About SpeechTime

SpeechTime converts a word count into speaking time and back again. That is one division, and a dozen sites already do it. This one exists because of the number the division depends on.

The problem with the usual number

Ask the web how fast people speak and you get 125–150 words per minute. We looked for something we could read and cite behind that range and did not find it. We are not saying it is wrong — we have no basis for that claim either. We are saying we could not show you where it came from, and building a calculator on a figure we cannot source would make this site one more link in the chain that keeps it circulating.

So the site measures instead. Not the reader — we cannot do that from here — but real recorded deliveries where a full transcript and exact timings are published for the same performance.

How the measurement works

The material is public talk pages on ted.com, which publish a complete English transcript in which every caption cue carries a millisecond offset into the recording. We are not affiliated with TED and nothing here is endorsed by them; theirs is simply the best public material we found where a transcript and exact timings describe the same delivery. The words of one cue are spoken between that cue's timestamp and the next one's, so across the window from the first cue containing speech to the instant the last one ends, the speaker delivered exactly the words in between. Both edges are observed timestamps. Nothing is estimated, interpolated or rounded up.

We ran that over 39 single-speaker talks — 78,841 words across 8.0 hours. The median came out at 165.7 words per minute, with the set spanning 101.8 to 210.3. Against the usual band, 10 of the 39 land inside 125–150 and 28 are faster.

Every talk, its word count, its measured window and its rate are published on the speaking rate page, each linking back to its source. The script that produced them writes the data file directly, so no figure in that dataset was typed in by a person and none can drift from the measurement. Figures that are not measurements — the page descriptions search engines show, the words-per-page and minutes-per-slide divisors, the section splits — are written by hand and labelled as conventions wherever they appear.

What the presets mean

Three of the four pace buttons are percentiles of that dataset, rounded to the nearest five: Presentation is the 10th percentile at 140 wpm, Speech is the median at 165 wpm, Podcast is the 90th percentile at 185 wpm . The fourth is the commonly cited band, offered because people arrive needing to match a figure someone else gave them, and labelled as untraceable wherever it appears.

The names describe situations those rates tend to suit. They are labels on a percentile, not measurements of presentations, speeches or podcasts as categories — we measured recorded conference talks and nothing else.

What we cannot tell you

• Your rate. It depends on your voice, your material, your nerves and your room. The pace meter measures it in about a minute, which is why it sits on five pages.
• Conversational speech. Our sample is prepared, rehearsed delivery. Conversation is a different activity and we have not measured it.
• Other languages. English only. Words carry different amounts of meaning in different languages, so a word-per-minute rate does not travel.
• Whether a rate is good. We report how fast talks were delivered, not how well they landed. Nothing here is advice about how you should speak.
• Filler words. Published transcripts are lightly cleaned, so hesitations are largely absent from the counts. That matches a written script, which has none either, but it means the acoustic rate of those deliveries was somewhat higher than our figures.

How the tables are built

The conversion tables are not selected examples. They are complete cross products — every word count on the axis against every rate on the axis — computed when the site is built and asserted cell by cell in an automated test suite that runs as part of every build. Between the two matrices on the front page that is 1,680 cells, none of them hand-written, which is the only way we know of to publish a table this size and be sure it is right.

The same suite reads the site's own prose as text and fails the build on a specific list of sentence defects: a claimed source we did not open, a retired superlative, a figure typed into a page description that disagrees with the calculator, a speaking rate typed in rather than interpolated, and a gap between two clocks computed in the one order that can disagree with the clocks either side of it. Writing is where mistakes actually live, so it is worth testing — but the list is a list of known defects, not a proof that every sentence is right.

Privacy, in one line

Your script never leaves the tab. There is no server, no account and no upload — the counting and the timing happen in JavaScript on your device. Open your browser's network tab and paste something if you would rather check than trust: advertising and analytics requests may fire on their own schedule, and none of them carries what you typed. The full statement is on the privacy page.

Corrections

If a cell disagrees with the calculator, a claim is wrong, or a source does not say what we say it does, we would rather know. The contact page has the address, and corrections to measured figures get the measurement re-run rather than the sentence softened.