Transcribe each session verbatim with speaker labels, search a transcript for the theme you're chasing, and quote a participant with a timestamp you can click back to and hear. Your recordings stay in your account.
100+ languages · speaker labels · timestamps throughout · never used to train a model
Thirty hours of recordings is thirty hours you can't re-listen to. But coding a theme means going back to what someone actually said — not to your note about what they said. A verbatim transcript with the timestamps intact is what makes the second read cheap enough to do properly.
Upload the recording or paste a link. Long single-session interviews are the normal case — no short length cap, and timestamps run the whole way through. Both voices are labelled, so the moderator's prompt and the participant's answer read as separate turns.
The transcript keeps the language it was spoken in — no silent translation into English, which is where nuance goes to die in a multi-market study. Over 100 languages.
Type a phrase and get every line in that interview containing it, each with its timestamp, so you can judge relevance before you jump. Accents and case are ignored — typing codigo finds código.
“Where did she talk about switching costs?”Ask what a participant objected to, or what they said about price, and get an answer grounded only in that transcript — with timestamps cited, so the claim is checkable rather than merely plausible.
Click a citation and the transcript jumps to that line while the audio plays from it. Hear the tone, confirm the words, then paste the quote and its timestamp into the deck.
Export TXT, DOCX, JSON, SRT or VTT. JSON carries per-segment speaker labels and timings — the format to use when you're importing into a coding tool or a script of your own.
The words, the speaker, and the moment — so a reader can check you.
“We didn't leave because of the price. We left because every time we asked for something, it took three emails and a call.”
Illustrative example. Speaker labels are anonymous (SPEAKER_00, SPEAKER_01) — you map them to moderator and participant, and the participant ID is yours to assign.
Hour-long one-on-ones where the useful admission arrives forty minutes in, unannounced, and quoting it means finding it exactly.
Diarization separates the voices in the room, so a disagreement reads as a disagreement instead of a paragraph.
Each transcript stays in its own language, so a local researcher reads the original rather than a translation of a translation.
The customer's own phrasing for why they left, quotable, rather than a summary line in the CRM written by whoever lost the deal.
Search a session for the screen name or the feature word and pull every moment a participant stumbled on it.
The Chrome extension records the audio of a session playing in a browser tab, or your own mic, and submits it as a job.
Search is per transcript, not per corpus. There is no single box in the app that searches all thirty interviews at once. To sweep the whole study, pull each transcript through the API as JSON or TXT and search them together in your own tool — or connect the MCP server and have an agent walk the transcripts for you.
It is machine transcription. Strong on ordinary speech, weaker on brand names, figures and jargon — exactly what a quote often turns on. It is not a certified verbatim record. Listen to the passage before it goes in a report; that's why every citation here links to the audio instead of being a number on a page.
Participants agreed to talk to you, not to a vendor's training set.
Transcripts and recordings sit in your account. Nothing is published, and nothing appears on a public page unless you deliberately put it there.
Audio is transcribed on our own GPUs running open-source Whisper. It isn't handed to an external speech-to-text service.
Your recordings and transcripts are not training data — a policy statement you can put in an ethics submission.
Uploaded audio is deleted after the retention window for your plan. The transcript stays until you delete it — and you can delete either.
Priced per minute of audio, so a study is a line item you can put in a proposal before you run it.
| Plan | Price | Minutes | Roughly |
|---|---|---|---|
| Starter | $4 | 300 | 5 hours — a pilot round |
| Pro | $12 | 2,000 | 33 hours — a full study |
| Team | $24 | 5,000 | 83 hours — a research group |
Worked example: 30 interviews at 60 minutes each is 30 × 60 = 1,800 minutes. That sits inside Pro's 2,000 minutes at $12, with 200 minutes spare for the two sessions that ran long. A 10-interview pilot at 45 minutes each is 450 minutes — over Starter's 300, so Pro is the cheaper single purchase. Credits don't expire. Full pricing →
Submit each recording, poll the job, and pull JSON with speaker labels and timings — the same transcripts, in a folder you control.
# one interview, diarized
curl https://whipscribe.com/api/v1/transcribe \
-H "X-API-Key: $WHIPSCRIBE_KEY" \
-F "file=@P14-2026-08-04.m4a" \
-F "diarize=true" \
-F "source=api"
# poll GET /api/v1/jobs/{job_id} until status is "done", then
# GET /api/v1/jobs/{job_id}/result?format=json → text, language, segments[]Full reference in the API docs; keys come from /keys. Building an analysis agent on top? The developer walkthrough →
Put one session through, search it for the theme you already suspect is there, and see whether the quote you find is better than the one in your notes.
Transcribe an interview →It transcribes what was said, in the language it was said in, without paraphrasing — but it's machine-generated, not a certified verbatim record. Strong on ordinary speech, weaker on brand names, figures and jargon. Listen before you publish a quote.
Search runs inside a transcript. For a corpus-wide sweep, export each transcript through the API and search them together in your own tool, or connect the MCP server and have an agent iterate them. We'd rather say that plainly than let you discover it on interview twenty-nine.
Yes — diarization splits the recording into labelled turns. The labels are anonymous, so you map speaker to moderator or participant once per session.
They live in your account. Audio is transcribed on our own GPUs running open-source Whisper — not passed to a third-party transcription vendor, and never used to train a model. Audio is deleted after your plan's retention window; the transcript stays until you delete it. See /security.
Export TXT, DOCX, JSON, SRT or VTT — JSON if you want speaker labels and timings preserved for a coding tool. There's no packaged integration with a specific analysis product.
Up to 10 hours per file, so a three-hour group session is unremarkable. Timestamps and speaker labels run the whole way through. If your material is hearings or lectures rather than interviews, /study-transcripts is the closer fit.