Speaking vs Typing Speed: What the Research Shows

In controlled studies, speech enters text about three times faster than a keyboard. In the best-known test, English speech entered text on an iPhone at 153 words per minute, against 52 words per minute on the built-in keyboard, 2.93 times faster ^{1 2}. That count included the time spent fixing mistakes. The multiplier is not a fixed number. It changed with the recognizer, the typist's skill and the setting, and the studies below show how.

This guide walks through the studies, from 1999 to 2022 ^{3 4}, says what each one measured, and sets out where typing still wins. For the wider picture of how fast people type and talk, see the typing speed overview. For the short answer, see Is dictation faster than typing?.

Speech is about three times faster than typing in controlled studies

The best-controlled comparison is a 2018 study from Stanford University and the University of Washington. Participants entered short messages on an Apple iPhone 6 Plus, once by speech using Baidu's Deep Speech 2 recognizer and once with the iOS keyboard. The study measured the finished text, so correction time counts against speech as well. In English, speech reached 153 words per minute ¹ and the keyboard 52 ². In Mandarin Chinese, speech reached 123 words per minute, 2.87 times the 43 words per minute of the Pinyin keyboard ⁵.

Speech was not only faster. It also made fewer errors while entering text, but left slightly more errors in the final text. During entry, speech had a corrected error rate of 5.30 percent against 11.22 percent for the keyboard ⁶. In the final text, speech left 1.30 percent uncorrected errors against 0.79 percent for typing ⁶. So speech is faster, and typing is cleaner after the last edit.

The study had limits. Participants read short phrases in a quiet room, and the recognizer dated from 2016. The table shows what the main studies measured.

The four studies do not measure the same thing, which is why their numbers never line up neatly. The 2018 study timed short phrases in a lab and counted the finished text. The 1999 study looked mainly at correction. The 2005 study followed people composing over time, and the 2022 study compared methods on a transcription task with statistics instead of published speeds. Read each row for what it counted before you compare two rows.

Yap, which we make, is an on-device dictation app for Mac and Windows, so we have a stake in this topic. The numbers above come from independent researchers, and we link to each original paper so you can check them.

The chart mixes lab tests with online averages, so compare the bars with care. The 153 ¹ and 52 ² come from the same experiment. The 36.2 ⁸ and 51.56 ⁹ averages come from different people typing in their own setting. The conversational rate counts all elapsed time in telephone speech ¹⁰.

The gap is largest on phones, and for repairing errors it favors average typists over fast ones

Phone typing is slower on average, so speech looks most dramatic against it. In an online study of 37,370 volunteers typing on their own phones, the average was 36.2 words per minute, with 2.3 percent uncorrected errors ⁸. In an online typing test that produced 136 million keystrokes, the average participant typed 51.56 words per minute ⁹.

If speech holds at 153 words per minute, the ratio against each average is about 4.2 on phones and about 3.0 on the online keyboard test ^{11 12}. That is arithmetic across two different groups of people, not a measured result. The practical version is simpler. At 153 words per minute, 1,000 words take about 6.5 minutes to speak ¹³. That is the lab rate, and it leaves out editing.

For expert typists the picture is different. In a 1999 study, the average typing rates on plain text were 23, 35 and 40 words per minute for slow, average and fast typists ¹⁴. The same study measured how fast people repaired recognition errors. Multimodal correction was slower than correcting by typing for typists with good skills ³. Fast non-secretarial typists could correct up to 15 errors per minute by keyboard and by choosing from the recognizer's list ¹⁵.

This suggests that the advantage of speech for fixing errors shrinks as typing skill rises. It does not show that speech input gets relatively slower for fast typists, because the study did not compare entry speed by skill. We found no controlled study that does.

In the late 1990s, correcting recognition errors made dictation slower than typing

Early recognizers made many errors, and the time to fix them could outweigh the speed of speaking. Suhm and colleagues found that participants first preferred speech for correction, then learned to drop the methods that did not work ¹⁶. Correcting by speaking again is also unreliable, because the spoken repair is often misrecognized too ¹⁷.

The composition figures from the same period are the clearest measure of the problem. Feng and colleagues, citing earlier work, reported that composition with speech software was 8 to 15 words per minute, far below normal speech at 125 to 150 words per minute ^{7 18}. Feng and colleagues then followed users over several sessions to see how their productivity changed.

Their 2005 longitudinal study shows where the time went. Navigation and error correction took a large share of the work in the early stages ¹⁹. The largest productivity gains came from better dictation quality and from navigation commands ²⁰. In other words, the largest gains came from the software and from commands that users learned to use.

Modern recognizers changed the result: speech now wins even after corrections

The 2018 phone study already counted correction time, so speech's 2.93 times advantage includes the time spent fixing mistakes ¹. The 2022 study from Simon Fraser University compared the same two methods and found the same direction. Speech was significantly faster in words per minute, with a large effect size ⁴. Typing still had the lower error rate ²¹. Time spent reviewing each phrase did not differ significantly between the two methods ²².

The 2022 paper is a short workshop report, so treat it as one data point. It also reports the statistics rather than the words per minute of each condition, which appear in a figure ⁴. We found no controlled study from 2023 to 2026 that measures speech and typing speed side by side with the newest engines. A reader who needs that number will have to test it, which the last section explains how to do.

Recognizers still differ a great deal. In a 2025 test of one episode of about seven and a half minutes, Apple's speech API had a word error rate of 10.3 percent ²³. Engines differ, so one number from one test is not a verdict on any app. Word error rate counts the substitutions, deletions and insertions in a transcript against a correct version, and it is explained in our glossary entry on word error rate. The same speed advantage can therefore produce very different finished text, depending on the engine.

Composing slows both methods, because thinking sets the pace

Speech is faster than typing at putting words down, but composing is mostly thinking. In a 1978 IBM study of letter writing, about two-thirds of the time spent composing was planning, whether the letter was written, dictated or spoken ²⁴. Speaking faster does not shorten the planning.

Experience helps, but only so much. Dictators with years of experience composed letters between 0 and 25 percent faster than novices, depending on how complex the letter was ²⁵. Planning still takes most of the composing time whichever method a writer uses ²⁴.

This matters for anyone who expects voice input to make writing several times faster. The speed of entering words can rise sharply, while the time spent deciding what to say does not depend on the method. A fast first draft by voice is a real gain, but it is a gain in production, not in thought. For the broader question of how dictation affects writing, see dictation vs typing.

Court reporters set the benchmark for keyboard transcription of speech

Court reporters set the benchmark for keyboard transcription of speech. The National Court Reporters Association's skills test for the Registered Professional Reporter sets the testimony leg at 225 words per minute ²⁶. Stenographers use a stenotype machine, a specialized keyboard (see stenotype). The history and speeds of that method are covered in stenography and shorthand.

The comparison puts speech in context. Ordinary people speak at about 125 to 150 words per minute ¹⁸, and a voice recognizer in a controlled study entered 153 words per minute ¹. The 225 mark ²⁶ is a court-reporting test standard, so it is a benchmark for trained stenographers rather than for typical typists. Dictation does not need that training, and that is the trade-off: the speed is lower, but anyone who can speak can use it.

A fair test of your own speech and typing speed takes ten minutes

The studies above used fixed phrases in controlled settings. You can run a rough version yourself. The key is to count the finished text, which is what the 2018 study did ¹.

1. Choose one passage of about 150 words that you can read aloud. Use the same passage for both tests.
2. Type it on your usual keyboard. Start a timer, type the passage, then fix every mistake until it matches the original. Stop the timer and note the minutes.
3. Dictate it with the speech tool you plan to use. Read the passage aloud, then fix every mistake by the same rule. Stop the timer.
4. Calculate words per minute for each method: the number of words in the corrected text, divided by the minutes, including the time spent fixing errors. The words per minute glossary entry explains the counting rules.
5. Count uncorrected errors in each final text, because the 2018 study found that speech left slightly more of them ⁶.

Record the results in the same shape each time, so the two methods can be compared side by side:

Read your results against the averages, not as a verdict. An average phone typist types about 36 words per minute ⁸, and most people speak faster than that ¹⁸. Your own result depends on your recognizer, your microphone and your room. If you want to try speech without installing anything, free dictation in your browser runs the same kind of test.

Typing still wins for code, small edits, noisy rooms and shared spaces

Speech is not the best method in every setting. Noise is the most measurable limit. A 1996 study reported that the performance of a speech recognition system degrades rapidly in ambient noise ²⁷. The Whisper Large V2 model had a word error rate of 2.7 percent on the LibriSpeech Clean test set ²⁸, but 25.5 percent on the CHiME6 test set ²⁹. The two test sets differ in more than noise, so the gap does not measure noise alone. For more on this, see dictation background noise.

Small edits are the second limit. Correcting a single word by speaking is error-prone, because the correction is often misrecognized ¹⁷. New users also do a lot of correction at first. In a 2013 study of radiologists, new users of speech recognition corrected 15 characters in a report of 433 characters at the start ³⁰. For a one-word change, a keyboard and a cursor are usually faster.

Code is a case where we make no speed claim. We found no controlled study that timed dictation of source code. Symbols and brackets are awkward to say aloud, and that is our judgment, not a measured result. Typing is the safer default there, and voice is most useful for prose and comments.

Shared spaces are a judgment, not a measurement. In an open office, a meeting or a train, speaking aloud can disturb other people and expose what you are writing. A keyboard is quieter, and the speed advantage of speech may not matter when you cannot speak freely.

Put simply, speech is the better choice for a first draft, a message or notes in a quiet room, where you are happy to fix mistakes afterwards. Typing is the better choice for precise edits, code, formulas, and work in places where speaking is impractical.

For short answers to the common questions on this topic, see can you type faster than you talk?, how many words per minute can you dictate? and how do you test your voice typing speed?. The overview of typing speed covers the keyboard side in more depth, including how fast most people type and how typing speed is measured.