Voice-to-Text for Dental Notes: UK Practice Guide
Voice-to-text for UK dental notes means producing clinical records by speaking rather than typing — either as raw dictation that the clinician transcribes into a template, or via an AI scribe that records the appointment and structures the note automatically. Both involve speech-to-text technology, GDPR considerations for processing patient identifiable audio, and accuracy variation by clinical environment.
Voice dictation has been around dentistry for years. AI scribes are the newer evolution. Both promise hands-free note-taking but both face the same fundamental challenges: clinical environment noise, regional accents, and the GDPR surface of processing patient audio. This guide covers what actually works in UK practice.
Three voice-based approaches
- Plain dictation (Dragon Dental, native macOS/Windows speech-to-text): you speak, words appear on screen, you correct. The clinician produces the structured note structure manually.
- Voice-driven template: you speak field values; the app maps them to template fields (e.g. "tooth UR6, MOD, composite, A2 shade"). Hybrid of voice and structured templates.
- AI scribe: you speak naturally during the appointment; the AI listens, transcribes, then structures the transcript into a clinical note format that you review.
Hardware and environment
Voice accuracy depends on the microphone and the environment more than on the software. Bluetooth headsets (Plantronics, Jabra Engage) outperform built-in laptop mics in clinical settings. Lapel mics work well but introduce hygiene considerations between patients.
Surgery noise (suction, ultrasonic scaler, compressor) significantly degrades transcription accuracy. Some AI scribes use noise-cancelling and speaker isolation; others don't. Test the vendor in YOUR surgery with YOUR equipment running.
Accuracy benchmarks
| Scenario | Plain Dictation | AI Scribe |
|---|---|---|
| Native English, quiet room | 95-98% | 93-97% |
| Regional accent, quiet room | 85-92% | 85-90% |
| Native English, surgery noise | 75-85% | 80-88% |
| Regional accent + surgery noise + mask | 65-78% | 70-85% |
| Multiple speakers (dentist + nurse + patient) | 60-75% | 75-88% |
Real-world accuracy in busy UK general practice typically lands at 80-90%. The remaining errors are usually clinical specificity (wrong tooth number, wrong material, wrong dose) — exactly the errors that matter for compliance and safety.
GDPR considerations specific to voice
Voice processing has unique GDPR considerations because:
- Patient speech is captured (consent discussions, social history, dental concerns) — identifiable special category data.
- Audio files are larger than text and have longer retention windows by default.
- Voice patterns are themselves potentially identifiable (biometric considerations).
- AI providers often retain transcripts/audio for "model improvement" — check vendor terms carefully.
- Cross-border transfer of audio to US-based AI providers requires Standard Contractual Clauses.
Best practice: obtain explicit patient consent before audio recording, document the consent in the note, choose vendors with documented "no model training" clauses, prefer UK/EEA data processing where possible.
When voice works vs when it fails
- WORKS: long consultations with detailed patient history (oral medicine, orofacial pain, paediatric anxiety management) — voice captures detail that templates would miss
- WORKS: clinicians with dexterity or vision impairments where typing/tapping is slow
- WORKS: locums or hospital-based clinicians moving between sites where template setup is impractical
- FAILS: high-volume routine general practice — voice review/edit time exceeds time to tap a template
- FAILS: noisy clinics with poor mics or untrained clinicians who speak unclearly
- FAILS: practices with strong regional accents or non-native English clinicians (accuracy drops below practical threshold)
- FAILS: complex periodontal charting (BPE, 6-point pockets) — structured templates dramatically outperform voice for grid data
Setup and training
- Invest in a quality microphone — Bluetooth headset or quality desk mic (typically £80-£200).
- Train each clinician on the software's vocabulary — dental terminology (FDI tooth numbering, material brand names, technique acronyms) often needs custom vocabulary uploads.
- Establish a consistent dictation style — "tooth UR6 mesial-occlusal composite, A2 body, A3 cervical" beats free-form narrative for AI scribe accuracy.
- Test in real conditions during the trial period — running suction, masks, regional accent dentist, busy clinic.
- Plan for review time — voice approach always requires clinician review and correction. Budget 2-5 minutes per note.
- Have a fallback — when voice fails (broken mic, very noisy day), what's the backup? Templates work as standalone fallback.
Where Nosht fits
Nosht is structured templates first — for the high-volume routine appointments where templates beat voice on speed. For practices that also want voice for complex consultations, pair Nosht (templates) with a dedicated AI scribe (complex cases). Combined cost is typically less than an AI-scribe-only solution because Nosht handles the 80% of appointments where templates win.
Start with structured templates
Nosht — £5/clinician/month (beta). Add voice later if specialist consultations justify it.
Try Nosht freeFrequently asked questions
Is voice-to-text accurate enough for UK dental notes?
In ideal conditions (native English, quiet room, quality mic): yes — 95%+ accuracy is achievable. In real UK dental practice (regional accents, surgery noise, masks): typically 80-90%. Always requires clinician review. Remaining errors often involve clinical specifics (tooth numbers, materials, doses) which matter for safety and compliance.
Do I need patient consent for AI scribes?
Best practice and increasingly expected: yes, explicit consent before recording. Document in the note ("patient consented to AI-assisted note-taking"). Refusal must not affect care. Some practices include in general consent forms; others ask per appointment. Either is acceptable if documented.
Which dental AI scribes work in the UK?
As of 2026, UK-available AI scribes for dentistry include Kiroku (UK-built, dental-specific), Heidi (UK + AU), DentalScribe (US-based with UK availability), and several emerging entrants. Compare on: UK GDPR compliance, dental vocabulary training, integration with your PMS, per-clinician pricing, accuracy in masked/noisy conditions, and trial flexibility.
Can I use Dragon NaturallySpeaking for dental notes?
Dragon Dental Edition exists and is widely used. It's plain dictation (not an AI scribe — it doesn't auto-structure notes). Best paired with a structured template or for clinicians who prefer typing-by-voice into a free-text field. Slower than AI scribe for unstructured notes, faster than typing for trained users.
Will my dental nurse need to use voice too?
Depends on workflow. AI scribes typically record dentist speech only (or dentist + patient). Plain dictation is dentist-only. Nurse remains free to assist patient and handle chairside tasks. Some practices use nurse for tap-completion of structured templates while dentist focuses on procedure — but this requires the nurse to know clinical terminology.
What's the cheapest voice setup?
Apple's built-in macOS / iOS speech-to-text is free and acceptable for unstructured dictation if you have a Mac/iPad. Microsoft's Windows Speech Recognition is similar for PC. For dental-specific accuracy, Dragon Dental costs ~£600 one-time plus updates. AI scribes (Kiroku, Heidi) are subscription, typically £30-60/clinician/month.