Review a recorded interview without uploading the source audio
Create a first-draft transcript for journalism, research or user interviews while keeping the selected recording on the current device.
Local processing reduces one data handoff: the interview recording is not sent to Whisper Web. You still need permission to record, a suitable device and a careful review process before quoting or sharing the transcript.
A review-first interview workflow
- 1
Keep the original recording
Preserve an unchanged source file and work from a copy when the interview matters as evidence.
- 2
Transcribe locally
Choose the recording, language, model and browser compute backend, then start transcription.
- 3
Check every important quote
Use timestamps to return to the audio. Verify names, numbers, technical terms and negations.
- 4
Export the minimum needed
Share the corrected text instead of the source audio when that is sufficient for the next reviewer.
Where the local workflow helps
Journalism
Search a source interview and return to the timestamp before using a quotation.
Qualitative research
Create a draft for coding and analysis without adding a transcription upload to the data path.
User research
Review a recorded session locally, then move only approved excerpts into the research repository.
Privacy is a workflow, not a badge
Local transcription does not by itself make an interview process compliant with a law, contract or research protocol. Recording consent, device security, backups and exports remain your responsibility.
Whisper Web does not identify speakers. For multi-person interviews, add labels during review and verify every passage where attribution matters.
Questions about this workflow
Is local transcription safe for confidential interviews?
It avoids uploading the selected recording to Whisper Web, but confidentiality also depends on the device, backups, permissions and how you share the exported transcript.
Can I publish the transcript without editing it?
That is not recommended. Treat automatic transcription as a draft and check quotations, names, numbers and sensitive passages against the recording.
Can it separate the interviewer and participant?
No. Speaker diarization is not implemented. Add speaker labels manually after comparing the timestamped segments with the recording.