Dictation and clinical notes: a transcription workflow

Clinicians dictate, the API turns the audio into text, and the clinician reviews and corrects it before anything enters a record. This page describes that workflow; it is not a compliance statement.

Get an API key Quickstart Supported languages

Last updated: · Markdown

The workflow

  1. The clinician records a dictation or, with the patient's consent, a visit.
  2. Your system uploads the file with POST /upload and starts a job.
  3. The transcript returns as JSON segments with timestamps.
  4. The clinician reviews and edits the text against the audio, then signs the note in your record system.
{
  "audio_url": "<fileUrl from POST /upload>",
  "model": "oracle_pro",
  "language": "en",
  "output_format": "json"
}

Settings that help

  • oracle_pro or oracle_max for fast speech, accents and medical vocabulary.
  • include_diarization for visit recordings with a clinician and a patient.
  • Set language explicitly; mixed-language visits are common in many countries.
  • Keep word timestamps so the reviewer can jump from a word to the audio.

Your responsibilities

Medication names, doses, numbers and negations are where machine transcripts go wrong. A qualified person must review every transcript before it is used for care. You are responsible for patient consent, for where audio and text are stored, and for the privacy and medical-record rules that apply to you. KolWrite makes no regulatory or certification claim for this API. Its models page states that audio submitted through the API is not used to train its models. For contract terms, contact KolWrite Enterprise.

Questions and answers

Is the KolWrite API certified for medical use?

This page makes no certification or compliance claim. Check your own obligations before sending patient audio.

Can it transcribe medication names correctly?

Often, but not always. Review every name, dose and number against the audio.

How fast is a dictation transcribed?

Jobs are asynchronous: poll the job or use a webhook. Processing time depends on the recording length and the queue, so measure it with your own dictations.

Related pages