For compliance, programming & comms teams

Know what went out — by searching it, not re-listening to it

Your as-broadcast recordings become text you can search by word, by speaker and by hour. Type a name and see every time it was said across months of output, each hit a timestamp you click to hear the audio in context.

Transcribe an hour of output → Talk about archive volume

Recordings you send us · not a live listener · no alerting · speaker labels · up to 10 hours per file

What this is: the record afterwards, not a listener on the line

Whipscribe is asynchronous. You send us recordings — the hourly as-broadcast files your logger already writes, a show export, a rendered clip — and text comes back once the job has run. We do not connect to a stream, there is no always-on capture, and nothing here alerts you on audio as it airs.

So this does not replace whatever captures your output. It sits behind it: your recordings stay the authoritative material, and the transcript is the index that makes them findable. If you need a word flagged the moment it is spoken, that is a live monitoring product and this isn't one.

An archive you can't query is a filing cabinet with no drawers

Broadcasters keep what went out. Almost nobody can search it. One 18-hour day is 1,080 minutes of speech; a month of those is 32,400; a year is roughly 6,500 hours sitting as audio files named after timestamps. When a question arrives — did we say that, when did we cover it, who was on air — the only tool is a person with headphones and a guess about which hour to start.

  1. 01

    Send the hours you care about

    Upload files, submit them through the API one call each, or give us read-only access to the bucket or Drive folder your logger writes into. Up to 10 hours per file, so an hourly logger file is one job and a whole day is 18 or 24 of them.

  2. 02

    Every hour becomes timestamped, speaker-labelled text

    Diarization splits the audio into labelled turns, so the presenter, the guest and the caller are separated in the record rather than merged into one voice. Timestamps run the whole way through the file.

  3. 03

    Search for the word, get the hour and the second

    Type a person, a brand, a place, a claim. Search returns every line containing it with its timestamp, so you read the surrounding lines and judge relevance before you listen to anything. Accents and case are ignored.

  4. 04

    Or ask the archive a question

    Ask in plain language and get an answer grounded only in that transcript, with citations. Every citation is a link: click it and the transcript jumps to that line while the audio plays from it — so the answer is a route to the evidence, not a substitute for it.

    “When did we discuss the roadworks, and who raised it?”
  5. 05

    Keep it, export it, hand it on

    Export TXT, DOCX, JSON, SRT or VTT with timings intact. Attach the passage and its timecode to an internal note, a response, or a report — so whoever reads it next can go and hear it themselves.

The questions a searchable record answers

All of these are the same operation underneath: find the words, then hear them.

“When did we mention this?”

A name, a company, a claim, a piece of music. Search across every hour you've transcribed instead of guessing which show it was on and scrubbing from there.

search · every hit timestamped

“What exactly was said?”

A complaint or an internal query usually paraphrases. The transcript gives you the verbatim line and the audio behind it, which is what settles the disagreement.

clickable timestamp → audio

“Who said it?”

Speaker labels separate presenter from guest from caller, so attribution comes out of the record rather than out of somebody's memory of the shift.

diarization on by default

“How often, and where?”

Count occurrences across a week or a month of output — coverage of a topic, mentions of a sponsor, how a story travelled across dayparts — from text rather than sampling by ear.

JSON export for tallies

What we will and won't say about compliance

A searchable text record makes it much faster for a person to locate a segment, read what was said around it, and hear the audio in context. That is genuinely useful when you have to answer a question about your own output under time pressure. That is the whole claim.

We do not claim that using Whipscribe satisfies any regulator's rules, licence condition, code, or retention requirement, and we make no statement about any specific regulatory regime. Your recording, retention and reporting obligations remain yours; your own recordings remain the authoritative material; machine transcription mishears names, figures and technical terms, so verify the audio before you rely on a line. Nothing here is legal advice. For jurisdiction-by-jurisdiction notes written up separately, see the broadcast hub.

Feeding an archive in, one file per call

An API key, one submit per file, then poll. A day of hourly logger files is a loop; a back catalogue is the same loop pointed at a longer list.

# submit every hour of yesterday's as-broadcast audio
for f in 2026-08-05-*.mp3; do
  curl -s https://whipscribe.com/api/v1/transcribe \
    -H "X-API-Key: $WHIPSCRIBE_KEY" \
    -F "file=@$f" -F "diarize=true" -F "source=api"
done
# each call returns {"job_id":"…","status":"queued"}

# list what you've submitted, then pull a finished transcript
curl -H "X-API-Key: $WHIPSCRIBE_KEY" "https://whipscribe.com/api/v1/jobs?limit=100"
curl -H "X-API-Key: $WHIPSCRIBE_KEY" "https://whipscribe.com/api/v1/jobs/<job_id>/result?format=json"

Your bucket, read-only

Prefix-scoped access to your own S3 (or S3-compatible) bucket or a Google Drive folder, so whole folders of logger output get transcribed without anyone opening a browser.

MCP server

Whipscribe ships an MCP server, so an assistant can submit files, search transcripts and pull quoted passages as part of a review workflow rather than a separate tool.

whipscribe.com/mcp

Chrome extension

For output that isn't in your own archive — a catch-up player, another broadcaster's clip, an embedded recording — the Whipscribe extension sends browser audio into the queue.

API reference

Every field, status and export format, including how jobs are scoped to your key and how results are fetched. For whoever wires this to the logger.

What it costs to keep a searchable record

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's just under half a cent a minute — so the decision stops being “which hours can we afford to transcribe” and becomes “how much of the schedule do we want indexed”.

What you indexMinutesCheapest way to buy itCost
One 18-hour broadcast day1,0801,080 min drawn from a $24 / 5,000-min pack~$5.18
One flagged hour a day, for a month1,800one $12 / 2,000-min pack, 200 min spare$12
18-hour days, every day for a month32,4006.5 × the $24 / 5,000-min pack~$156
Round-the-clock, 24 h a day for a month43,2008.7 × the $24 / 5,000-min pack — talk to us~$207
One 18-h day
1.1k
1 h/day, a month
1.8k
18 h/day, a month
32.4k
24 h/day, a month
43.2k
Minutes per month, to scale. Most teams start with the flagged hour and widen once the search box has earned it.

There is no free tier and no daily free allowance — you buy minutes, not seats, and unused credits stay yours. Backfilling a year of archive is a volume conversation: contact sales → · full pricing →

Test it on the hour you'd struggle to find

Pick a logger file where you know something was said, put it through, and search for it. That single test tells you whether a searchable record would have saved you the afternoon you spent last time.

Questions

Do you monitor our stream live?

No. Whipscribe works on recordings you send us — the hourly as-broadcast files your logger writes, or a show export. There is no live listener, no always-on capture, and no alerting on audio as it airs. What you get is the record afterwards, in text you can search.

Does this make us compliant with our obligations?

No, and we won't claim it does. This is a transcription tool that makes recordings you already keep searchable, so a person can find a segment fast and hear it in context. Your obligations remain yours, your recordings remain the authoritative material, and nothing here is legal advice.

How do we find when something was mentioned?

Search returns every line containing your phrase with its timestamp, so you read the context before you listen. Or ask the transcript a question and get an answer grounded only in that transcript, with citations you click to play the audio from that exact moment.

Can it separate speakers so we know who said it?

Yes — diarization splits the audio into labelled turns, so presenter, guest and caller are distinguished. Labels mark speakers as distinct rather than naming them; you rename once and it carries through.

How accurate is it on broadcast audio?

Strong on clean studio speech, weaker on phone lines, crosstalk, music beds and unfamiliar proper nouns — which are exactly the words a query often turns on. Treat the transcript as the index and the audio as the record: click the timestamp and listen before you act on a line.

Where is the audio processed?

On private infrastructure running faster-whisper and whisperX. Your recordings are never handed to a third-party transcription service and are never used to train a model.