Radiology Voice Recognition Built for Data Privacy

When radiology departments evaluate voice recognition systems, they run into the same problem: most options on the market route audio through cloud infrastructure. That creates a data exposure question, a BAA requirement, and an IT dependency that many departments would rather not carry.

VoicePrivate Healthcare Edition works differently. It installs on the radiologist's workstation — Mac or Windows — and processes audio entirely on that machine. No cloud routing. No external API calls. No audio leaves the device.

This page covers the system architecture, deployment options, and what the transition looks like for departments moving from legacy voice recognition platforms. If you are evaluating radiology voice recognition at a department or enterprise level, this is the technical context you need.

What “Voice Recognition System” Means Here

The word “system” in enterprise purchasing contexts usually implies server infrastructure, network integration, and a vendor support contract. VoicePrivate is architected differently — and that difference matters for your procurement and compliance evaluation.

Here is what VoicePrivate is, specifically:

• A local software system. It installs on each radiologist's workstation. There is no shared server to configure, no network integration to build, and no vendor infrastructure to depend on after deployment.
• An on-device AI engine. The voice recognition model runs entirely on the workstation processor. On Apple Silicon Macs, this happens on the Neural Engine. On Windows machines, on the CPU or GPU. Audio is never transmitted anywhere.
• On-device architecture — no external data transmission. Audio is processed entirely on the workstation and nothing is transmitted to external servers. There is no cloud data routing in the architecture. Your compliance team should assess what that means for your organization's specific obligations.

For departments evaluating data exposure requirements, the architecture is direct: audio is captured on the workstation, processed on the workstation, and text output stays on the workstation. Nothing leaves the device during dictation.

Architecture for Radiology Departments

The architecture is straightforward. Each component is on-device.

Photo by Pexels

Radiology Vocabulary

General-purpose speech recognition fails in radiology reports because it is not trained on the vocabulary radiologists use. Terms like “subcentimeter hypodense hepatic lesion,” “mild left basilar atelectasis,” and “no acute intracranial abnormality” are not in a general conversational corpus. A system that mishears these terms forces the radiologist to correct output, defeating the purpose.

VoicePrivate Healthcare Edition includes a vocabulary set built for clinical documentation, with particular coverage for:

• Imaging modality terminology (CT, MRI, PET, ultrasound, fluoroscopy, X-ray)
• Anatomical structures with standard reporting names
• Finding descriptors used in structured reporting (attenuation, enhancement, morphology, signal characteristics)
• Common measurement conventions and laterality expressions
• Impression and recommendation phrasing patterns

Custom vocabulary is also available on paid plans. If your department uses subspecialty terms, equipment model names, or protocol-specific shorthand that a standard vocabulary set would miss, those terms can be added and will be recognized in subsequent dictation sessions.

Deployment Options

VoicePrivate is sold as individual workstation licenses, departmental plans, and enterprise agreements. The deployment model is the same across all tiers: install on each machine, activate with a license key, download the model once, and the system is operational.

IT requirements are minimal. VoicePrivate is a standard macOS or Windows application with no server-side components and no persistent network requirements after first-run model download. There is no directory integration, no SSO requirement, and no agent to deploy. License key distribution and activation is handled locally on each machine.

For current pricing, see the Healthcare Edition pricing page.

Migration from PowerScribe and Dragon Medical

Departments considering a move from Nuance PowerScribe or Dragon Medical One are typically motivated by one or more of the following: cost structure changes following Microsoft's acquisition of Nuance, cloud-dependency concerns, or contract renewal evaluation cycles.

Here is what migration to VoicePrivate does and does not require:

No special integration with your RIS or PACS. VoicePrivate dictates text directly into whatever application the radiologist is using. LIVE dictation mode types text into any active text field — including browser-based reporting systems. There is no RIS integration module to configure and no PACS plugin to install. If the radiologist can click into a text field, VoicePrivate can type into it.

No acoustic model training required. Legacy systems like Dragon Medical often required extended voice profile training sessions before the system was reliable for a given radiologist. VoicePrivate's underlying AI model works accurately without per-user training. Some radiologists prefer to add custom vocabulary for their specific reporting style, but it is not a prerequisite for productive use.

No historical dictionary migration. If radiologists had extensive custom macros or auto-text built into their legacy system, those will need to be rebuilt. VoicePrivate's custom vocabulary and AI command mode cover most of what auto-text handles, but there is no automated migration path from PowerScribe or Dragon Medical custom dictionaries.

Different report workflow. PowerScribe and similar systems have purpose-built report editors and RIS integrations. VoicePrivate works at the OS level, typing into whatever is on screen. For departments that need a specific structured report editor, that editor stays in place — VoicePrivate just handles the voice input.

Privacy Architecture Summary

For compliance teams and IT security reviewers, here is the complete data flow during a VoicePrivate dictation session:

• Audio is captured by the workstation microphone.
• Audio is processed by the on-device AI engine. It is not buffered, stored, or transmitted at any point.
• Transcribed text is output to the active application (your reporting system, a text editor, a web form).
• No audio, text, or usage data is sent to VoicePrivate servers during or after a dictation session.
• No account is required. There is no user profile, no session ID tied to a clinician identity, and no cloud-side data accumulation.

For a detailed explanation of the architecture, including how the local model operates and what telemetry (if any) is collected during app updates, see the HIPAA Architecture page.

We will not tell you that VoicePrivate is HIPAA certified — that is not a meaningful claim, and no software tool carries such a certification. What we can tell you is the technical architecture. Your compliance team determines what that means for your obligations.

Evaluate Healthcare Edition

VoicePrivate Healthcare Edition is available for evaluation with no commitment required. The free tier includes 5,000 words of transcription, which is enough to dictate a representative set of radiology reports and evaluate the vocabulary engine against your actual report types.

Paid plans include the full Healthcare Edition vocabulary, extended file transcription (audio and video files up to your plan limit), AI command mode for post-dictation text transformation, and multi-seat licensing for departmental use.

All paid plans include a 30-day money-back guarantee. Enterprise pricing is available for 6+ seat deployments.

For a departmental evaluation, the recommended path is:

1. Download and install on one workstation.
2. Complete the one-time model download (requires internet; after this, the app works offline).
3. Dictate a representative sample of your report types on the free tier.
4. If the vocabulary performance meets your requirements, contact us about a departmental or enterprise evaluation license.

Start with the Healthcare Edition overview or go directly to the pricing page for plan details.

For more on VoicePrivate's approach to healthcare documentation, visit the Healthcare Edition overview. For architecture details relevant to compliance review, see HIPAA Architecture. For pricing and plan comparison, see Healthcare Edition pricing.