How accurate is voice recognition software?

Voice recognition software is accurate on clear speech spoken close to the microphone, and much less accurate when the conditions change. Your microphone, your speech and the task all move the result, so no single accuracy figure applies to everyone. Accuracy is usually reported as word error rate (WER), the share of words the software gets wrong. Read any accuracy claim with its conditions beside it.

Clear speech is the easiest case for the software

For typical speech, a 2026 study notes that automatic speech recognition reaches word error rates below 5 percent ¹. That is the easiest case: a clear voice, a good microphone and ordinary vocabulary.

Speech affected by dysarthria is much harder for these systems. The same 2026 study found that every commercial system it tested exceeded 51 percent word error rate on severe dysarthria ². The study used a small set of speakers and default settings, so it shows the size of the gap, not the best each system can do. For the mild and severe cases in detail, see the question on dysarthric speech.

Dragon's accuracy claims come from the vendor and from old tests

Dragon is the best-known product in this area, so its figures matter. Nuance, which makes Dragon, says Dragon Professional v16 lets employees dictate documents 3 times faster than typing, with up to 99 percent recognition accuracy ³. The page does not say which audio or vocabulary produced that figure, and "up to" marks the best case. Treat it as the vendor's own claim.

Independent measurements are older. In a 2000 peer-reviewed study, physicians dictated medical reports into Dragon NaturallySpeaking Medical Suite 3.0, and its vocabulary recognition error rate was 14.1 to 15.2 percent ⁴. That study tested software from 1998 and 1999, and it measured vocabulary errors rather than the same word error rate, so it shows where the technology started. It does not show what Dragon does today. We did not find a recent independent test of general Dragon accuracy.

For a more recent figure, a 2025 pilot in neurosurgery found that Dragon Medical One had a mean word error rate of 1.54 percent, counting all errors ⁵. Dragon Medical One is the clinical edition, and the pilot was small. Treat it as one data point, not a ranking of products.

Your microphone can change the result a great deal

Microphone placement changes the result a great deal. In 2017 figures from Ravanelli and colleagues, a single distant microphone gave a word error rate about 3.2 times the close-talking rate ⁶. The two conditions were not identical, so read this as a rough ratio. A close headset is the simplest fix.

Accuracy also varies between speakers. Across five commercial systems in a 2020 study, the average word error rate was 0.35 for Black speakers and 0.19 for white speakers ⁷. That study used interview speech, not dictation apps, but the gap is a reason to test any tool with your own voice.

Recognized text still needs checking

A recognized draft is not a finished text. In a 2018 study of clinical notes, 96.3 percent of speech recognition drafts contained at least one error ⁸. Clinical notes are long and full of drug names and numbers, so a short message has fewer chances for an error. The habit still applies: read the draft, then correct it. For the method, see the question on proofreading dictated text.

Test accuracy on your own writing

The fastest way to get a real answer is a short test. Use the same passage for every tool you compare, so the difference is the software and not the text.

1. Dictate a page of real work in the app you use most, with the microphone you will really use.
2. Count the corrections you make, then divide the errors by the number of words. The result is a rough word error rate for you.
3. Add your own terms, such as names and product names, using a custom vocabulary where the tool supports one.
4. Repeat in a noisy room if that is where you will dictate, because the drop from a close microphone is large.

Accuracy is only one reason to choose a tool. Is AI dictation more accurate than built-in dictation? compares the two kinds directly, and what dictation software is explains where each kind of tool runs. For how the error rate is calculated and read, see the guide to speech recognition accuracy.