Recording and Transcribing
Record a meeting or a thought, get a speaker-separated transcript you can search and read along with, and keep it encrypted throughout.
Some material arrives as a document. Some arrives as a conversation.
LumaVista records audio, transcribes it, separates who said what, and puts the result where the rest of your work is — searchable, linkable, and encrypted from the moment it leaves the microphone.
Recording
Record in the browser or upload audio you already have. A recording in progress survives the tab; uploads that have not finished are held and resumed rather than silently dropped.
The audio is encrypted with your key before it is stored. This has a consequence worth stating plainly rather than discovering later: if your session is locked, transcription cannot run. There is no key to open the audio with, so the job waits rather than failing or falling back to something unencrypted. Nothing is processed on your behalf while you are away unless you have separately turned on background execution.
The transcript
Transcription produces more than a wall of text.
Speakers are separated. The system identifies distinct voices and labels their turns, so a three-person meeting reads as three people rather than one long paragraph. Labels start generic — Speaker 1, Speaker 2 — because it knows there are three voices, not who they are.
Timing is per word. Every word carries its position in the audio, which is what makes the next part work.
Playback and text are synchronised. Play the audio and the transcript follows. Click a line in the transcript and the audio jumps there. Finding the thirty seconds you actually need in a ninety-minute meeting stops being a scrubbing exercise.
Highlights mark passages worth returning to.
Search runs across transcripts, so you can find the meeting where something was said without remembering which meeting it was.
Recognising your own voice
Optionally, LumaVista can learn your voice and label your turns automatically — so you read “You” instead of “Speaker 2” and can filter for what you actually committed to.
This is a voiceprint, which is biometric data, so it is handled accordingly: explicitly opt-in, encrypted under your key like everything else, and covering only the account holder. It is not offered as a way to identify other people in your recordings, and it does not do that.
Availability. Recordings and transcription are in beta and off by default, and need a suitably equipped host. Voice recognition is a separate switch on top. If you do not see the recordings surface, it is not enabled for your account — ask and we will turn it on.
Turning a recording into working material
A transcript on its own is a record. What makes it useful is what happens next.
Promote it to a document. One action turns the transcript into a document in your library — converted, indexed, and searchable alongside everything else. From that point research and chat can draw on what was said in the room the same way they draw on a PDF. The recording keeps a link to the document it produced, so you can always get back to the audio.
This is opt-in per recording. A recording is transcribed for you to read; it becomes part of your searchable corpus only when you say so.
Or let it act. An opted-in recording can start a workflow that reads the transcript and does something with it — files the commitments, starts the research, drafts the follow-up. That is Voice to Action, and it has its own consent model because it is a much larger step than storing text.
What is protected, and how
| Stage | Where it is |
|---|---|
| Audio at rest | Encrypted under your key |
| Transcript | Encrypted under your key |
| Voiceprint, if enabled | Encrypted under your key |
| Transcription itself | Runs on infrastructure we operate, not a third-party speech API |
That last row is the one that separates this from most transcription products. Your meetings are not sent to an external speech service to be turned into text.
Live voice chat is a different feature with a different answer — it does reach an external provider, which is why it has its own explicit consent. See Where Your Data Is Processed.
Related
- Voice to Action — when an opted-in recording starts a workflow
- Notes and Wikilinks — where transcript material tends to end up
- Setting Up Your Encryption — why a locked session stops transcription