Reading speed vs listening speed: the research on why you read faster than anyone can talk

The figure you've probably heard is 300 words per minute. The best evidence says lower. In 2019 Marc Brysbaert pulled together 190 studies covering 18,573 participants and spanning from 1901 to 2019, and asked a plain question: how many words do we actually read per minute?

The answer for silent reading of English is 238 words per minute for non-fiction and 260 for fiction. Fiction reads faster because its words are shorter. Read the same text aloud and the rate collapses to 183 words per minute, measured across 77 studies and 5,965 participants – the voice becomes the bottleneck the moment you stop reading silently.

These are averages, not limits. Most adults reading non-fiction fall somewhere in the 175-300 wpm range; for fiction it's 200-320. But the headline is that the long-cited 300 figure sits at the fast end of normal, not the middle of it.

Now the other side of the comparison. How fast is speech actually delivered to you? Brysbaert notes that audiobooks are spoken at 140-180 words per minute, a band that lines up with other speech-rate estimates in the literature.

Even professionally paced spoken word stays in that zone. As reported in Brysbaert 2019, Rodero (2016) concluded that the ideal speech rate for radio news is 170 wpm for high-density messages and 190 wpm for lighter ones. That is people whose job is to speak clearly, at their best, and they still land far below the speed at which you read the same words on a page.

Line the two up and the gap is the whole story. You take in written words at roughly 238 a minute; audio hands them to you at 140-180. You read faster than anyone can talk.

The obvious counter is: play it at 1.5x, at 2x, close the gap. There's a ceiling on that, and it's a real one.

As reported in Brysbaert 2019, Foulke and Sticht (1969) found that listening comprehension declines slowly as the word rate climbs, holding up until around 275 words per minute, and then falls off much faster past that point. Listening and reading aren't the same channel running at different speeds; they have different ceilings.

The cleanest demonstration is a head-to-head. Jester and Travers (1966), again as discussed in Brysbaert, took the same prose and found the optimal presentation rate was 300 wpm when it was read silently but only 200 wpm when it was heard. Same words, same reader, two channels – and the eye's ceiling sits well above the ear's.

Reading speed isn't only an academic question. The accessibility standards bodies who put words on screen for a living have to decide how fast a human can comfortably read text, and they've published their answers.

Netflix's English (USA) Timed Text Style Guide caps subtitle reading speed at up to 20 characters per second for adult programs and 17 characters per second for children's. (Netflix specifies characters per second, not words per minute – the two don't convert cleanly without assuming an average word length.) The BBC Subtitle Guidelines take the words-per-minute route, recommending a subtitle reading speed of 160-180 wpm. (Netflix and BBC caption limits last verified: 2026-07-13.)

Different units, same instinct: there's a rate past which text stops being comfortable to read, and it's a design constraint serious enough to standardize. If you care how those limits get set, our guide to caption accuracy standards walks through what the style guides actually require.

The most relatable version of the listening ceiling comes from UCLA. In 2021, Murphy and colleagues had 231 students watch lecture videos at 1x, 1.5x, 2x and 2.5x, then sit a 40-question test.

Immediately after, scores barely moved with speed: about 26 out of 40 at normal pace, about 25 at both 1.5x and 2x, and only then a real drop to about 22 at 2.5x. A week later the pattern held – 24, 21, 21 and 20 respectively. Comprehension stayed essentially flat until playback passed twice normal speed, then it broke.

So there is real room to speed audio up, but it runs out around 2x. Past that, you're trading understanding for time. Reading has no equivalent penalty at the speeds most people actually read.

Put the rates into an hour and the payoff becomes concrete. At 238 words per minute, an hour of silent reading covers roughly 14,280 words (that's 60 minutes times 238).

An hour of recorded audio, meanwhile, is an hour of your life to re-listen to – maybe 40 minutes if you push it to 1.5x, half an hour at the 2x ceiling, and less than that costs you comprehension. The same content as a transcript, you can move through far faster, because you're reading at your own rate instead of waiting for a voice to catch up. If you want the underlying counts, we broke them down in how many words are in an hour of audio.

This is why "just listen to it again" is such a poor answer to "where did they say that thing about the budget?" Re-listening is linear and slow. The transcript is the fast path.

Even the raw speed gap undersells it, because speed isn't the only difference. Audio is linear. It plays at whatever rate it plays, in order, and to find one sentence you scrub a timeline and hope.

Text is random-access. You skim the boring parts, slow down on the dense ones, jump straight to the quote, search for a name, re-read a sentence three times without rewinding. None of that has a words-per-minute figure attached, and it may matter more than the raw rate. The reason a transcript feels faster than re-listening isn't only that you read quicker – it's that you stop being a passenger to the playback speed.

None of this argues against listening. Audio is the right medium for a commute, a walk, a first pass through a podcast you're enjoying. The research just explains a specific, common frustration: when you need to find something, quote something, or review something, going back to the audio is the slow way to do it, and it isn't close.

That's the narrow job Pepys is built for – turn the recording into text once, then read, search and skim at your own speed instead of the speaker's. Point it at a file with audio to text, and for the long recordings where this gap hurts most, our guide to transcribing long audio covers the mechanics. The studies above are the why; the transcript is the how.