Medical Dictation: Software, Devices and How It Works

Medical dictation is the clinician's spoken account of a visit, turned into text for the patient record. Three workflows do this today. In back-end dictation, the clinician speaks, speech software makes a draft, and a medical transcriptionist corrects it before the physician signs. In front-end dictation, the clinician speaks straight into the electronic health record and corrects the text as they go. In ambient documentation, an AI scribe listens to the visit and drafts the note afterwards. Each workflow puts the editing in a different place, and that decides which one suits a given clinician.

This guide explains each workflow with the evidence we could check, shows the prices we could verify, and says where the evidence runs out. For the broader subject, see our overview of dictation for work.

Medical dictation turns a clinician's words into the record in one of three ways

The words "dictation" and "transcription" are often used for all three, so check which one a product actually does before you compare prices. A product that types into your EHR and a service that returns a transcript to your inbox solve different problems.

The back-end role is the one medical transcriptionists fill: they listen to recorded dictation and transcribe it into medical reports ¹. If you are weighing that career, read our guide to medical transcription.

Back-end dictation sends a machine draft to a transcriptionist, and the draft still needs review

The 2018 study of clinical notes measured each stage of the back-end process. Notes were dictated in 2016, and the speech recognition draft had an error rate of 7.4 errors per 100 words ². After a medical transcriptionist edited the drafts, the error rate fell to 0.4 percent, and to 0.3 percent in the notes that physicians signed ³.

The lesson is that a clean-looking draft is not a finished note. The transcriptionist's review is what brings the error rate down, and that review has a cost in staff time and in the wait for the edited note. Back-end dictation suits a clinician who does not want to edit at all, or a practice that already has transcription staff and a review step.

Front-end dictation puts the editing on the clinician, and the learning curve is real

In front-end dictation the clinician sees the text appear and fixes it, so the editing stays on their side of the desk. A 2013 hospital study of radiologists measured how much correction that takes. Experienced radiologists corrected 6 characters per report ⁴. New users corrected 11 characters per report ⁵. The gap is the learning curve: new users need time before the software and their own dictation habits line up.

We did not find a peer-reviewed trial that compares front-end and back-end dictation directly on accuracy and turnaround. The evidence we could check comes from separate studies in different settings, so treat it as a guide to what to measure, not as a ranking. Our glossary explains the measure used in most of these studies: word error rate.

Dragon Medical One is the tool one health system's Epic guidance names for dictating into the chart

Dragon Medical One is the product that a Canadian health system's Epic tips name for dictating into a patient's chart. The post says it "remains the app to use when dictating into a patient's chart on a desktop or tablet in Hyperspace" ⁶. The post names Hyperspace as the desktop route for a clinician working at a workstation. Check which Epic client your own system runs.

Microsoft now describes Dragon Copilot as combining and extending Dragon Medical One and DAX Copilot ⁷. Buyers should therefore ask which product their quote covers, and whether it is the desktop dictation tool, the ambient assistant or both.

Dragon Medical One is a per-user subscription, not a one-time purchase. Resellers quote it "per user per month", and Nuance publishes no price ⁸. The reseller rates we found are below.

The rates are $99 per user per month on a one-year term, $89 on two years and $79 on three years ⁹. A three-year commitment at $79 comes to $2,844 per user before setup ¹⁰. Expect a one-time implementation fee on top, which one vendor guide gives as $525 ¹¹.

Not every source agrees on the annual figure. Another vendor's blog says Dragon Medical One costs "$99 per user per month or $990 per user annually" ¹². A full year at the $99 monthly rate comes to $1,188 ¹³, more than that $990 figure, so the two sources probably assume different terms or prepayment. Ask a reseller for an itemized quote before you budget. For how these tools compare, see our overview of dictation software.

The general Dragon product and the medical product are sold for different work

The general Dragon edition is sold as a one-user license. A US reseller listed Dragon Professional v16 at $685.99 for one user when we checked on 16 October 2026 ¹⁴. Dragon Medical One is a different purchase: a per-user subscription aimed at clinical notes and the EHR. Compare them on the workflow you need, not on the Dragon name. For the full product comparison, see Dragon Medical One vs Dragon NaturallySpeaking.

Accuracy studies give mixed signals, so test vocabulary and EHR fit on your own notes

Accuracy numbers from one study do not settle which product is medical. A 2000 comparison tested speech software on physicians dictating medical reports. The lowest vocabulary error rate went to IBM ViaVoice 98, at 7.0 to 9.1 percent ¹⁵. Dragon's rate in the same study was 14.1 to 15.2 percent ¹⁶. The software dates from 2000, so treat the result as a historical benchmark, not a verdict on today's products.

A 2025 pilot in a neurosurgical practice shows the same caution in the other direction. Dragon Medical One had a mean word error rate of 1.54 percent, counting all errors ¹⁷. A modified Whisper-based model had 1.75 percent ¹⁸. The difference was not statistically significant (p = 0.080) ¹⁹. With a small pilot like that, the honest conclusion is that accuracy alone did not separate the two products.

What separates the products in practice is what they do with your vocabulary and your chart. On a Mac, Voice Control accepts up to 1,000 custom vocabulary terms, and you can teach it how to say each one ²⁰. That is a general-purpose workaround, not a clinical vocabulary. We could not find a published list of Dragon Medical One's specialty vocabularies, so treat any claim about its coverage as unverified until you test it on your own notes. The testing step is in the final section below.

Epic's physician data links how a note is written to the time spent on it

The Epic analysis groups physicians by how their note text was produced. In Epic's adjusted models, physicians who mainly used transcription and dictation spent an average of 4.8 minutes per visit on notes ²¹. Physicians who mostly used copy and paste spent 13.1 minutes ²². The comparison is observational, so it shows what physicians did, not what a controlled change would achieve, but the gap is large.

Time on notes matters because of where the rest of the day goes. In a 2016 time and motion study of physicians, 27.0 percent of the office day went to direct face time with patients, and 49.2 percent went to EHR and desk work ²³. Any tool that moves note work out of the EHR window is competing for that second number.

Templates have a limit. A 2024 review of DotPhrases, the short text shortcuts that EHRs use for templates and phrases, warns that excessive use "can lead to an excessively long and cumbersome note" ²⁴. Set up a few templates for the notes you write most, and resist adding one for every case. Our glossary entry on dictation templates defines them.

Ambient AI scribes draft the whole note from the visit conversation

An AI scribe is a different tool. It listens to the visit and writes a draft note, so the clinician does not dictate at all. In a 2025 randomized trial of physicians, the scribes gave mixed results. Physicians who used Nabla cut their time in the note by 9.5 percent against the control group, the only arm with a statistically significant reduction ²⁵. Physicians who used Microsoft DAX Copilot showed no statistically significant change, at -1.7 percent ²⁶.

Those two results are the reason to test before you buy. The scribes did not help equally, so a tool that saves time in one practice may not do the same in another, and the draft still has to be checked against what was said in the room. For how these tools compare, read our guide to AI medical scribes.

Microphone position and room noise change what the engine hears

The microphone decides much of what the recognizer receives. Nuance's guidance for a headset is to keep the microphone element "about an inch from the side of your mouth, but not touching it" ²⁷. Distance changes the signal the engine gets, and a microphone held farther away picks up more of the room.

Noise is the bigger problem. A 1996 paper opens by stating that the performance of a speech recognition system "degrades rapidly in the presence of ambient noise" ²⁸. A clinic has fans, doors, corridor talk and other staff, so a quiet room and a close microphone are worth more than a feature list.

Volume matters for the people wearing the headset. NIOSH investigators found that the dictation equipment in medical transcription has a potential for excessive noise exposure ²⁹. Keep playback volume moderate, and choose equipment you can set to a comfortable level.

If you buy a dedicated device, read the model's setup guide before you test software, because the engine's accuracy depends on the audio it receives.

Free and general dictation can draft letters, but patient data raises HIPAA questions first

Free built-in dictation is a good first step for correspondence and general notes. Its limit is clinical vocabulary: the Voice Control workaround above is general-purpose, not clinical. Patient information also brings compliance rules, and those apply whether the tool is free or paid.

Under HHS guidance, a cloud service provider that holds health information for a covered entity must sign a business associate agreement: "the covered entity (or business associate) and the CSP must enter into a HIPAA-compliant business associate agreement (BAA)" ³⁰. The Security Rule also requires a risk analysis, an accurate and thorough assessment of the risks to the confidentiality, integrity and availability of electronic protected health information ³¹. A dictation tool that touches patient information belongs in that assessment.

On-device processing reduces where audio travels. Apple says that on a Mac, voice inputs and transcripts for general Dictation are processed on your device and not sent to Siri servers ³². That is useful, but it is not a business associate agreement, and it does not replace the risk analysis. Ask the vendor for the BAA in writing before any patient data goes in.

Yap, which we make, is a general dictation app for Mac and Windows. It is not built for clinical documentation, so this guide does not compare it with EHR products. The question of a free option has its own page: is there a free medical dictation app?.

Choosing medical dictation software comes down to five checks

1. Workflow. Decide who corrects the words: you, a transcriptionist, or an AI scribe. The table near the top shows the trade-offs.
2. EHR fit. Ask whether the product types into your EHR, on the device you actually use. The desktop route in Hyperspace is one described above ⁶, but check your own system and version.
3. Vocabulary test. Dictate ten real notes from your specialty and count the errors. Use word error rate as the measure, and compare the vendors on your own notes, not on their demos.
4. Privacy paper. Get a signed business associate agreement and complete your own risk analysis before any patient data goes in. Our guide to HIPAA-compliant dictation explains the terms.
5. Total cost. Add the subscription for your chosen term, the one-time setup fee, the microphone and the support tier. A three-year Dragon Medical One commitment at $79 comes to $2,844 per user before setup ¹⁰.

If you are learning the work rather than buying software, sample reports are a free way to see what medical notes look like. MTSamples lets you "print, share, or link to these sample reports for educational purposes", but its reports are user-contributed and "not guaranteed to be complete or error free" ³³. Read them as examples of format and phrasing, not as verified records. For more on samples, see the question on medical dictation examples.

For the basic meaning of the term, see what is medical dictation. For how the same tools serve other roles, read our guides to dictation for nurses and therapists and how much time dictation saves at work.