VOICE · ASR · NLP · trained on medical data
Voice input for radiologists, pathologists and ultrasound specialists
Russia’s first RIS-integrated module for filling in reports by voice. Recognition of Russian-language medical vocabulary, voice control of the DICOM viewer and shortcut snippets — with no data sent to the cloud.
ASR · MEDICAL MODEL
Speech recognition
An acoustic and language model fine-tuned on radiology documents. Vocabulary of CT, MRI, ultrasound, endoscopy and pathology.
Learn more
REPORT · TEMPLATES
Voice template filling
Structured report templates, switching between fields by voice, phrase snippets, auto-correction of numbers, dates and abbreviations.
Learn more
COMMANDS · VIEWER
Voice control
Window/level, MPR, series switching, slice navigation, comparison with the prior study, signing and sending — without touching the keyboard.
Learn more
Module 01 · ASR
A model fine-tuned on a medical corpus
Most general-purpose ASR engines lose accuracy on medical vocabulary: Latin terms, chemical formulas, numeric ranges, abbreviations. ArchiMed-ASR is our own model, fine-tuned on radiology documents and audio recordings of physicians’ dictation. It recognizes punctuation, numbers, units of measurement, abbreviations, ranges, and Latin and English terms within Russian-language dictation.
- Domain vocabularies by specialty. Radiology, endoscopy, ultrasound, pathology — each has its own set of terms, diagnoses and units of measurement.
- Auto-correction of numeric forms. “Seven millimeters” → “7 mm”, “zero point six three two” → “0.632”, dates in DD.MM.YYYY format, Latin segments.
- Local model — no cloud. Runs on the physician’s workstation, inside the healthcare facility’s secure network. Patient personal data never leaves the facility’s perimeter.
- Adaptation to the physician’s voice. Accuracy improves with use thanks to the user’s acoustic profile.
- Noise and echo cancellation. Works in a room next to a CT scanner, in an open-plan radiology department and with a lavalier microphone.
- Windows and Linux. Desktop application and web version; support for USB microphones and pedals: PowerMic / Olympus / Philips SpeechMike.
ArchiMed-ASR · RU-MED v3.2
Live transcription
local model · 0% cloud · ~120 ms latency
REC 02:14
At the level of segment S6 of the right lung, a rounded lesion with spiculated margins measuring 14×11 mm is seen.
Mediastinal lymph nodes are not enlarged; the pleura is thickened to 3 mm on the right.
Impression: suspected lesion in S6 on the right, follow-up CT recommended in…
Template · chest CT · screening
Chest CT
patient M-62 · STUDY-2024-77321 · Dr. Sokolova, PhD
Lung parenchyma FILLED
Solid nodule in S6 of the right lung, 14×11 mm, with spiculated margins. No other focal or infiltrative changes identified.
Mediastinum DICTATING
Lymph nodes not enlarged, vascular configuration preserved, trachea and main bronchi patent…
Pleural cavities PENDING
Say “Go to pleura” or press the pedal
Impression SNIPPET
/normal_chest
→
expands into the standard block “no focal or infiltrative changes”
SR · DICOM-encapsulated
Ready for QES signing
Module 02 · Templates
A structured report — by voice
Report templates by diagnosis and study type. The radiologist moves between sections by voice, dictates findings and expands phrase snippets. The text can be edited before saving, and the final report is signed with a qualified electronic signature (QES).
- Templates by diagnosis. Chest CT, contrast-enhanced abdominal CT, neuro MRI, mammography — dozens of ready-made frameworks with sections and mandatory fields.
- Voice navigation between fields. “Go to impression”, “Back to mediastinum” — the focus jumps between template sections without touching the mouse.
- Phrase snippets. The voice command “Normal chest” expands into a standard block of text. Custom snippets can be edited in 30 seconds.
- Retrieval by pathology. Voice request “Lung-RADS lung nodule template” — the system opens the right template.
- Editing before saving. The radiologist reads through the text, corrects it with the mouse or by voice, and only then gives the “Save and sign” command.
- Direct entry into the RIS. The finished report goes to the RIS / HIS (hospital information system) via API — no copy-pasting into a separate text editor.
Module 03 · Commands
Viewer control — hands on the image
The radiologist never looks away from the image: window/level, series switching, MPR, comparison with the prior study — all by voice command (commands are spoken in Russian). Reading goes faster, and hands and neck get less tired.
- W/L presets. “Lung window”, “Bone”, “Soft tissue” — presets switch by voice in half a second.
- Series and slices. “Open DWI”, “Compare with last year’s CT”, “Slice 142”, “Scroll to the diaphragm”.
- Reformats. “Turn on MPR”, “Show coronaries”, “Volume reconstruction” — rendering switches without the cursor.
- Measurements. “Measure with ruler”, “Place HU marker” — the tool is activated by voice, and the number is taken from the latest transcription.
- Final actions. “Sign and send” — the report is signed with a QES and sent to the RIS / EGISZ (Unified State Health Information System of Russia).
- Custom hot phrases. Each command is mapped to a viewer function in the settings — for non-standard diagnosis-specific checklists.
- Command synonyms. The same action can be triggered with different words — no need to memorize the exact wording.
- Working with the report. “Export”, “Print”, “Add to favorites”, opening and searching for a template by body region — also by voice.
DICOM Viewer · 32 commands active
PRESET
SERIES
RENDER
PRIOR
TOOL
FINAL
Recognized command log
W/L presets · series · MPR · signing · sending
Lung window
W/L preset → Lung (-600 / 1500)
Open DWI series
Series.load("DWI b=1000") · MR/Brain/3
Turn on MPR
Render.mode = "MPR" · 3 viewports · ax/cor/sag
Compare with last year’s CT
Prior.load(latest, modality=CT) · split-screen
Measure 14 millimeters with ruler
Tool=Ruler · annot.size = 14 mm
Sign and send
Sign(QES) · Send → RIS · ACK 200
Who it’s for
Use cases for voice input
One engine — six workflows: radiology, pathology, endoscopy, ultrasound, on-call service and remote reading at centers of expertise.
Radiologist · radiology department, CT/MRI
Dictation of CT, MRI, X-ray and mammography reports. Eyes and hands on the image, voice on the text. QES signing on command.
Pathologist
At the microscope, both hands are busy. Voice dictation of gross and microscopic descriptions, WSI navigation, ICD-O snippets, TNM auto-fill.
Endoscopist
Reporting during the procedure: probe, biopsy, photo markers. Sterile control of the viewer window with a foot pedal + voice.
Ultrasound diagnostics
Transducer in one hand, voice for measurements and the report. Templates for obstetrics, abdomen, thyroid, limb vessels.
On-call service · CITO
CT stroke protocol, polytrauma, acute abdomen — reports in minutes. Voice is faster than the keyboard, especially at night.
Remote reading
Teleradiology center of expertise: a high volume of studies per shift — voice is faster than the keyboard.
Integrations
Fits into your existing workflow
Open API, ready-made connectors to RIS/HIS, support for medical microphones and pedals. Desktop for Windows and Linux, web version for thin clients.
ArchiMed RIS / Workstationnative integration
Third-party RIS / HISREST API · JSON · HL7
DICOM ViewerSR · signing · sending
Web Viewerbrowser · WebRTC audio
Windows 10 / 11desktop client
LinuxRHEL / Astra Linux SE
PowerMic · SpeechMikeUSB microphones
Foot pedalOlympus · Infinity
Snippet editorphrase templates
QES · CryptoProreport signing
EGISZ · electronic medical documentsSEMD submission
Open SDKPython · C# · Java
Regulatory compliance
Local model — patient data never leaves the facility’s perimeter. Russian software, an FSTEC certificate (Federal Service for Technical and Export Control of Russia), a Roszdravnadzor registration certificate (Russian medical device registration).
Register
Unified Register of Russian Software
Rights holder — ArtVision LLC
FSTEC
Certificate of conformity
personal data protection
Law
152-FZ · 323-FZ
federal laws on personal data and health protection · local, no cloud
Roszdravnadzor
Medical device registration
Aivory Voice voice assistant
Web
no additional
installation
installation
LAT · EN
Latin and English
terms in the text
terms in the text
E-sign
report signing
by voice command
by voice command
100%
local processing,
nothing sent to the cloud
nothing sent to the cloud
Раньше я печатал описание 6–8 минут на одно КТ ОГК и приходил домой с ноющим запястьем. Сейчас диктую за две минуты, голос — пресет лёгочного окна, голос — следующая серия. К концу смены остаются силы.