Words per minute speaking calculator for voiceover scripts
Updated November 6, 2026
I used to plan my voices at 130, 150 or 170 words per minute. Then I timed 233 of their Eleven v4 clips. I measured my voices at 178 words per minute overall, faster than any of those guesses, and that page lists every voice’s number. For a script read by one of my voices, plan on about 190 for the quick ones, about 175 for the middle group and about 155 for the two unhurried ones. Human speakers vary a lot more, from person to person and from one situation to the next. Paste your script to see how long it runs at each pace, or choose a length and get a word budget.
Typical speaking rates
| Kind of speech | Words per minute | Where the number comes from |
|---|---|---|
| Everyday conversation, measured | about 173 on average | Phone conversations in the Fisher English corpus, analyzed on Language Log. Individual speakers spread widely around that figure. |
| Conversation, common planning figure | about 150 | Widely cited for US English and attributed to the National Center for Voice and Speech |
| Audiobooks, radio, podcasts | 150 to 160 | VirtualSpeech’s estimate |
| Human narration and explainers | about 150 | A common rule of thumb for planning a human read |
| Human fast social read (Shorts, Reels, TikTok) | about 170 | A rule of thumb I used before I measured my voices |
| Human relaxation, sleep, meditation | about 110 to 130 | A rule of thumb, since the pauses carry the read |
| My 15 ElevenLabs voices on Eleven v4 | 178 overall | Measured from 233 clips at default settings |
People are surprised by the measured conversation figure. My guess is that casual talk runs fast because nobody is pacing it for a listener. Narration should go slower on purpose, since the listener can’t stop you to ask what you meant. That’s why human voiceover is usually planned below the speed people chat at. My AI voices ignore that habit. In my clips they came in at 178, a little above the Fisher conversation average.
How many words fit in a given length
These are rounded down so the script fits.
| Length | Unhurried, 155 wpm | Medium, 175 wpm | Quick, 190 wpm |
|---|---|---|---|
| 15 seconds | 38 | 43 | 47 |
| 30 seconds | 77 | 87 | 95 |
| 60 seconds | 155 | 175 | 190 |
| 90 seconds | 232 | 262 | 285 |
| 3 minutes | 465 | 525 | 570 |
| 5 minutes | 775 | 875 | 950 |
| 10 minutes | 1,550 | 1,750 | 1,900 |
For tutorials and screen recordings, knock about 20 percent off. Viewers need quiet moments to watch each click, and narration that never stops gives them no time to look.
How AI voices differ from people
The voice sets its own pace. ElevenLabs says the pacing of generated audio is highly influenced by the audio used to create the voice. Hand two of my voices the same 1,500 words and they can finish more than two minutes apart. At their measured rates, Drift would take about ten minutes and Reel under eight. My fifteen voices fall into three groups by measured rate.
- Pulse, Reel, Dexter, Timber and Bilbop are quick, at 185 or more. I plan them at about 190.
- Snap, Derek, Jett, Heisenberg, Havoc, Reverend, Felix and Titan sit in the middle, from 170 to 184. I plan them at about 175.
- Titan Epic and Drift are unhurried, under 170. I plan them at about 155.
Pulse, Snap, Bilbop and Drift each rest on six clips or fewer, so their places in these groups are rough.
Punctuation adds time. Eleven v3 and v4 have no break tags, so pauses come from full stops, line breaks and ellipses. A script chopped into many short lines runs longer than its word count suggests, and a dense block of text runs shorter.
Numbers hide words. On the page, “2026” is one word. Spoken, it’s three (“twenty twenty-six”). A price or a percentage can turn into five spoken words, so spell numbers out before you count or the estimate will come in short. My voiceover script guide covers how I write them.
Takes vary a little. Nerves never speed an AI voice up. Even so, every generation is a slightly different performance, and ElevenLabs’ Stability setting controls how different. The same script can land a touch longer or shorter from one take to the next.
Using the result in ElevenLabs
- Budget the script here, then generate one take. Don’t spend an hour polishing a script you haven’t heard once.
- Measure the real pace. Divide the word count by the take’s length in minutes. A 150-word script that runs 58 seconds is reading at about 155 words a minute. Use that figure for the same voice next time, because it beats any table, mine included.
- Fix length in the script. ElevenLabs’ v4 docs say the Style and Speed sliders aren’t available on Eleven v4. On other ElevenLabs models a speed setting runs from 0.7 to 1.2, and ElevenLabs warns that extreme values can affect quality. Either way, I would cut a sentence before I’d squeeze a read.
- Split long scripts. Through the API, Eleven v4 takes up to 10,000 characters per request, which ElevenLabs’ models page puts at roughly ten minutes of audio. The website caps each generation separately, and that cap depends on your plan. For anything long I split at paragraph breaks and generate in sections.
When the pace keeps coming out wrong for the job, the fix is usually a different voice. My narrator voices and TikTok voices pages sort the voices by the kind of read you need, and Eleven v4 vs v3 explains which controls you still have on each model.