Whisper · speech-to-text

Whisper integration: speech-to-text in your app

Whisper integration means an audio file or a recording from your app becomes text in your database — with a timestamp and a user on it. It is one of the AI features that actually saves time in the field: a tradesperson dictates on the way from the job. A technician closes a call. Clinic-adjacent notes, if you have a legal basis. The text is a draft until a human hits approve. Whisper mishears jargon. A “15 mm” that becomes “50 mm” must not land straight in the quote. Live transcription is a different project from “after recording”. GDPR: delete the audio after X days if you only need the text.

Whisper is OpenAI’s speech-to-text. It is not the same socket as GPT drafts. The other direction — text to speech — is ElevenLabs. AI in the field, not on the homepage.

Design it as a draft

Jargon, measurements and SKUs are where Whisper fails. A wrong millimetre in a quote costs more than the hour you saved on dictation. The approve button is the product.

Udkast

Prompt

Menneske godkender før udsendelse

The text waits. It is not “the case” until someone has read it.

Operations

  • File size and timeout — a 40-minute recording is a different job from 40 seconds
  • Delete audio after X days if you only need the text
  • Language set to da when you know it — fewer hallucinations on code names
  • A legal basis before clinic-adjacent or staff conversations are recorded
1

Record

The app stores the file with you, not in a random cloud folder.

2

Transcribe

Whisper runs with language da when you know it.

3

Approve

A human fixes jargon before the text is truth.

4

Delete audio

If you only need to search the text, the audio should not live forever.

After recording is the robust track

What to do now

Record one real close-out from the field (with consent). If the text is usable after five minutes of edits, you have a case. Send us the kind of recording you have.

FAQ about Whisper integration

Live or after recording?

After is simpler and more robust. Live is a different project.

May we record customers?

Only with a legal basis and clear notice. Clinics, employment and secret rooms are separate rules. That is not an API question.

How long should the audio be kept?

If you only need to search the text: delete the audio after a short period you can explain.

Does Whisper fix millimetres and SKUs?

No. That is why approval is first-class. A 15 that becomes 50 must not reach the quote.

Turn field notes into searchable text