More than
two voices

Demonstrations of documentation tools tend to feature one clinician and one articulate patient in a quiet room. Real Australian practice looks different. A parent gives the history for a febrile toddler. An adult child explains what has changed in their mother's memory. A support worker reads from a medication chart. An interpreter on speakerphone relays a patient's answers from another language. Two of these can happen in the same consult.

The extra voices are often the clinically important ones. Collateral history is central to paediatrics, aged care, cognitive assessment and much of mental health, so a documentation tool that only coped with tidy two-person audio would fail exactly where good notes matter most. The practical questions are how ambient scribes handle several speakers, what happens when an interpreter sits between the clinician and the patient, and what the reviewing clinician should check before the draft becomes part of the record.

Almost everything below applies to any ambient AI scribe. One section near the end describes how aurii approaches these consults specifically, and the review habits apply regardless of which tool a practice uses.

An ambient scribe starts with the full audio of the consult and transcribes everything said in the room. Modern systems also separate speakers, using the acoustic signature of each voice to mark where one person stops and another starts. The drafting step layers context on top of that: the person asking about medication changes and examination findings is the clinician, the person answering in the first person is usually the patient, and a voice that refers to the patient by name or as mum or dad belongs to someone else in the room.

That combination lets a draft attribute information sensibly. A well-constructed note distinguishes what the patient reported from what a relative observed and from what the clinician found on examination, because those three things carry different clinical weight. Speaker separation makes the distinction possible, and the clinician's review confirms it landed correctly.

Overlapping speech is the hardest case for any system. When two people talk across each other, fragments can blur in the transcript and the draft may compress that moment. Consults do not need to be stage-managed to avoid this, but it is worth knowing that the messiest thirty seconds of audio is where a draft is most likely to be thin, and checking that part of the note against memory takes a moment.

Interpreted consults are routine in Australian healthcare. Practices reach phone interpreters through TIS National, which the Australian Government funds for medical practitioners in private practice, or book on-site interpreters credentialled by NAATI for scheduled appointments. The consult audio then carries two languages, with the interpreter's English rendering sitting between the patient's words and the clinician's ears.

For documentation purposes, the interpreter's English is the working account of what the patient said. It is what the clinician heard, weighed and acted on, so it is the right basis for the note. The draft should read the way the clinician experienced the consult, with the history attributed to the patient and identified as taken through an interpreter. Reviewing these drafts means checking the clinical content against the interpreter's rendering, since that version is the one the clinician can verify from memory.

Two record-keeping habits matter here. First, note that an interpreter was used, including the language, the service and the interpreter identifier where one is given, because it explains how the history was obtained and helps the next clinician arrange the same support. Second, seek recording consent through the interpreter at the start of the consult so the patient genuinely understands what is being asked, and tell the interpreter the consult is being recorded as well.

A third voice in the room usually means collateral history, and collateral history is often the point of the consult. The parent describing a child's breathing overnight, the spouse who noticed slurred speech and the carer with the medication list each contribute information the patient cannot give. A scribe that captures the whole room captures this material along with everything else.

Attribution is the thing to protect. A note that records a relative's observation as the patient's own report misstates the history in a way that can matter later, particularly in cognitive assessment or safeguarding contexts. During review, check that the draft credits collateral history to the person who gave it and that the patient's own account is identifiable as theirs.

Consent stays simple. The patient consents to the recording of their consult, and anyone else present who will speak should be told the consult is being recorded, which one plain sentence at the start achieves. If a family member asks for something they said to be kept out of the note, that is a clinical judgement the clinician makes at review, exactly as it would be with typed notes.

Australian consult rooms hold a wide range of accents on both sides of the desk. A substantial share of the medical workforce trained overseas, and patients bring every accent in the country. Older dictation software dealt with this by training a voice profile for a single user, which never helped with the patient's side of the conversation. Ambient scribes are built on speech recognition trained across a broad range of accents and speaking styles, and they transcribe accented English as ordinary input without any per-user training.

Medical vocabulary is a separate challenge from accent. Drug names, anatomical terms, eponyms and local place names are where general-purpose transcription slips, and clinical systems are tuned for exactly this vocabulary. The residual risk sits in predictable places: drug names that sound alike, numbers and units in doses, and negatives, where the difference between chest pain and no chest pain is one short word spoken quickly.

In practice, accent alone rarely determines the quality of a draft. The checks worth keeping are vocabulary checks. Read every drug name, dose and negative finding before signing, because those are the errors that matter clinically when they occur.

Nothing an ambient scribe produces becomes part of the clinical record until a clinician has reviewed it and signed it. That discipline carries the whole workflow, and in a multi-speaker consult the review gains a few extra targets. The clinician's memory of the consult is the reference, and the checks take seconds once they are habitual.

• Attribution: statements credited to the patient were said by the patient, and collateral history is labelled with its source.
• Interpreter details: the note records that an interpreter was used, the language, and the service or interpreter number.
• Medication: every drug name, dose, unit and frequency matches what was actually decided in the room.
• Negatives: denied symptoms and normal findings match your memory, since a dropped word reverses their meaning.
• Names and places: patient names, family names, suburbs and facility names are spelled correctly.
• Busy moments: the parts of the consult where people spoke over each other are captured, and anything material that was lost is added by hand.