light, dark and muted modes.

Important limitations

Do not rely on automated transcription or OCR as an authoritative record. Obtain consent where required before recording people or capturing their screens. This app does not replace a secure records system, a legal record, or accessibility support provided by the original meeting platform.