Is 2x Speed Bad for Audiobooks? What the Evidence Actually Says

The short answer, with the caveats attached

If you listen at 2x and quietly suspect you absorb nothing, the evidence is on your side, with conditions. The most cited experiment on the question, published in Applied Cognitive Psychology in 2022, put 231 undergraduates through recorded lectures at four speeds and found no statistically significant comprehension gap between normal speed, 1.5x and 2x. Only 2.5x produced a reliable drop. That result has been repeated on audio without pictures, and a 2026 study of 328 adults reached the same conclusion at 2.0x.

Three other things are true at the same time, and the confident version of the answer ignores all of them.

So 2x is defensible for a large share of what people listen to and a bad setting for a smaller share that matters more. Getting there gradually decides which you get, and the method lives in our guide to the best audiobook listening speed and how to raise yours. This page stays on the evidence.

Where the evidence sits, speed by speed

Narration is recorded at roughly 150 to 160 words a minute. Multiply by your playback speed for the effective rate your ear handles, which is what the research is about. Old and new literature disagree about where that rate becomes a problem.

Audible supports 0.5x to 3.5x in 0.1 steps across its apps, the web player and Alexa, so the dial runs far past anything a researcher has validated. Apple Books stops at 2x. Our comparison of where to listen to audiobooks covers the controls each service ships, and how to listen to audiobooks handles setup.

One framing to fix before we go further. The question is not whether fast listening matches slow listening. It is whether the loss is worth the hours. A 13 hour book at 2x hands back six and a half hours, and if that costs three percent of a comprehension score on something you were hearing for fun, the trade is fine. On a book you will be tested on it is terrible. Everything else is detail on that one calculation, which is why we built the speed calculator next to it.

Sixty years of compressed speech research, before anyone had an app

People were speeding up recorded speech long before Audible existed. The work started in the 1950s and 1960s, driven by reading machines for blind students and by military training. Emerson Foulke and Thomas Sticht reviewed the field in Psychological Bulletin in 1969, and their summary set the number that dominated thinking for fifty years.

That number is 275 words a minute. Writing shortly afterward, Sticht put it plainly: speech may be accelerated up to around 275 wpm without seriously impairing the comprehensibility of the message. Push to 300 wpm and comprehension fell significantly against a normal 165 wpm baseline. On a 155 wpm audiobook recording, 275 wpm is 1.77x. The old literature, in other words, drew the line at about 1.75x.

Two details of that work matter, and they point opposite ways. The first is that Sticht put the bottleneck in the listener, not the machine: the factors limiting the comprehensibility of rapid speech, he wrote, reside more with the inability of the listener to process rapid rates of speech than with the signal distortion produced by the equipment. Inconvenient for anyone explaining the modern results away by blaming old tape machines.

The second detail undercuts the first. A 1998 paper by Michele Covell, Meg Withgott and Malcolm Slaney at Interval Research, describing a system called Mach1, noted that comprehension of linearly compressed speech often breaks down above 225 to 270 words a minute, far short of the 500 wpm people manage when a human simply talks fast. Their explanation: uniform compression is the wrong operation. A person speeding up crushes pauses and unstressed syllables and leaves stressed vowels roughly intact. An algorithm that shrinks everything equally destroys that structure.

Mach1 compressed pauses hardest, stressed vowels least, unstressed material in between. Tested at global rates between 2.56x and 4.15x it produced a 17 percentage point comprehension gain over linear compression at the same overall speed, and listeners chose it 94.8 percent of the time. Part of what feels wrong at high speed is an artifact of how the compression is done, not a limit of your brain.

Why the old ceiling and the new results disagree

If the old literature capped useful listening near 1.75x and modern experiments find nothing wrong at 2x, one is measuring something the other is not. Four differences do the work. Old studies compressed uniformly with mechanical methods, while digital time-scale modification preserves pitch and handles transitions better. Old studies often used free recall, modern ones use multiple choice recognition, an easier task. Old participants had never heard compressed speech, modern ones grew up on 1.5x podcasts. And lecture video supplies slides, which prop up an overloaded ear in a way an audiobook cannot.

The recognition point changes how you should read every percentage here. Picking the right answer from four options is not the same as remembering a book. For the wider argument about what listening delivers compared with print, our piece on whether audiobooks count as reading works through the comparison studies, including the ones that found no difference and the ones that did.

The 2022 study everyone quotes, read properly

Here is the paper behind nine out of ten confident internet claims about 2x speed. Dillon Murphy, Kara Hoover, Karina Agadzhanyan, Jesse Kuehn and Alan Castel, “Learning in double time: The effect of lecture video speed on immediate and delayed comprehension”, in Applied Cognitive Psychology in 2022, from Alan Castel’s memory lab at UCLA. It deserves better than the one line summary it gets.

The setup: 231 UCLA undergraduates aged 18 to 41 watched one of two recorded lectures at 1x, 1.5x, 2x or 2.5x. One ran 12 minutes 56 seconds with 2,031 spoken words, the other 14 minutes 27 seconds with 2,403. Do the division and the instructor was speaking at about 157 and 166 words a minute, almost exactly narration pace. That coincidence is why the study transfers to audiobooks at all.

Participants watched in fullscreen, could not pause and could not take notes. Then they answered 20 multiple choice and true or false questions, once immediately and again a week later, with different versions each time. A separate control group of 123 people answered the questions without watching anything, which is the most useful number in the paper and the one nobody repeats.

The actual scores

Proportions correct, with the never watched control included so you can see how much of each score is learning and how much is guessing.

Read the middle rows carefully. The 1.5x group scored slightly lower than the 2x group, which is not what a clean speed penalty looks like. It is what noise looks like. The only pairwise comparison to reach significance was 1x against 2.5x, at d = 0.24, small by any standard. Speed and time of test did not interact: the week long delay hurt every group by roughly the same six points.

The experiment nobody mentions

The same paper ran a second study that is more useful than the first. One group watched the lecture once at normal speed, the other watched it twice at 2x, which takes identical clock time. Tested immediately they were indistinguishable, .64 against .63. Tested a week later, with the second 2x viewing placed just before the test instead of back to back, the double speed group won clearly: .63 against .55, d = 0.54.

That is a real argument for fast listening, and not the one fast listeners usually make. Speed did not improve a single pass. It made a second pass affordable, and spacing the two beat one careful pass by a wide margin. For audiobooks the equivalent is re-listening at 2x a month later, which costs the same clock time as one slow pass and leaves you with more. It also argues for keeping titles, so check what you keep if you cancel Audible before assuming a re-listen will be there, and think twice about deleting books from your Audible library.

The limitations, stated plainly

None of that makes the study weak. It makes it narrow, which is what a good experiment is. The error sits in the leap from “no significant loss on a 13 minute appraisals lecture” to “2x is free on everything”, a leap made constantly by people recommending whichever audiobook they think you should start with and by services like Audiobooks.com selling the format on convenience alone.

Does it hold when there is nothing to look at?

The obvious objection to the lecture research is the slides. Strip the pictures away and you have an audiobook, so somebody had to test audio alone. In 2024 a team including the same lead researcher did: Chen, Kumar, Varkhedi and Murphy, “The Effect of Playback Speed and Distractions on the Comprehension of Audio and Audio-Visual Materials”, in Educational Psychology Review.

Their first experiment used podcast clips, one on the biology of anorexia at 2,289 words and one on black hole mergers at 2,412. Two hundred undergraduates listened at 1x, 1.5x, 2x or 2.5x and answered 30 multiple choice questions per topic, immediately. No slides, no instructor on camera, nothing but a voice: the closest any published study comes to an audiobook.

The overall test came out at F(3,219) = 2.40, p = .069, a null by the usual convention and a suspiciously orderly one. Every step up the dial moved the mean down, and the gap between top and bottom landed a hair outside significance. If you want to argue that audio without pictures degrades more smoothly than the lecture data suggested, this table is your best support, and it is not proof.

Two more things. Scores sat low across the board, between .40 and .46, so the podcasts were hard. And the authors, having found no significant effect anywhere, still wrote that they do not advise exceeding 2x because performance tends to decline quickly past that point. When the people who ran the study and got a null attach that warning, the warning is the finding.

A third study arrived from a different direction. In 2026 Osorio-Andrade, Rodriguez Orejuela and Arango Espinal published work in Frontiers in Education using 328 participants aged 18 to 73, a far broader range than the undergraduate samples, on a 16 minute 42 second technical lesson at 1.0x, 1.5x or 2.0x. Objective comprehension did not budge. Perceived clarity, measured separately, dropped sharply at 2.0x. The authors call it a dissociation between the learning outcome and the subjective experience of processing.

What happens when you pool every study

Single experiments answer narrow questions and often disagree. Meta-analysis is how a field settles an argument, and two arrived recently. Neither flatters high speed.

The bigger is Tharumalingam, Roberts, Fawcett and Risko, “Increasing video lecture playback speed can impair test performance: a meta-analysis”, in Educational Psychology Review in 2025. Twenty four studies, 110 effect sizes, and the headline is in the title. The qualifier: the cost stays small, and often not significant, at 1.5x and slower. Above that it becomes measurable.

Two of their null results inform as much as the main effect. Test type did not moderate the outcome, so the penalty was no larger on delayed tests than immediate ones, and lecture length did not moderate it either. Little that theory said should matter did much moderating, which points toward individual differences and context doing the real work.

The second, by Yang, Du and Huang in Frontiers in Psychology in 2025, pooled 12 studies, most of them Chinese language work that rarely reaches English discussion of this topic. The results run the same way and are blunter about the mechanism.

Sit with the cognitive load number. Effort rose substantially while performance fell modestly, the signature of compensation: listeners worked harder to stay level. That fits what heavy fast listeners report, that 2x is fine but tiring, and it explains why a speed you sustain for twenty minutes falls apart in hour three of a drive. It is also why the sensible ceiling is lower for a book you want to enjoy.

How do you square a meta-analysis saying speed impairs performance with three studies saying it does not? Individual experiments are underpowered for small effects, and a gap of three or four percentage points needs a far bigger sample than 200 people to detect. A null result is not evidence that two conditions are equal. Pooling recovers what single studies missed. The cost of 2x is real and small, not zero, and small costs are worth paying for large time savings, which the speed calculator puts in hours.

Immediate comprehension versus what survives a week

Almost every claim about playback speed rests on a test taken minutes after the audio stopped. That is the wrong measurement for a book you want to still know next month, and it is where the folk wisdom expects to be vindicated. The data is more awkward.

Start with what is established. In the lecture study, delay cost every group about six points regardless of speed, and the test for an interaction between speed and time of test came back flat. The 2025 meta-analysis looked across two dozen studies and also found no moderation by test type. On current evidence, listening fast does not make forgetting faster. Anyone claiming the research shows fast listening wrecks long term retention is describing a study that has not been published.

Now the awkward part, which is arithmetic and not a finding. Subtract the never watched control score of .41 from every cell and you get the portion of each score attributable to watching at all.

Treat that table with suspicion, and anyone quoting it without this sentence with more. Subtracting a baseline magnifies noise, none of these differences were tested for significance, and the 2.5x row breaks the pattern by holding on better than 2x, a hint that the ordering is not real. The table is good for scale. A week later the 2x group sat about twelve points above people who never saw the lecture. That is what one pass leaves behind, and it is not much at any speed.

The conclusion is not “slow down”. It is “one pass is not a plan”. Two spaced passes at 2x beat one careful pass at 1x on a delayed test, decisively. If you want a book to stay with you, the lever is repetition and spacing, not the dial. Our guide to how Audible credits work explains why owning a title rather than streaming it makes the re-listen possible, and downloading your Audible books keeps them playable offline.

One caveat about their design applies directly to you. Participants could not pause, rewind or take notes. You can. Every rewind is a repair the study forbade, so real world fast listening probably performs better than the experiments suggest, and your rewind count is the most reliable measurement available to you. The self-test below turns it into a number.

Cognitive load, and why dense material breaks first

Speed alone is not what hurts you. Speed multiplied by difficulty is, and the pooled evidence shows that shape: cognitive load rose by g = 0.59 under acceleration while performance fell by less than half as much. Listeners spent more to get almost the same, and the spending is finite.

Think of it as a budget. Following speech in real time means holding a sentence in working memory while the next arrives, resolving who “she” refers to and tracking an argument across paragraphs. At 1.0x there is slack. Every increment spends some. On cheap material plenty remains at 2x. On expensive material you run out, and you run out without noticing, because the sensation of following along is generated by the parts you did catch.

Nothing in that table is a research finding. It is the load principle applied to what people own, a hypothesis to test on yourself. What is a finding: the studies used material near the middle of that range, and even there the authors flagged complexity and difficulty as boundary conditions they had not established.

Two load multipliers get ignored. First, what your body is doing: a speed that works on a walk collapses in traffic, because driving draws on the same attention pool. Second, the audio environment. A gym, a train, a kitchen with a fan running all force your ear to rebuild words from a degraded signal, expensive before you multiply anything. Our guide to what to do while listening to audiobooks ranks activities by attention cost, and reading along with the text is the standard fix for material that outruns your ear.

What speed does to the narrator’s performance

Here is the part no comprehension test can reach, and the reason the answer for fiction differs from the answer for a textbook. A professional narrator makes hundreds of decisions a page about pace, emphasis, pitch and above all silence, and those decisions carry information. The path into that craft is covered in how to become an audiobook narrator.

Prosody is not decoration on top of the words. Margaret Kjelgaard and Shari Speer showed in the Journal of Memory and Language in 1999 that prosodic phrasing changes how listeners resolve ambiguous sentences: matching prosody eases a structure that would otherwise trip you, and conflicting prosody misleads. The pauses and stresses do syntactic work. They tell you where the clause ends.

Now consider what uniform compression does to that. Halve a recording and everything halves: vowels, consonants, the breath before a revelation, the beat a narrator leaves after a line of dialogue so you know it landed. The Mach1 team built their system around this, and mimicking natural fast speech instead of compressing uniformly bought 17 percentage points of comprehension. No major audiobook app documents using non-uniform compression. Assume your pauses are squeezed by exactly the same factor as the words.

This is the asymmetry at the center of the question. For a book whose value is information, compressing the performance costs little, because the performance was packaging. For a book whose value is the experience, you are taking a degraded copy of what you paid for. A comprehension test cannot see that loss because it never measured it. You can, and the sign is specific: if you follow a story without ever being moved by it, the dial is a candidate explanation.

Narrator quality changes the calculation too. A brisk, clearly articulated professional survives 2x far better than a quiet, breathy performance full of natural pausing. Volunteer recordings vary enormously here, worth weighing in our roundup of where to download free audiobooks and the wider list of free audiobook sources. The word audible means able to be heard, and at 2.5x on a poor recording that is what is at stake. Budgets are part of why audiobooks cost what they do.

Why experienced fast listeners are not bluffing

When someone says they listen at 2.5x and follow everything, the reflex is to assume they are kidding themselves. Sometimes true. But adaptation to compressed speech is one of the better established effects here, and it is fast.

Emmanuel Dupoux and Kerry Green tested this in the Journal of Experimental Psychology: Human Perception and Performance in 1997, with 160 participants in the first experiment alone. Listeners heard sentences compressed to 38 or 45 percent of their original duration, roughly 2.2x to 2.6x. Accuracy climbed sharply over the first two or three sets of sentences, then leveled off. Adaptation took a couple of dozen sentences, not a couple of weeks.

Two follow up experiments matter for real listening. Switching talker after ten sentences produced only a small, immediately compensated dip, so the adjustment is not locked to one voice, and inserting normal speed sentences in the middle did not send performance back to baseline. Your ear keeps the skill. That is why the first ten minutes at a new setting sound terrible and the next hour does not, and why judging a speed by its opening minutes is the commonest mistake in the business, which our guide to finding your best listening speed treats as the core of its method.

The upper end is remarkable. In a 2007 study presented at the International Congress of Phonetic Sciences, Anja Moos, Ingo Hertrich, Susanne Dietrich, Juergen Trouvain and Hermann Ackermann worked with a blind man in his twenties who used a screen reader daily. He understood synthetic speech at 16 syllables a second. Six sighted controls topped out near 10, and imaging showed his visual cortex engaging. The sample is one person, so draw no population conclusions, but as an existence proof of how far trained listening goes it is hard to beat.

Adaptation has limits, and age is the clearest. Murphy, Hoover and Castel published a study in Memory in 2023 comparing younger adults averaging about 20 with older adults averaging about 72, samples of roughly 150 to 200 per group per experiment, at 0.75x, 1x and 2x. Younger adults held steady. Older adults declined at faster speeds. If you are recommending a setting to a parent, the undergraduate result is not the one to quote.

Mind-wandering, which cuts both ways

Every argument so far has treated 1.0x as the safe baseline. There is reason to doubt that, from the same UCLA lab. In the 2023 Memory study, probes asked participants what their mind was doing: focused, thinking about their own performance, distracted by the room, or zoning out. Faster playback tended to reduce mind-wandering, though the pattern wobbled across the three experiments, so treat it as a lean and not a law.

The mechanism is intuitive once you see it. A habitual reader takes nonfiction off a page at around 238 words a minute and fiction nearer 260, according to Marc Brysbaert’s 2019 review of 190 studies. Narration at 155 wpm is far slower, and the gap is where attention goes shopping. If you have ever finished a chapter at normal speed with no memory of the last four minutes, you have met the failure that speeding up fixes.

So the trade is not comprehension against time. It is one failure against another.

Both are real, and which you are prone to depends on your reading speed, the material and what your hands are doing. For a fast reader on mainstream nonfiction, 1.0x may well produce worse retention than 2x, because at 1.0x they are barely in the room.

The verdict: for whom, and for what

Two is not a magic number and there is no cliff at 2.0x, but there is a slope, and how steep it feels depends on who you are and what is playing.

The clearest line the evidence supports: up to 1.5x the cost is low enough to ignore on any material. Between 1.5x and 2x the direct experiments find nothing significant and the pooled work finds a small penalty, so it depends on the book. At 2.5x you have the one replicated deficit, and the researchers who found it advise against it.

One group deserves a warning. If you are inside a free trial deciding whether audiobooks work for you at all, do not run that experiment at 2x. You will blame the format for what is really the setting. Start at 1.0x or 1.25x with fiction you want, and read how to use Audible in your first week before touching the dial. Same if you are testing through a library app or whatever audiobooks Kindle Unlimited includes.

Find your own limit in one week

Every number here is an average over people who are not you, and averages are the wrong tool for a personal setting. This is a protocol, not a study, and it tests what the research could not: your material, your ear, your attention, measured after a delay.

Once you have a number, the ramping method for holding and building on it belongs to our listening speed guide, which sets out a four week protocol, the stopping rules and settings by material and situation. This page tells you what the evidence says. That one tells you what to do about it.

What the speed change buys you

The comprehension side of this trade is measured in single percentage points on a multiple choice test. The time side is measured in hours of your life. Ignoring one half is how people end up at 1.0x out of guilt or 2.5x out of ambition.

Take a 13 hour audiobook, near the top of the standard range.

Look at the fourth column, the argument this page has been building toward. Going from 1.0x to 1.25x buys two hours thirty six minutes at no measurable cost. The next step buys an hour forty four. The step from 1.75x to 2.0x buys fifty six minutes, and it is where the evidence turns mixed. Savings shrink as risk grows.

Run your own titles through the audiobook speed calculator instead of trusting a 13 hour example, because the arithmetic is nonlinear and your intuition about it is wrong. It sits with our other free lifestyle calculators, and how many audiobooks a year is realistic turns the hours into finished books.

One case to watch. On Spotify audiobooks, speed does not stretch your monthly allowance, because Spotify counts listening at normal speed. Finishing a 15 hour book at 2x still spends 15 hours of quota, so there the calculator describes your calendar and not your entitlement. The details are in how many audiobook hours Spotify gives you. Audible has no such limit, and how to get credits on Audible covers the supply side. Apple sells outright too, though not through Apple Music.

Five ways people misread this evidence

One more, aimed at a claim you will meet repeatedly: there is no evidence that listening fast damages your attention span, rewires your brain or degrades your ability to read slowly. That is folklore. What the research supports is duller: a small, speed dependent cost to test performance, a larger cost in effort, and less enjoyment.

Fast listening: frequently asked questions

The short version

Related reading