Why does dictated text sound different from typed text?

Dictated text sounds like speech because it started as speech. Spoken language has a lower share of content words and carries more fillers, restarts and loose clauses than writing does. Dictation keeps those habits unless the software or the writer removes them, so a first draft keeps some of the voice of the conversation. Many of the differences can be fixed in revision. Our guide on whether dictation changes how you write covers the research behind this.

Speech has a lower share of content words

Researchers measure the difference with lexical density, the share of content words such as nouns, verbs, adjectives and adverbs in a text. Ure (1971) placed the boundary at 40 percent, with speech below it and writing above it ¹. Eggins (1994), in a corpus analysis, found 33 percent in speech and 42 percent in writing ². Both figures reach us through a 2015 paper that summarizes them, and definitions of lexical density vary, so read them as a direction rather than a precise measure.

We have not measured lexical density in dictated prose. Our inference is that a dictated draft carries more of the spoken pattern at first, which may be why a dictated sentence can sound like talk even when every word is correct.

Speech carries fillers and restarts that writing usually drops

Speech is full of the pauses, restarts and fillers that show a speaker is still working out what to say. A corpus study of task-oriented conversation found that disfluency rates rose when speakers discussed abstract figures, which the authors read as harder planning ³. Dictation picks up the same material. Readability studies cited by Lease and Johnson in 2006 found that fillers and speech repairs can be deleted from transcripts without losing meaning ⁴. Because the studies found that the deletion does not compromise meaning, removing fillers is a reasonable first edit.

Some software now does the removal for you. Microsoft says Fluid dictation in voice typing on Copilot+ PCs automatically corrects grammar, punctuation and filler words as you speak ⁵. The page describes corrections to grammar, punctuation and filler words, not to sentence structure, so the draft can still read as speech.

Listeners miss recognition errors by ear

In Hong and colleagues' 2018 study, people listening to the audio of recognized text missed more than half of the recognition errors ⁶. We found no study that compares screen and audio review directly. Listening may also hide a clumsy sentence, so check the draft by reading it.

Which differences you can fix, and which you cannot

The table lists the fixes that work on the text itself. The rhythm of a passage is harder to change, and it takes a full pass to hear.

What to do

1. Search for fillers first. Find "um", "uh" and "so basically" and remove them. How to stop saying um when dictating covers the spoken side of the problem.
2. Split long sentences. Break each one where you paused while speaking.
3. Read on screen. Check word choice and tone once the fillers are gone.
4. Check names and terms by hand. A recognizer can get them wrong, and a sound-alike error is easy to miss by ear.

Filler words are the main subject of the filler words entry. If you want to know how far AI editing goes, does AI cleanup make writing sound generic addresses the register change that some tools make. For the revision steps in full, see how to dictate. For the related questions, see is dictated writing worse than typed writing and can dictation help with writer's block.