from audio to insight
What happens after you hit stop — the four-step pipeline, the transcript view, retrying, exporting.
The moment a recording ends (or an import lands), it flows through a local pipeline of four steps. You'll see the same four icons on every recording card, each one lighting up as it completes:
| step | what it does |
|---|---|
| listen | voice → text |
| recognise | who said what — voices are separated and recognised |
| understand | meaning: punctuation, names, sentiment — and making it searchable |
| insight | the AI: title, summary, and every lens you've enabled |
Nothing here blocks anything else — a recording is safe on disk the moment you stop, and the rest happens after, at its own pace. If a step fails (say, a model wasn't downloaded yet) the recording shows exactly which step failed, with an inline retry.
the transcript
A finished transcript reads as turns: who spoke, what they said, with detected entities and a sentiment hue per turn. Speakers you've named show their names; unknown voices show as Speaker N until you name them.
Not happy with the words? Retranscribe lets you re-run speech-to-text with a different (say, bigger) model — useful after upgrading models on a recording that mattered.
exporting
Export transcript writes the whole thing to a file: Markdown, plain text, WebVTT, SubRip (.srt), JSON or CSV. Subtitle formats carry timestamps, so a transcript drops straight onto its video.