Send us the recorded show. You get back a speaker-labelled transcript with chapters, so a three-hour programme becomes a navigable log — every segment, every guest, every phone-in, findable by typing a word.
Recorded files, not a live feed · speaker labels · chapters · 100+ languages · up to 10 hours per file
Whipscribe is asynchronous. You give us recorded audio — a show export, a rendered interview, the hourly file your logger already writes — and the transcript comes back once the job has run. We do not connect to your stream, there is no always-on listener, and nothing here alerts on words as they go to air.
That is the honest trade. In exchange you get what a live captioner can't give you: diarized speakers, chapters, summaries, a searchable archive that keeps growing, and a per-minute cost low enough to run across everything you air rather than the one show you can afford to caption.
A station airing 18 hours a day produces 1,080 minutes of speech every day — roughly 6,500 hours a year. It is recorded, it is stored, and almost none of it is text. So the guest who said the useful thing in April is unfindable, the interview never becomes a web page, and producing show notes means someone scrubbing the tape they just made.
Upload the show export, point the API at each file, or let us read a folder in your own S3 bucket or Google Drive. Up to 10 hours per file, in mp3, m4a, wav, flac, ogg, mp4, mov or webm — so a three-hour programme is one job and a full day is the hourly files your logger writes anyway.
Diarization is on by default and splits the audio into labelled turns, so a two-hander plus three phone-ins reads as a conversation rather than a wall of text. Labels mark speakers as distinct; you rename them once and the name carries through.
The transcript is broken into chapters with timestamps, which is what a programme log actually is — segment starts, the interview, the travel, the competition. Skim the chapter list, click the one you want, land on the second it begins.
Type a name, a sponsor, a track title and get every line that contains it with its timestamp. Or ask the transcript a question and get an answer whose citations are clickable — the audio plays from the moment being quoted, so a producer verifies before publishing.
“What did the guest say about the festival line-up?”The summary and chapter list are most of a show-notes page. Pull quotes with their timings for social, export SRT or VTT to caption a video cut of the interview, or take TXT, DOCX or JSON into your CMS. The hour that aired once becomes a page that keeps earning.
Long-form talk, guests worth crediting, and an archive that should be an asset.
Breakfast or drive, five days a week, mostly talk. Chapters give you the running order after the fact; search gives you every time a topic came up across the run.
Speaker-separated text makes it trivial to pull exactly what a guest said, credit them accurately, and send them the quote — which is how you get the next booking.
Summary, chapters and quotes come out of the same job, so the page goes up the same day rather than never. Your on-air talk starts answering search queries.
Years of recordings nobody queries, because querying means listening. Run them through in batches and the archive turns into something a producer or a researcher can search.
One call per file, an API key, and a poll for the result. Same pipeline as the web app; the transcript comes out as TXT, JSON, SRT, VTT or DOCX.
# submit one hour of logger audio curl https://whipscribe.com/api/v1/transcribe \ -H "X-API-Key: $WHIPSCRIBE_KEY" \ -F "file=@2026-08-06-0900.mp3" -F "language=en" \ -F "diarize=true" -F "source=api" # → {"job_id":"35f4be54-…","status":"queued"} # poll, then fetch the transcript curl -H "X-API-Key: $WHIPSCRIBE_KEY" https://whipscribe.com/api/v1/jobs/35f4be54-… curl -H "X-API-Key: $WHIPSCRIBE_KEY" "https://whipscribe.com/api/v1/jobs/35f4be54-…/result?format=json"
Give us read-only, prefix-scoped access to the S3 bucket or Google Drive folder your logger writes into, and whole folders get transcribed without anyone touching a browser.
Whipscribe ships an MCP server, so an assistant like Claude can submit files, read transcripts and pull quotes inside a producer's workflow instead of a separate tab.
For the one-off — a guest's own upload, a rival's interview, an archive page — the Whipscribe extension sends browser audio straight into the queue.
Endpoints, fields, statuses and formats, including how jobs are scoped to your key. Written for the engineer who has to wire this to the logger.
Credit packs are one-time and don't expire: $2 / 60 min, $8 / 1,000 min, $12 / 2,000 min, $24 / 5,000 min. At the 5,000-minute pack that is just under half a cent a minute, which is what makes station-scale volume arithmetic rather than a project.
| What you send us | Minutes | Cheapest way to buy it | Cost |
|---|---|---|---|
| One three-hour show | 180 | 180 min drawn from a $24 / 5,000-min pack | ~$0.86 |
| A weekday three-hour show, one month (20 shows) | 3,600 | one $24 / 5,000-min pack, 1,400 min spare | $24 |
| Six hours of original programming a day, 30 days | 10,800 | 2.2 × the $24 / 5,000-min pack | ~$52 |
| A full 18-hour broadcast day, 30 days | 32,400 | 6.5 × the $24 / 5,000-min pack — talk to us | ~$156 |
There is no free tier and no daily free allowance — you buy minutes, not seats, and unused credits stay yours. Full pricing →
Put one programme through and look at the chapter list, the speaker labels and the search box. That tells you more about whether this fits your station than any page can — and it's one file, not a project.
No. Whipscribe is asynchronous — you send recorded audio and the transcript comes back once the job has run. We never connect to your stream, there is no always-on listener, and nothing here alerts on audio as it airs. If you need live captioning on transmission, that is a different kind of product.
Up to 10 hours. A three-hour show is one job; a full broadcast day goes in as the hourly files your logger already writes — which is also easier to search later, because the file name carries the hour.
Diarization is on by default and separates each voice into its own labelled turn. It marks speakers as distinct rather than naming them — you rename once and the label carries through the transcript and the exports.
Strong on clean studio speech, weaker on a bad phone line, crosstalk, or a name it has never seen. Machine transcription mishears proper nouns and figures first, which is exactly what a credit or a quote turns on — so click the timestamp and hear it before you publish it.
Over 100 languages, and the transcript stays in the language it was spoken in. Useful for multilingual stations and for community programming inside an otherwise English schedule.
Whipscribe runs faster-whisper and whisperX on private infrastructure. Your audio is never handed to a third-party transcription service and is never used to train a model.