Is speaking 3x faster than typing? The research, by language

The "3x" comes from a smartphone study by Stanford University and others

The study most often cited for "speaking is three times faster than typing" is by researchers at Stanford University and elsewhere. A preprint came out in 2016, and a peer-reviewed version was published in a journal in 2017¹.

If you cite it, cite the peer-reviewed 2017 version. The numbers are slightly lower, but "about 3x" is a fair summary.

But what it tested was copying short sentences on a smartphone

When reading this study's numbers, pay attention to what was measured:

• Compared with an on-screen smartphone keyboard: the device was an iPhone 6 Plus, not a computer keyboard
• A copying task: participants typed out short sentences shown to them, not text they were composing
• Seated, in a quiet room
• On errors, speech had fewer corrected errors along the way (5.30% vs 11.22%), but more errors left at the end (1.30% vs 0.79%)

Whether speech is 3x faster when you compose your own text on a computer is something this study can't tell us.

On a computer, combining separate studies gives roughly 3.8x (a rough estimate)

For typing on a computer, there's a large-scale dataset of 168,960 people copying English sentences²:

• Average: 51.56 words per minute (standard deviation 20.20)
• The top 10% type about 78 wpm or more; the bottom 10% under about 26
• 85% of participants were native English speakers, and almost all used QWERTY

Meanwhile, a study of 2,438 English telephone conversations found an average speaking rate of 196 words per minute³.

Dividing one by the other gives 196 ÷ 51.56 ≈ 3.8x (calculated). But this combines two separate English-language studies with different participants and tasks, so treat it as nothing more than a rough estimate.

For Japanese, we found no comparison of speaking and typing using today's speech recognition

For Japanese, we found no study that compared speaking and typing with today's speech recognition, using the same people on the same task. Older comparisons exist (one from 2014, for example, with recognition accuracy of about 77%), but their premise is too different from today's recognition, so we don't use them here. There are separate figures:

A mora is a beat of sound in Japanese. Morae don't match the number of written characters, so the two figures can't simply be divided. The certification figure is a minimum passing line, not an average.

Typing Japanese adds the step of converting kana into kanji. Since typing takes longer than in English, speaking may have a bigger advantage, but we found no published study that measured the cost of conversion itself.

Saying the same thing takes about as long in any language

One study compared speaking rates in 17 languages by having people read aloud texts with the same content⁶. We calculated per-language values from the paper's supplementary data.

Syllable rates differ by almost 2x between languages, with Japanese at the fast end. But once you multiply by how much information each syllable carries, the "information rate" differences become much smaller.

In other words, saying the same thing takes about the same time regardless of language. Languages spoken quickly carry less information per syllable; languages spoken slowly carry more.

What this tells us is only that the speaking side doesn't vary much by language. The study measured the information rate of speech read aloud, not typing or input speed. Typing speed should vary a lot with the writing system and input method (kana-kanji conversion, pinyin conversion, Hangul, Arabic script and so on), but apart from English and Chinese, we found no published figures measured across large numbers of people.

Beyond English and Chinese, these were the studies we found that directly compared speaking and typing⁷:

• Arabic: 40 people wrote emails on a desktop PC. Including corrections, speech was about 1.4 to 2.3 times as fast as keyboard and mouse (the voice screen also had commands that insert greetings and the like)
• Hindi: 20 beginners. Adding speech to the keyboard was 2.5x faster (confirmed from the abstract; we did not verify the figures in the full text)

"3x" holds for English and Chinese on smartphones, and no further

• Copying short sentences in English or Mandarin on a smartphoneConfirmed (about 2.9x)
• English on a computerAbout 3.8x by combining separate studies (rough estimate)
• JapaneseNo study found comparing the two with today's speech recognition
• Other languagesSpeaking time varies little by language; typing data is lacking

SayonaraType's English website says speaking is "about 3× faster than typing," and states the condition next to it: on a smartphone, when copying short messages (Ruan et al. 2017). That is the condition the evidence supports. On a computer, or when you compose your own text, we don't have a measured English figure, so we don't claim one. Our Traditional Chinese page gives the same "about 3×" with the same smartphone condition (the study used Mandarin with a pinyin keyboard). Our Japanese page says speaking is "about 3× faster than typing" as a comparison of rates: speaking at about 340 to 440 characters per minute versus typing at about 70 to 120. That compares speeds, not the measured time it takes to finish a piece of writing. Our pages in the other 11 languages give no multiplier and simply say speaking is faster than typing, because we found no study showing "3×" in those languages. Once we have figures measured for each language, we will update both this article and the website. The basis is listed on our Sources page.