Voice Dictation for Medical Professionals: Privacy and HIPAA

Yes, voice dictation can be HIPAA-compliant if the audio never leaves the device. Local-first tools like Parrot's on-device mode process everything using Whisper.cpp on your Mac - no cloud, no third-party servers, no PHI transmission. With clinicians spending an estimated 49% of their time on documentation (Annals of Internal Medicine, 2017), private voice dictation is a practical solution for reducing that burden.

The HIPAA problem with cloud dictation

Most voice dictation tools send your audio to a cloud server for processing. When that audio contains protected health information (PHI) - patient names, diagnoses, treatment plans - you have a HIPAA compliance issue.

Using a cloud transcription API doesn't automatically violate HIPAA, but it requires a Business Associate Agreement (BAA) with the provider. Some providers offer BAAs (OpenAI does for enterprise accounts), but many don't. And even with a BAA, you're trusting a third party with patient data.

Local mode: the simplest solution

The cleanest way to handle PHI in voice dictation is to keep everything on-device. When audio never leaves your machine, there's no transmission of PHI to worry about.

Parrot's local mode does exactly this. Whisper.cpp runs the transcription model on your Mac. Ollama handles the AI cleanup locally. Your audio and text stay on your laptop - no cloud, no network requests, no third parties.

This doesn't mean you can ignore all HIPAA requirements. You still need appropriate device security (FileVault encryption, strong passwords, auto-lock), and your organization's policies around dictation software still apply. But you've eliminated the biggest risk: data in transit.

Custom vocabulary for medical terminology

Medical dictation has a specific challenge: terminology. Drug names, procedures, anatomical terms, and abbreviations that general-purpose transcription engines frequently get wrong.

Parrot's custom vocabulary lets you add your terms - "metformin," "echocardiogram," "HbA1c" - so the transcription engine recognizes them. You build this list once and it applies to every dictation going forward.

Workflow: dictating patient notes

Here's how medical professionals typically use Parrot:

• Between appointments, press the hotkey to start recording.
• Dictate the patient note conversationally: "Patient is a 54-year-old male presenting with chest pain radiating to the left arm. BP 150 over 95. History of hypertension."
• The AI cleanup formats it into clean clinical text, removing filler words and applying consistent formatting.
• The note is pasted directly into your EHR or documentation system.

The entire process takes 30 seconds to a minute, compared to 3–5 minutes of typing the same note. Over a day with 20+ patients, that adds up to over an hour saved.

Cloud options with BYOK

If local mode doesn't fit your workflow (older Mac, limited disk space), Parrot's BYOK (Bring Your Own Key) mode lets you use cloud providers like OpenAI Whisper with your own API key. Some providers offer BAAs for enterprise accounts. You control the relationship with the provider directly.

The key difference: in BYOK mode, audio goes directly from your Mac to your chosen provider. Parrot never sees or stores it.

What about Dragon Medical?

Dragon Medical is the incumbent in medical dictation, and it's good at what it does. But it's expensive ($99/month for individual licenses), requires a Windows PC or specific integrations, and is optimized for large hospital systems.

Parrot isn't trying to replace Dragon in a hospital IT department. It's for individual practitioners, small practices, and medical professionals who want something that works on their Mac. Local mode keeps everything on-device, BYOK mode gives you control over your cloud provider.

Getting started

If you're a medical professional interested in trying voice dictation with local processing, join the waitlist to get access when Parrot launches. Choose local mode during setup, and add your medical terminology to the custom vocabulary. The whole process takes a few minutes, and you can start dictating immediately.