Dictation vs typing speed: the case for talking first

Most people can speak about three times faster than they can type. The exact figure depends on the person and the task, but the shape of the gap is stable across the research: speaking sits near 150 words a minute, typing sits somewhere between 40 and 52, and no amount of keyboard practice moves those two numbers close together.

The most cited measurement comes from a 2017 study by researchers at Stanford, the University of Washington and Baidu. In a controlled test, English speech input reached 153 words a minute against 52 for the same people typing on an iPhone keyboard. That is 2.93 times faster, which rounds to the headline everyone repeats: speech is roughly 3x faster than typing.

The rest of this piece unpacks each side of that ratio – why typing is slower than it feels, why speaking is quicker than you would guess, and why the head-to-head gap survives even against a decent phone keyboard. Then the practical part: what to actually do with the difference.

The best data on how fast people really type comes from Aalto University, which in 2018 analysed 136 million keystrokes from about 168,000 volunteers – the largest typing study ever run. The average came out at 52 words a minute, roughly a third of a normal speaking rate.

The distribution matters as much as the average. In that dataset the fastest typists topped out near 120 WPM, and most people sat between 30 and 60. Reference tables tell the same story from a different angle: professional typists work at 43 to 80 WPM, and only the rare specialist role asks for 80 to 95. Very few humans type anywhere near the speed at which they talk.

The uncomfortable part is that training does not fix this. Aalto found that self-taught typists using fewer than ten fingers can be as fast as formally trained touch typists, and that people who had taken a typing course typed at about the same speed as people who had not. Practising your keyboard is a weak lever. Switching input method is a strong one.

The other side of the ratio is speech. The average conversational rate for US English is about 150 words a minute, a figure attributed to the National Center for Voice and Speech. That is the baseline against which every typing number looks slow.

Speaking rate shifts with the setting, but it stays well above typing in all of them. Presentations run slower and more deliberate, around 100 to 150 WPM. Ordinary conversation lands at 120 to 150. Audiobooks and podcasts, read to be easy on the ear, run about 150 to 160. Even the slowest of these clears the average typist.

If you want the flip side of speaking speed – how fast the same words can be read or heard back – we pulled that apart in reading speed vs listening speed. The short version is that talking is a fast way to get words out, and reading is a fast way to take them back in. Typing is the slow step in the middle.

Averages from separate studies are suggestive, but the Stanford, University of Washington and Baidu team put both methods in front of the same people. Thirty-two participants each ran 100 trials – 50 by voice, 50 by keyboard – dictating to Baidu's Deep Speech 2 recognizer or typing on the built-in iOS keyboard.

For English, speech came in at 153 WPM against 52 for the keyboard: 2.93x faster. The result was not an English quirk, either. In Mandarin, speech input reached 123 WPM against 43 for a Pinyin keyboard – 2.87x faster – so the roughly-3x gap held across two very different writing systems.

This is the detail worth holding onto: the comparison was not voice against handwriting or voice against hunt-and-peck. It was voice against a modern smartphone keyboard, the fastest most people ever get, and voice still nearly tripled it.

Speed would be a hollow win if it came with a mess of errors, so the same study tracked accuracy. During entry, speech had a lower corrected error rate than typing – 5.30% against 11.22% – meaning people fixed fewer mistakes as they went. By that measure, voice was not just quicker but tidier to compose with.

The honest caveat: the study also reports that the final uncorrected error rate was slightly worse for speech, 1.30% against 0.79% for the keyboard. So the accurate claim is not "dictation is always more accurate." It is "dictation is cleaner while you correct, and leaves a little more to catch on a final pass." That is exactly why dictated text wants a proofread, not a straight publish.

If you are weighing how much cleanup a machine transcript needs before it is usable, how accurate is AI transcription and word error rate explained walk through the metrics without the marketing gloss.

There is a subtlety the raw WPM numbers hide. Copying text you can already see is faster than inventing it. Reference figures put transcription typing – copying from a source – at about 32.5 WPM, but composition, writing original sentences as you think of them, drops to about 19. Drafting, not copying, is the writing most of us actually do, and it is the slower kind.

That gap is the strongest argument for talking first. When you dictate, you compose at speaking speed instead of composition-typing speed, which is where the keyboard is at its worst. Handwriting is slower still, around 13 words a minute, a useful reminder of how far human text entry lags the voice.

And the keyboard, not the hand, is the bottleneck. Court reporters on a purpose-built chord machine – the stenotype – reach up to 360 WPM, faster than anyone speaks. The hands can clearly go quicker. It is the ordinary QWERTY layout, one letter per press, that holds everyone else to 40-something.

The takeaway is not "never type." It is to split writing into two jobs and use the right tool for each. Capture is a speed problem – get the raw thoughts out before they evaporate – and speech wins it outright at three times the pace. Shaping is a precision problem, and that is where a keyboard, on a transcript that already exists, earns its keep.

So the workflow that respects the numbers is: talk the first pass, turn the audio into text, then edit the text. You compose at 150 words a minute and reserve the slow keyboard for polishing, not for the blank page. A rambling ten-minute voice memo becomes fifteen hundred words you can cut down – far more raw material than ten minutes of typing would ever produce.

The catch is the middle step. Dictation software transcribes live but drifts on long stretches, accents and crosstalk; a recording plus a good speech model handles those better, then you correct once. If your captures run long or noisy, how to improve transcription accuracy covers the upstream habits – mic placement, clean audio, a correction pass – that keep the draft usable.

If your capture tool is a phone that already records voice memos and meetings, the missing piece is turning that audio into an editable draft. That is the job Pepys does: upload the recording, get back a clean transcript in the formats an editor actually eats, and take it from there. It is an audio-to-text step, not a live dictation gadget – which suits the dictate-then-edit workflow the research points to.

None of this is a reason to abandon your keyboard. It is a reason to stop using it for the part it is worst at. Speak the first draft at 150 words a minute, transcribe it, then type only to make it good. The 3x is real – the trick is spending it on the right half of the work.

For the arithmetic of how much text a recording yields, how many words are in an hour of audio does the conversion, so you can guess how big a draft a given voice memo will become.