How do you improve Mac dictation accuracy?

The sound that reaches the speech model is the first thing to change on a Mac. Put the microphone close to your mouth, quiet the room, and set the language you are speaking. Each of those three steps can change how many words come out wrong. After that, speak in complete sentences and proofread for meaning. Apple Dictation has no voice-training step, so you cannot teach it your voice, but Voice Control lets you add names and terms with their pronunciations ¹.

Fix the input before you change any setting

The sound is the first factor to check. In the 2017 Ravanelli et al. figures, a single distant microphone gave a word error rate about 3.2 times the close-talking rate ². The study's conditions are not identical to a desk setup, so treat the ratio as a rough guide. The direction is clear, though: the authors say distant speech recognition is dramatically reduced compared with close talking ³.

Apple's Dictation help gives three checks, and they are the right first steps. Position yourself as you normally use your Mac, speak clearly at a volume that is neither too quiet nor too loud, and avoid background noise ⁴. Keep the Mac's microphone unobstructed by clothing or your body ⁵. If you use an external microphone, make sure its input volume is sufficient ⁶.

For a noisy office or a room with echo, Apple says a headset microphone may help ⁷. A headset keeps the microphone at a fixed distance from your mouth, which is the variable the distance figures above are about. The question on whether a laptop microphone is good enough covers the built-in option, and the question on microphones for Mac dictation covers the choice of device.

Set one language, then speak in full sentences

Apple describes switching between languages rather than recognizing two languages in one sentence ⁸. Check the dictation language before you start, and add the language you need through the Edit button next to Languages ⁹. The guide to changing the dictation language on Mac has the steps for the case where the wrong language is active.

Speak in full sentences at a normal pace. Dragon's guidance, from a different product, tells users to speak clearly and naturally at a conversational pace, with a consistent speed and volume ¹⁰. The same guidance says to pause briefly before and after dictating punctuation ¹¹. Apple's Dictation adds some punctuation automatically in supported languages ¹², so check where it puts commas and periods rather than dictating every mark.

Add names and jargon through Voice Control

Dictation has no custom word list in the pages we read, so the names, product names and technical terms are where a general model is most likely to fail. Voice Control accepts up to 1,000 custom vocabulary terms, and you can record how each one is pronounced ¹. You can also import terms from a text file with one term on each line ¹³. Recording a pronunciation requires Dictation to be turned on, although you can save a term without recording it ¹⁴.

The trade-off is that while Voice Control is on, standard Dictation is not available ¹⁵. Turn it on for the work that needs the vocabulary, and turn it off when you only need to type. The question on adding words to Mac dictation has the setup steps.

Domain vocabulary can help a speech model. In a 2022 study of UK Supreme Court hearings, adding an in-domain language model and common legal phrases can improve the word error rate of a generic model ¹⁶. That study covered a whole legal domain with a different engine, so its result does not show what a few names would do.

It depends: what helps, and what cannot help

• Helps most: a close, steady microphone and a quiet room. These change every word, so they come first.
• Helps for specific words: the Voice Control vocabulary, for names and jargon you use often.
• Cannot be fixed by a setting: a sentence that mixes languages, and rare names the model has seldom heard. For these, proofread and dictate one language at a time.
• No voice training: Apple's Dictation has no step where you read text aloud to train it. Dragon for Mac did in version 6, and its help says Voice Training asks you to read a story aloud so the software learns how you speak ¹⁷. If you need voice training, that is a different product.

Personalization can cut errors a lot for some speakers. In Google's Project Euphonia study of 432 people with disordered speech, personalized models reached a median word error rate of 4.6 percent, against 31 percent for general models ¹⁸. That result is for speech that general models handle poorly, and it used a research system, so it does not promise the same for typical speech.

What to do: a short routine

1. Check the microphone. Use the one you normally dictate with, close to your mouth and unobstructed.
2. Check the language. Set one dictation language for the document you are writing.
3. Dictate a 200-word passage. Count the words that came out wrong and divide by 200, to get a rough error rate.
4. Change one thing and test again. Move the microphone, quiet the room, or switch to a headset, then repeat the passage.
5. Add the words you keep correcting. Put names and jargon into Voice Control, and proofread them by sight.
6. Proofread for meaning. Apple underlines ambiguous text in blue, so you can click the underlined word and choose an alternative ¹⁹.

If the rate stays high after these steps, the problem is probably the engine or the vocabulary, and a dedicated app is the next option. The question on whether Apple Dictation is any good explains how to read the test, and why Apple Dictation is so bad covers the causes in more detail. The overview of how Apple Dictation works explains where the model runs, and the overview of dictation on Mac compares the options. For the general method that applies to any dictation tool, see how to improve dictation accuracy. The word error rate entry explains the percentage.