Voice documentation

Episode → Notes → Clinical documentation → open an unsigned note you wrote → Record this visit

The voice panel sits inside the note editor, above Document this visit. It appears only on your own unsigned note. When voice cannot run, you see a disabled microphone and the reason:

• Microphone permission. Your browser asks once for permission to use the microphone on this site. If you block it, recording stops with The microphone could not be opened. Check the browser's permission for this site. Allow the microphone in the browser's site settings and try again.
• Browsers and devices. We have not published a list of tested browsers or devices. Try a short recording on the computer and microphone you will use before relying on it.
• Choosing a microphone. The consent step lets you pick the microphone. The choice is remembered on that computer.
• The patient AI agreement. Where your clinic has added it, Record this visit and Record this evaluation run only for a patient who answered yes to audio recording. Dictate a summary does not check it.
• Everyone present. Tell everyone in the room that the visit will be recorded and transcribed, that no audio is kept, and that they can stop it at any time.

1. Select Record this visit.
2. Tick I told everyone present that this visit will be recorded and transcribed. and Everyone present agreed. Choose whether agreement was given verbally or in writing, the microphone, and the language spoken. Cancel backs out without recording.
3. Select Record. Text appears under Transcript - speakers not identified, not verified as each part is transcribed.
4. Use Pause and Resume as needed; nothing is captured or sent while paused, and the same consent covers the resumed recording.
5. If the patient changes their mind, select Patient withdrew consent. Recording stops and everything transcribed is deleted.
6. Select Stop when the visit ends.
7. Correct mistakes with Fix a line; your wording is what is used, and the original stays on file.
8. Select Add to visit notes. The transcript is added to Today's visit notes under a heading marking it as unverified speech-to-text. Then use Draft SOAP note or write the note yourself.

Other options after stopping: Copy transcript, Review transcript (rule-based checks that only display results), and Discard, which deletes the transcript.

Dictation is for the therapist alone, after the visit: a short spoken recap. Because only you speak, the AI SOAP draft can use it directly as Your dictated summary (speech-to-text, today).

1. Select Dictate a summary.
2. Tick I am alone - nobody else is in the room while I dictate. If someone else is present, tell them and record their agreement instead.
3. Select Start dictating, speak, then Stop. Discard dictation deletes it while recording.
4. Check the text under Your dictation - speech-to-text, not verified and use Fix a line where it misheard you.
5. Select Use in draft, then Draft SOAP note, and review every section before signing.

Every dictation you have not discarded is a source for this note's draft, whether or not you select Use in draft. To keep one out, select Discard. Word-rule matches in a dictation are shown for your attention only and are not escalated.

On an initial evaluation, re-evaluation, progress report or discharge summary, record with Record this evaluation (same consent and controls as above), stop, then select Propose entries from the transcript. A rule-based reader - not AI - looks for:

• answers to the guided screening questions for the body site on record (PT, OT and multidisciplinary episodes only; none for SLP);
• range-of-motion and strength values with joint, motion, side and method;
• mobility and self-care assistance levels.

Each proposal shows the words it came from and is either explicit or uncertain. Under Proposed from the recording - not recorded until you accept, for each one:

Accepting or correcting records the entry straight away against the episode and visit. The note's text changes only when you select Add accepted entries to the note, which inserts what is recorded for you to review and edit before signing.

• Nothing is proposed for diagnoses or impressions.
• A measurement heard without its side or method is counted, not proposed.
• Silence is never read as No.
• A red-flag question recorded Yes cannot be cleared from a recording; reassess the patient and correct it yourself. While an urgent alert is unacknowledged, no red-flag question can be answered No from the recording.

Speech is captured in parts of about 15 seconds. Each part is sent and transcribed while you carry on speaking; silent parts are not sent. A recording can run to about 100 minutes, and the consent covering it lasts four hours. If your browser closes mid-recording, reopening the note offers to Resume it (while the consent is still active), End it or Discard it.

• No audio is kept. Each part is held in memory while it is transcribed and then discarded.
• Transcripts are encrypted, as are your corrections and the words behind each proposal.
• Retention. Transcripts are erased after your organisation's retention period - 90 days unless your administrator changes it - by a daily clean-up. Discarding, or the patient withdrawing consent, deletes them at once.
• What stays. Text you added to a note, and entries you accepted or corrected, are part of the record and stay when the transcript is erased.
• Audit. Starting, stopping, withdrawing consent and viewing a transcript are recorded in the audit trail.
• The consent you record here is for this visit only. It is separate from the patient AI agreement form.

An administrator turns Voice on or off for the organisation in the Voice settings. Your organisation's AI switch (Settings, AI) does not affect voice; it controls AI drafting.

• Speech-to-text availability has been intermittent; there are no confidence scores on transcribed text.
• Accuracy has been tested on synthetic speech only, not on real clinic audio, accents or background noise.
• A failed part can cause the rest of that recording to be refused (above).
• Plain-language finding capture and the measurement reader in Quick Capture read typed text, not audio.