Chapter 5: Live Captions and Live Notes

While a recording is in progress, two things run alongside it in the browser: a live transcript and (for signed-in users) a live notes editor. This chapter covers what each is, what it can do, and how they persist through interruptions.

Live captions

The transcript comes from SeaMeet's live speech-to-text service. Audio streams from the recorder to the STT service; caption segments stream back and append to the transcript pane as they arrive.

What you'll see:

• Timestamped segments ([00:03:12] prefixes) at the start of each caption line.
• Speaker labels ("Speaker 1", "Speaker 2") — the AI attempts speaker separation once distinct voices are detected, typically after the first minute.
• A short delay between spoken word and printed caption — usually 1–3 seconds, because STT works on rolling windows of audio.
• A "Live transcription paused" pill in all 23 languages if you pause the recording (see below).

Timestamp accuracy — the reason captions stay aligned

Timestamps in the caption pane are the same clock as the audio. Concretely (fixed in v1.21.0):

• Timestamps are derived from sample counting on the audio actually sent, not from wall-clock time.
• Caption clocks no longer re-anchor to wall-clock on every STT reconnect (which used to accumulate 0.5–2 s of drift per roll).
• The first caption now lands at 00:00, not late by the startup window.
• On long recordings, the transcript stays aligned across reconnects, network gaps, and clock adjustments.

This means when you replay a recording later and click a transcript segment, the media seeks to exactly that moment — every time, for the whole recording.

Pause and resume — transcript-safe

Pressing Pause on the recorder suspends both the media recorder and the STT socket. That's a real behaviour change (v1.21.0):

• Anything you say while paused is neither captioned nor recorded.
• The captions pane shows a "Live transcription paused" pill.
• After Resume, the next caption is timestamped at the exact point in the audio where recording picks up — no drift.
• The very first fraction of a second of speech right after Resume, while the STT connection reopens, may go uncaptioned.

Before this fix, pause only stopped the recorder; STT kept running against silence and captions accumulated with the wrong timestamps.

When captions stop coming

• STT service temporarily unavailable. A Retry captions banner appears above the transcript. Click it — the recording itself is unaffected.
• Free transcription pool exhausted. For long recordings on free tier, live captions may end mid-session. The transcript is still generated from the audio on upload; you just lose the live preview.
• You're offline. Live captions are a cloud service; the recording continues locally and the transcript is regenerated from the audio after upload.

Live notes (signed-in only)

Signed-in users get the full WYSIWYG notes editor on the recording screen — the same editor as the post-recording Notes tab, right below the timer and Pause/Stop controls (or in a collapsible "Notes" card on mobile).

What the editor can do:

• Headings, lists, quotes, bold, italic, links.
• Markdown shortcuts as you type (**bold**, - list, etc.).
• Formatting toolbar with buttons for the common cases.
• Full undo / redo.

Timestamp inserts. The ⏱ toolbar button drops a [mm:ss] marker at the cursor using the current recording time, so you can jot "decision on Q3 rollout at ⏱" and get back to the exact moment later. In playback, the timestamp becomes a clickable link that seeks the media.

Crash-safe local draft. Notes mirror to IndexedDB while you type — if the tab dies mid-recording, the draft survives on disk (24-hour retention). Discarding an unfinished recording from the recovery modal shows a warning if unuploaded notes would be lost with it.

What happens on Stop. The draft uploads to the cloud together with the recording (never mid-recording). It continues seamlessly into the recording's Notes tab, and is weighted into the AI summary when the summary is regenerated — the notes are treated as first-class context for the AI, alongside the transcript.

What anonymous sessions get. No live notes editor in the current release. Anonymous recordings capture audio + video + live transcript, but the notes editor requires a signed-in identity to attach the draft to on the server.

Markdown rendering in the wiki and elsewhere

A convention worth knowing: single line breaks in notes render as line breaks (as of v1.21.0) — "comment-style" formatting the mobile app and Chrome extension also use. Blank lines still separate paragraphs. Caveat: text formatting must start and end on the same line — **bold** split across a line break shows the asterisks literally.

Quick Reference

Task	How
Retry captions if they stop	Retry captions banner above the transcript
Insert a timestamp linked to the recording	⏱ toolbar button (or type [mm:ss])
Bold / italic / list	Editor toolbar, or markdown as you type
Pause safely (no caption drift)	Pause button in the recorder controls
Recover a note after a tab crash	Open /record again — the draft appears in the notes card

---

Last updated: 2026-08-28

← Chapter 4: The Per-App Record Pages | Chapter 6: Stopping and Saving →