Voiceover speed for videos

A voiceover should be spoken at a natural rate, around 150 to 160 words a minute for English, and a video should then give each line enough scene time for the pictures to land. Those are two different numbers, and mixing them up is where most pacing problems start. Across the 20 video examples on this site, narration ran at a median of 76 words per minute of running time in promos, 84 in tutorials and 98 in creator style ads, roughly half the speaking rate, because scenes hold while people look.

The example under this post is the expense claim how to video, a fictional portal walkthrough in eight scenes, 48 seconds and 85 words of narration. It is the densest tutorial of the five counted here, at 106 words per minute of running time, and it still sounds calm.

What the research says about speaking rate

The most useful study for recorded video is Philip Guo, Juho Kim and Rob Rubin’s 2014 analysis of 6.9 million video watching sessions on edX. They approximated each video’s speaking rate from its time coded subtitles: spoken words divided by speaking time. Rates ranged from 48 to 254 words per minute, with a mean of 156.

The finding runs against common advice. Within each length band, engagement usually rose with speaking rate, by up to two times. The authors note that practitioners often recommend 160 words per minute as the optimum for presentations, a figure first made in 1967 for live lectures, and suggest online viewers can follow faster speech, especially when the same information is on screen. They are careful about the cause: faster speakers also sounded more enthusiastic, so their advice is not to force people to speak faster but to stop them artificially slowing down. The edX producers they interviewed also edited out filler words and pauses.

The takeaway for narration is simple. A natural, slightly brisk read beats a deliberately slow one. Slowness is not clarity.

Words per minute by video type

To see what that means in a finished video, the scripts of all 20 video example pages were counted on 13 September 2026. For each video, the words in every narration line were added up and divided by the total scene seconds, then multiplied by sixty. That gives words per minute of running time, including the moments when a scene holds with nothing said. It is a different measure from Guo’s speaking rate, which only counts the time someone is talking.

Promos, five videos, median 76. The lowest of the whole set is the 48 hour sale video at 55: 16 words over 17.6 seconds, with lines like “Ends Sunday midnight.” given four seconds each. A countdown needs the number to sit on screen, not a paragraph over it.

Tutorials, five videos, median 84. Steps hold for five to nine seconds with six to fifteen words, because the viewer needs time to find the control. The dashboard tour runs at 80.

UGC ads, ten videos, median 98, the highest of the three. The seven English ones have a median of 100. In the skincare serum ad, a talking line such as “Not a cure, just my skin, and I keep buying it.” puts 11 words into 3 seconds, 220 words per minute of scene time, while its proof scene “Then this. Ten percent niacinamide.” runs at 71. Creator speech is fast; the overall pace comes from alternating it with quiet proof shots. Voice style tracks the same split: the six UGC examples with an energetic voice had a median of about 108, the three with a warm voice 86.

Why the count changes by language

Word counts do not compare cleanly across languages, because languages split words differently on the page. The Hindi course ad counts the highest of all 20 examples, 128 words per minute, and part of that gap is how Hindi text divides into words rather than how fast the voice speaks. The Spanish restaurant promo counts 69 and the Portuguese listing 78.

So for any language other than the one you usually write in, time the line instead of counting words. Read it aloud, or play the generated clip, and compare the seconds with the scene length. The per language voice detail, including where quality drops, is in AI voiceover languages and voices.

When to change the speed setting

A speed setting is a finishing tool, not a writing tool. Reach for it in three situations.

A line overruns its scene by a fraction. If a narration clip is half a second too long, 1.1x brings it to about 91 percent of its length and usually still sounds natural. At 1.25x it takes 80 percent of the time, which starts to sound pushed on long sentences.

A line has to be understood exactly. An instruction with a control name, a price or an offer code can go to 0.9x, which stretches it to about 111 percent, so a viewer catches “Submit for approval” or a discount code on the first listen.

The voice should start later. A delay before the line lets a picture establish itself first, which often fixes a scene that felt rushed without touching the speed at all.

What a speed setting should not do is rescue an overwritten line. Cutting three words from a twelve word line saves more time than the fastest setting, and it keeps the delivery relaxed.

Captions and pace

Burned in captions follow the narration, so the voice sets their reading speed. The BBC’s subtitle guidelines recommend 160 to 180 words per minute, or 0.33 to 0.375 seconds per word, and suggest leaving a subtitle up for at least about 0.3 seconds per word. The DCMP Captioning Key is stricter for education: no more than 130 words per minute for lower level educational media, 140 for middle and 160 for upper level.

A talking line at 220 words per minute of scene time is past all of those. That is acceptable for a three second creator aside that viewers hear, and a problem for a tutorial step that someone watching muted has to read. The W3C’s guidance is a reminder of who depends on captions: people who are Deaf or hard of hearing need a text version of the speech and important sounds. If captions are part of how your audience watches, keep individual lines under about 160 words per minute of scene time.

Common mistakes

Slowing the voice to sound clear. In Guo’s study, engagement usually rose with speaking rate. Clarity comes from short lines.

Speeding up instead of cutting. 1.25x on a long sentence sounds hurried; the same sentence minus a clause does not.

Judging pace by word count in another language. Time it in seconds.

Filling every second with speech. A countdown or a price needs a silent beat. Pace is words and pauses together.

Forgetting muted viewers. Captions inherit the narration’s speed, so a line nobody can read in time is lost to them.

Energetic reads on instructions. The tutorial examples all use instructional voices for a reason: steps need a measured delivery. For creator ads the opposite holds, as what a UGC ad is explains.

Build it

In the builder, every scene’s narration has its own voice, a delay in seconds before the line starts and a speed of 0.85x, 0.9x, 1x, 1.1x or 1.25x, along with a caption toggle that is on by default. Captions are derived from the narration and burned into the picture as a dark pill at the bottom of the frame, inset above the safe area. Voices come in four styles across two genders, and a tutorial is never given the energetic style. Tutorial steps are planned at 5 to 9 seconds with 10 to 25 words, promo scenes at 2.5 to 4 seconds with 5 to 8 words. The voice options are on AI voice options and the steps are in add a voiceover, captions and music to a video.