150 words a minute against 40.
The gap between how fast you can say a sentence and how fast you can type it is the whole argument for dictation. It is also widely misread, so here is what it does and does not buy.
The round numbers this site uses. The measured figures are below, with their studies: speech at 161 words a minute, typing between 32.5 and 51.6 depending on whether you are composing or copying.
What the research actually measured
The typing side has a very large answer. In Observations on Typing from 136 Million Keystrokes, Dhakal, Feit, Kristensson and Oulasvirta recorded 168,000 people and put the average at 51.6 words a minute. That is typing text that is already in front of you. When people have to compose rather than copy, the number falls a long way: the CHI 1999 study by Karat, Halverson, Horn and Karat measured 32.5 words a minute for a keyboard and mouse.
The speech side was measured directly against a keyboard by Ruan, Wobbrock, Liou, Ng and Landay in Speech Is 3x Faster than Typing for English and Mandarin Text Entry on Mobile Devices: 161.20 words a minute for English speech against 53.46 for typing, on the same task with the same people.
So the honest version of the headline is a range rather than a pair of round numbers. Speech lands around 150 to 161. Typing lands between 32.5 and 51.6, and which end you are at depends entirely on whether you already know what you are going to write. The gap is between two and five times, and it is widest exactly where most working writing sits: composing something new.
Measure your own, it takes four minutes
Averages are a starting point, not an answer. Do this instead. Pick something you actually have to write today: a reply, a summary, a handover note. Set a timer for two minutes and type it. Count the words. Then take a second, similar task, set the timer again, and say it out loud as if you were explaining it to someone across the desk. Count those words. Now you have your two numbers and the ratio between them, for your work rather than for a study population.
Most people are surprised twice. The typing number is lower than they thought, because it includes the pauses where they were deciding what to type. The speaking number is higher than they thought, because talking does not have those pauses in the same places.
What the gap actually buys
It does not make you finish a document four times faster. Writing is thinking, editing, checking and formatting, and none of those get quicker because you said the words instead of typing them. What the gap buys is the part in between: the stretch where you already know what you want to say and your hands are the only thing in the way.
That stretch is most of a working day for a lot of people. Replies. Commit messages. Handover notes. The paragraph explaining the decision. The summary nobody wants to write. Those are transcription tasks dressed up as writing, and they are where the three times number shows up honestly.
Why the number on the page is not the number that matters
Here is the part most comparisons skip, and it is the part the 1999 study is really about. Speech recognition in 1999 could take dictation at speed and still lose to a keyboard once corrections were counted, because fixing a wrong word cost more than typing the right one. A dictation tool that can do 150 words a minute but delivers them a second late is slower to use than typing, because you stop trusting it. You start waiting to see what appears. You start saying the sentence more carefully. The moment you do that, you are no longer talking at 150 words a minute.
The threshold is roughly a quarter of a second. Under it, the words feel like they came from you. Over it, they feel like they came back from somewhere. Zumbo composes on your own Mac in about a fifth of a second, which is on the right side of that line, and it is on the right side of it with the network off, because there is no network in the path at all.
The second thing that eats the gain
Wrong words. If a term in your field comes out wrong every single time, you pay for it twice: once to notice and once to fix. Three or four of those in a paragraph and the whole advantage is gone. This is why a general speech model is not enough for professional writing, and why Zumbo ships twenty one vocabulary packs and lets you correct a word once with Teach so it is heard your way from then on.
Ship the Q3 roadmap before Friday.
What to do with this
Take your two numbers from the exercise above. If the ratio is small, you are a fast typist doing writing that is mostly thinking, and dictation will not change your day much. If it is large, and most of what you write is a thought you have already had, then the keyboard is the bottleneck and you can simply remove it.
Related: the Mac dictation apps side by side, dictation for developers, and dictation with the Wi-Fi off.
Where do the numbers come from?
The measured ones are on this page with their studies. Typing averaged 51.6 words a minute across 168,000 people in the CHI 2018 keystroke study, and 32.5 words a minute in the CHI 1999 composition study where people were writing rather than copying. English speech was measured at 161 words a minute in the 2016 Stanford study. The 150 and 40 in the headline are the round numbers this site uses; the range above is what the research actually reports.
Is speaking really three or four times faster?
For getting a first version of a sentence out, yes, and the Stanford study measured almost exactly three times. It is not three times faster to finish a piece of writing, because thinking, editing and formatting are unchanged. The gain is in the part where you are only transcribing your own thought.
Does dictation make me write worse?
It makes a different kind of first draft: longer, looser and closer to how you talk. For messages, commit notes and replies that is usually an improvement. For anything that gets published, it is a draft to edit, the same as any other.
What slows dictation down in practice?
Two things. Latency, if the words arrive late enough that you start second guessing them. And wrong words, if you have to go back and fix the same term every time. Both are solvable, and both are what Zumbo is built around.
Do I need to talk in a special way?
No. Talk at the speed you talk. Punctuation is inferred, fillers are dropped, and you can restart a sentence part way the way people do.
Free for 3 days, everything unlocked. Then $15 once. macOS 14 or later, Apple silicon.