Talk through the visit.
Get the note.
GPU-accelerated clinical transcription that separates clinician from patient, structures the conversation into a SOAP draft, and runs on hardware you control.
And how long has the swelling been there?
About three weeks now, maybe a bit longer.
Any change in the medication since we last met?
Three weeks of swelling, gradual onset. No change to current medication.
Localised oedema noted on examination.
Working impression documented for review.
Investigations ordered; follow-up arranged.
Draft — requires clinician review and sign-off
Documentation is the part of the day nobody trained for
The clinical work in a consultation ends when the patient leaves. The typing does not. For most specialists the note is written later, from memory and shorthand, at a point in the day when the details have started to blur — which is how documentation becomes both the slowest task and the least accurate one.
AI Transcriber listens to the encounter itself. Because it separates who is speaking, the transcript is not an undifferentiated wall of text but an attributed conversation, and a conversation with turns in it can be structured. What comes back is a SOAP draft written from what was actually said, with the medications, findings and problems pulled out alongside it.
It is a draft, and it is meant to be read. Nothing it produces should enter a chart without a qualified clinician reviewing and signing it — the point is that reviewing a draft takes a fraction of the time that writing one does.
Under the hood
How the note gets made
- 1
Capture
The encounter is recorded — in the room, or from a virtual visit.
- 2
Transcribe
GPU-accelerated speech recognition turns audio into text at speed, on your own hardware.
- 3
Attribute
Diarization separates clinician turns from patient turns so the transcript has structure.
- 4
Structure
The attributed conversation is assembled into SOAP sections, with clinical entities extracted.
- 5
Review
You read, correct and sign. The draft is never the final word.
Where the audio lives is your decision
The deployment we recommend runs on your own GPU hardware, so recordings and transcripts never leave the practice network. That is a much simpler thing to explain to a patient — and to a privacy officer — than any assurance about a third party's data handling.
Features
What it does
Ambient capture
Record the encounter as it happens rather than dictating afterwards. The conversation is the input; nobody has to narrate for the machine.
Speaker diarization
The transcript knows who spoke. Clinician and patient turns are separated, which is what makes reliable structuring possible downstream.
SOAP structuring
Subjective, objective, assessment and plan, assembled from the conversation — a draft note in the shape you already write in.
Clinical entity extraction
Medications, findings, problems and negations pulled out as structured data alongside the note, so the chart gets more than a block of prose.
Runs on your hardware
GPU-accelerated and deployable on-premise. Audio and transcripts can stay entirely inside your network — the shortest possible answer to "where does the recording go?"
Review before it counts
Everything produced is a draft for a qualified clinician to read, correct and sign. It saves the typing, not the judgement — and the interface is built around that assumption.
Technology
FAQ
AI Transcriber questions
Where does the audio go?
Where you decide. AI Transcriber is built to run on hardware you control, using your own GPU, so patient audio and transcripts need never leave your network. That is the deployment we recommend for practices that want the shortest possible answer to the question.
Does it know who is speaking?
Yes. Speaker diarization separates clinician from patient, so the transcript attributes each turn rather than producing an undifferentiated wall of text — which is what makes automatic SOAP structuring possible.
What does the output look like?
A structured clinical note in SOAP form, with clinical entities — medications, findings, problems — extracted alongside it, plus the raw diarized transcript if you want to check anything against the source.
Does it replace my review of the note?
No, and it should not be used that way. Everything it produces is a draft for a qualified clinician to review, correct and sign. It saves the typing, not the judgement.
Does it work with PocketEMR?
Yes — it is the transcription engine behind PocketEMR's AI voice notes. It can also run standalone alongside another EMR.
See it on a real consult
The fastest way to judge clinical transcription is to watch it handle a consultation that looks like yours. Book a walkthrough and we'll do exactly that.