AI Voice Dictation vs Dragon Medical: Why We Built Our Own

We run a dental practice in Darwin. Like most clinics, our dentists spend a surprising amount of time typing — clinical notes, referral letters, treatment plans, emails. It's not glamorous work, and it adds up fast. A clinician who types for 15 minutes a day is losing over an hour a week that could be spent with patients.

Voice dictation is the obvious fix. Speak instead of type. But when we looked at the options, we found a market that charges a premium for something that modern open-source AI can do for free — or close to it.

So we built our own. Here's why, and how it compares.

The market: what's out there

If you're a healthcare professional looking for voice dictation in Australia, you'll typically land on one of these:

Dragon Medical One (Nuance / Microsoft)

The incumbent. Dragon has been the gold standard for medical dictation for over 30 years. It has an enormous medical vocabulary, understands clinical terminology across specialties, and integrates deeply with hospital EMR systems. It's genuinely excellent at what it does.

The catch? At time of writing, Dragon Medical One costs roughly ~$99–199/month per user, depending on your reseller and region. For a practice with three clinicians, that's $300–600 every month. It's cloud-based, so your audio is sent to Microsoft's servers for processing. And it works best inside supported applications — if you want to dictate into a random text field or a niche practice management system, you may hit friction.

Heidi Health

An Australian company building AI clinical documentation. Heidi listens to the patient consultation and generates structured clinical notes automatically. It's purpose-built for healthcare, understands AHPRA requirements, and is designed to slot into the clinical workflow.

Heidi's paid tiers run roughly ~$69–99/month per clinician at time of writing. It's a different product to general dictation — it's specifically for clinical notes from consultations, not general-purpose "speak and it types." If clinical note generation is your primary need, Heidi is worth a serious look. But it's cloud-based, and it's not designed for dictating an email or a referral letter into any arbitrary app.

Whisperflow and similar

Newer, lighter tools that use OpenAI's Whisper model (or similar) to provide dictation at a fraction of the cost — roughly ~$8–15/month. These are more affordable, but they're still cloud-based, meaning your audio leaves your machine.

What we built: VoxInk

VoxInk is push-to-talk dictation. Hold a key, speak, release, and the transcribed text appears wherever your cursor is. Any app, any text field — your practice management system, an email draft, a Word document, a web form. It doesn't matter.

It runs entirely on your local machine using Whisper-class speech recognition models. No audio leaves the computer. No cloud. No subscription. No per-user fees.

We built it because we needed dictation that:

• Works in any application, not just supported ones
• Keeps patient data completely local — no audio sent to external servers
• Doesn't cost $200/month per clinician for something a local AI model handles perfectly well
• Runs on both macOS and Windows, since our practice uses both

Head-to-head comparison

Here's how the options stack up across the features that matter most in a clinical setting:

Being fair: where Dragon and Heidi win

We're not going to pretend VoxInk is better at everything. That wouldn't be honest.

Dragon Medical has three decades of medical speech recognition behind it. If you're a surgeon dictating highly specialised operative reports with complex anatomical terminology, Dragon's vocabulary depth is genuinely hard to beat. It has been trained on millions of hours of medical dictation across every specialty. For large hospital environments with deep EMR integration requirements, Dragon is still the standard for good reason.

Heidi Health does something fundamentally different to dictation. It listens to an entire patient consultation and produces structured clinical notes — SOAP format, referral letters, the lot. If your main pain point is documenting consultations in real time, Heidi is purpose-built for that job and does it well. It's Australian, it understands our regulatory context, and it's designed by people who understand clinical workflows.

If you need the deepest possible medical vocabulary and don't mind the cost, Dragon is excellent. If you need AI-generated clinical notes from consultations, Heidi is excellent. We're not here to argue otherwise.

The real question: what are you actually dictating?

Here's what we noticed in our practice. The vast majority of dictation — easily 90% — isn't complex medical terminology. It's things like:

• "Please book Mrs. Johnson for a clean and check-up next Thursday at 2pm."
• "Referred to Dr. Singh for assessment of lower right wisdom tooth. OPG attached."
• "Patient presented with sensitivity on the upper left six. Cold test positive. Advised monitoring."
• "Hi Sarah, just following up on the lab case for Mr. Patel. Can you confirm the shade?"

None of that requires 30 years of medical vocabulary training. A modern Whisper model handles it with near-perfect accuracy. The words are common. The sentences are short. The context is straightforward.

You don't need a $199/month specialised engine to transcribe "please book Mrs. Johnson for a clean next Thursday." You just don't.

For the 10% of dictation that does involve complex terminology — detailed operative notes, specialist referrals with specific anatomical terms — VoxInk still handles it well through domain-specific prompting. We feed the model context about dental terminology, and it gets the right words. Is it as bulletproof as Dragon on rare surgical terms? Probably not. But for general dental practice, it's more than accurate enough.

The cost argument over time

Let's say you're a dental practice with three clinicians. Here's what dictation costs over three years:

• Dragon Medical: 3 users × ~$149/month (midpoint) × 36 months = ~$16,092
• Heidi Health: 3 users × ~$84/month (midpoint) × 36 months = ~$9,072
• VoxInk: 3 users × $0/month × 36 months = $0

That's $9,000–16,000 in savings over three years — on dictation alone. And VoxInk has no per-user pricing, so when you hire a fourth clinician, the cost stays the same: zero.

We covered the full cost picture across all practice software in our post on the real cost of AI tools. Dictation is just one piece of the puzzle, but it's a clear example of where local AI has caught up to expensive cloud services.

Privacy: the quiet advantage

Every time you use a cloud-based dictation service, your audio — which may contain patient names, conditions, treatment details — is sent to an external server for processing. For Dragon, that's Microsoft's cloud. For Heidi, it's their processing infrastructure.

Both companies have privacy policies and compliance frameworks. We're not suggesting they're being reckless with your data. But the simplest way to guarantee patient data privacy is to never send it anywhere in the first place.

VoxInk processes everything on your local machine. The audio never touches a network. There's no server to breach, no third-party data processing agreement to review, and no question about where your patient data ends up. It stays on the machine where it was spoken.

For practices that take data sovereignty seriously — or that simply prefer the peace of mind — local processing is a meaningful advantage.

Who should use what

Our honest recommendation:

• Large hospitals with deep EMR integration needs and complex multi-specialty dictation: Dragon Medical is still the gold standard. The cost is justified by the depth of integration and vocabulary.
• Clinicians whose main bottleneck is documenting patient consultations: Heidi Health is purpose-built for exactly this. It's not just dictation — it's AI clinical documentation.
• Everyone else — GPs, dentists, allied health, small practices: A local dictation tool like VoxInk covers 90% of your needs at 0% of the cost. Use the savings on something that actually moves the needle for your practice.

The bottom line

Speech recognition has reached the point where open-source models running on consumer hardware are accurate enough for the vast majority of clinical dictation. The gap between a $0 local model and a $199/month cloud service is real — but for most practices, it's not $199/month worth of real.

We built VoxInk because we needed it ourselves. It saves our clinicians 15 minutes each per day, works in every app we use, keeps patient data completely local, and costs nothing to operate. If that sounds like what your practice needs, get in touch — we're happy to show you how it works.