How fast people talk depends on the situation: measured lecturers run about 125–160 words per minute, British conversation about 190–230. Across 2,438 recorded American phone calls in the Switchboard corpus, people averaged 196 words per minute speaking, and a larger corpus in the same study came in at 193. The average typist in a 168,000-person study managed about 52. Set side by side, those studies put speech at roughly three to four times the speed of typing. Dictation — speaking to a computer and keeping the text — lands somewhere between the two, and the reasons matter before you plan around the gain.
I build a dictation app, so I have an obvious interest in this question. Every number below links to its source; where I could only reach a study through another paper's summary, I say so, and where the research runs out I say that too.
Average words per minute speaking, by situation
How many words per minute people speak depends mostly on context: conversation is fastest, and lectures and presentations are slowest. These are published figures, not calculator defaults:
| Situation | Words per minute | Source |
|---|---|---|
| Phone conversation, American English | 196 average (111–291 across conversations) | Yuan, Liberman & Cieri, 2006 |
| Same conversations, silences removed | 236 average | Yuan et al., 2006 |
| Conversation, British English | 190–230 | Tauroza & Allison (1990), as summarized by Huang & Gráf, 2020 |
| Interviews | 160–210 | same |
| Radio broadcasters | 150–170 | same |
| Lecturers | 125–160 | same |
| Audiobook narration | 140–180 | Brysbaert, 2019, stated in passing rather than measured |
| Reading aloud | 183 average | Brysbaert, 2019 (77 studies, 5,965 participants) |
Two findings from the Yuan study change how you should read these numbers. First, very short turns are slow: speaking rate rises steeply as turns grow from one to seven words, then holds roughly level for turns of about eight to thirty words. Second, the average rate depends on how you count. Measured against total call time, the corpus averaged 196 wpm; with silences stripped out it was 236. Any single "average speaking rate" quietly picks one of those definitions.
The same paper reports that older speakers talk more slowly, men slightly faster than women, and that people use longer but slower turns when talking with strangers.
How fast do people talk compared with typing and reading?
Talking is about three to four times faster than average typing when you set these studies side by side, and silent reading is faster than both. The one head-to-head lab test measured 2.93 times. Here are the numbers together:
| Activity | Words per minute | Source |
|---|---|---|
| Typing on a physical keyboard, average | 52 (SD 20) | Dhakal et al., 2018, 168,000 volunteers |
| Typing, fastest participants | 120 or more | Dhakal et al., 2018 |
| Typing on a phone keyboard, lab | 52 | Ruan et al., 2017 |
| Speech recognition on a phone, lab | 153 | Ruan et al., 2017 |
| Conversation | 196 | Yuan et al., 2006 |
| Reading aloud | 183 | Brysbaert, 2019 |
| Silent reading, non-fiction / fiction | 238 / 260 | Brysbaert, 2019 |
The average typing speed figure deserves a caveat. Dhakal's participants were self-selected volunteers on an online typing test, copying sentences shown on screen. That measures how fast fingers move, not how fast someone writes an email they have to think about.
The phone study is the closest thing to a head-to-head test. It compared the iOS keyboard against Baidu's Deep Speech 2 recognizer, in a quiet room, with participants copying short phrases. English speech came out at 153 wpm against 52 for the keyboard, 2.93 times faster. (A preprint of the same paper is on arXiv.)
Words to minutes: how long a speech or dictation takes
To convert words to minutes, divide the word count by the speaking rate. At presentation pace, a 1,000-word speech takes about 6 to 8 minutes to say; typing it at 52 wpm takes about 19.
There's no single answer to how many words per minute a speech runs, so I used three speaking rates: 130 wpm (a common default in online speech calculators, and inside the 125–160 range measured for lecturers), 150 wpm (inside both the lecturer and radio-broadcaster ranges above), and 196 wpm (Yuan's conversational average). The last column is the 52 wpm average typist.
| Words | 130 wpm | 150 wpm | 196 wpm | Typing at 52 wpm |
|---|---|---|---|---|
| 100 | 0:46 | 0:40 | 0:31 | 1:55 |
| 250 | 1:55 | 1:40 | 1:17 | 4:48 |
| 500 | 3:51 | 3:20 | 2:33 | 9:37 |
| 750 | 5:46 | 5:00 | 3:50 | 14:25 |
| 1,000 | 7:42 | 6:40 | 5:06 | 19:14 |
| 1,500 | 11:32 | 10:00 | 7:39 | 28:51 |
| 2,000 | 15:23 | 13:20 | 10:12 | 38:28 |
| 3,000 | 23:05 | 20:00 | 15:18 | 57:42 |
Times are minutes:seconds. For a speech, these are the numbers to plan with; to use your own rate, measure your words per minute speaking first. For dictation, treat the speaking columns as a floor you won't reach, for the reasons in the next section.
Why dictation speed is lower than speaking speed
Dictation is slower than speech because it adds thinking time and error correction, and neither appears in speaking-rate studies.
The famous numbers measure copying, not composing. Ruan and colleagues chose a copying task on purpose, to isolate the input method from the writer. Composition brings in thinking time and word choice. When you dictate a real email, you pause to decide what comes next. Those pauses count against your effective rate.
Corrections cost more than they look. In the same phone study, English speech made fewer mistakes while people were entering text (a 3.93% corrected error rate against 4.72% for the keyboard), yet slightly more of its mistakes survived into the finished text (0.55% against 0.35%).
Older desktop studies show how expensive fixing can get. In Koester's 2003 study of 23 experienced speech recognition users, all of whom had physical disabilities affecting keyboard use, text entry averaged 16.7 wpm, with a range of 3.5 to 31.7. On average, 37% of task time went to correcting recognition errors, and recognition accuracy ranged from 72% to 94%. Ruan's paper also cites a 1999 study by Karat and colleagues in which speech users managed 7.8 wpm composing and 13.6 wpm copying.
Those recognizers are two decades old, and today's error rates look nothing like that. The pattern still holds: a misheard word takes a second to say and many seconds to find and fix.
There is no good modern benchmark. I couldn't find a recent peer-reviewed study that measures desktop dictation of composed text, corrections included, on current speech models. Anyone quoting a precise "dictation speed" for real writing is either measuring themselves or guessing.
It's plain arithmetic: effective rate = words ÷ (speaking time + pauses + fixing time). As an illustration, not a measurement: 300 words spoken in two minutes, plus one minute of thinking and one minute of fixing, is 75 wpm. That is well under speaking speed, and still faster than most people type. To check whether dictation is faster than typing for you, time both on the same text.
Where dictation beats typing, and where it doesn't
Dictation wins on long, loosely structured prose. Typing wins wherever exact characters matter.
Dictation tends to help with:
- First drafts of emails, documents, and posts, where getting words down matters more than getting them right
- Long prompts for AI tools, where more context produces better results
- Notes and ideas captured before they fade
- Any writing when your hands or wrists need a break
Typing usually stays faster for:
- Code, file paths, URLs, and anything with symbols
- Tables, lists of numbers, and precise formatting
- Short replies, where starting dictation takes longer than typing three words
- Small edits inside existing text
Most people who stick with dictation end up mixing the two: speak the draft, type the fixes. If writing is most of your day, dictation software for writers goes deeper on that workflow.
Getting your dictation speed closer to your speaking speed
Much of the gap between speaking and dictating is correction time, so the biggest gains come from fewer misheard words and less fixing mid-flow.
- Speak in full sentences. Short bursts are slow, which matches Yuan's finding that turns of seven words or fewer are spoken most slowly.
- Don't edit while you talk. Dictate the whole thought, then fix it in one pass.
- Fix recurring errors at the source. If a name or term is misheard every time, teach the tool once instead of correcting it daily. Why Mac dictation gets words wrong covers the common causes.
- Start free. macOS Dictation is built in. Apple says you can "dictate text of any length without a timeout", and it stops after 30 seconds without speech.
If you want more control, Vext, the app I build, targets the correction-time part of the gap. Custom vocabulary rules ("heard as" → "replace with") apply to every transcript, so a name you fix once stays fixed. The Clean up style removes filler words and false starts while keeping your wording as close as possible to what you said, and the raw transcript is stored next to the cleaned version. More on that trade-off in AI text cleanup. For long takes, hands-free dictation lets you toggle recording on and off instead of holding a key.
Download Vext costs $49 once, with a free trial of 100 dictations. It needs macOS 14+ on Apple Silicon. Or install it with Homebrew:
brew install muvon/tap/vext