Best Ambient Sounds for Studying
What noise is best for studying? Start with the result nobody sells
That question gets asked in a dozen phrasings — what noise is good for studying, which noise is best for studying, best noise to study to — and the research answer disappoints all of them equally. A Bayesian meta-analysis pooled 65 experiments on reading with sound in the background and found that noise, speech and music each left performance slightly worse than quiet did. Speech you could make out, and music with words in it, did the most damage. Nothing in that analysis came out ahead of silence.
So the premise of the question needs replacing before the question can be answered. There is no sound that makes a person better at studying than quiet does. What exists, for particular tasks, is a sound that costs less than the room it covers up — and that is worth having when the room contains a flatmate on a call, a sibling watching videos or four strangers in a library annex. The practical question is not which sound is best. It is which sound is cheapest against the task in front of you, in the room you are actually in.
The study task decides, not the colour
Studying is not one activity. Drilling vocabulary, reading a dense chapter, working through problem sets and drafting an essay load memory and attention in different ways, and the experiments that tested them were separate experiments with separate results. The table below keeps each finding attached to the task it was measured on.
Read the rows as directions rather than as magnitudes. Each comes from different participants doing a different task, so a downward arrow in one row is not a bigger or smaller effect than a downward arrow in another.
Notice what the table does not contain: a column for noise colour. White, pink and brown noise are the same random signal with its energy weighted differently across the frequency range, and that difference is covered in white versus pink versus brown noise. No study below compared them against each other on a study task. What the evidence does distinguish is the task, so the sections that follow are organized that way.
Memorising lists, dates and vocabulary
Holding a sequence in order is where background sound does its most reliable damage. Salamé and Baddeley found that speech a listener was neither attending to nor able to understand still disrupted recall of lists presented on paper. The usual account is the phonological loop, the part of working memory that rehearses verbal material: incoming speech sounds get into it uninvited and compete with the list being held there.
A later experiment closed the obvious escape route. Perham and Vizard tested serial recall under music participants had chosen as liked and music they had rated as disliked, and recall was worse than quiet under both, with no advantage for the liked version. Enjoying a track changes how the session feels; it does not exempt the task from interference.
For flashcards and formula drills that gives a narrow, firm rule: nothing with words, and nothing picked because it is enjoyable. Where a silent room cannot be arranged, a steady broadband sound that covers the talking is the fallback, and trading an intermittent large cost for a constant small one is the real case for white or pink noise during memorisation.
Hear the difference: one colour plays at a time.
These buttons need JavaScript, which is off or still loading. Every colour is generated in your browser, so there is no audio file to fall back to.
Start quiet and switch between two colours rather than stacking them. Green and grey have no standard definition; these are the shapes this site generates, and another site's will differ.
Reading and comprehension
Reading is the task the meta-analysis covers head-on, and its verdict applies to everything: each category of sound tested carried a cost. The ranking that follows is quiet first, then featureless sound, then anything carrying words — which is also the answer to what the best type of noise for studying is when the reading is dense. A broadband hiss offers the reading system nothing to decode, and that absence is the whole of its value.
The size of the effect matters as much as its direction. These were small average differences, not the difference between finishing a chapter and failing to. A reader who has found that rainfall suits them is not contradicting the evidence; they are sitting inside its error bars. What the meta-analysis rules out is the strong claim — that a study sound is a comprehension aid — not the weak one, that a particular sound makes a particular room bearable.
Problem sets and mental arithmetic
Here the comfortable version of the advice breaks. Banbury and Berry played office noise to participants doing two tasks, memory for prose and mental arithmetic. With speech in the recording, both tasks suffered. With the speech taken out, prose memory recovered — and arithmetic was still disrupted. A further experiment in the same paper found the arithmetic disruption did not depend on whether the speech carried any meaning.
So wordlessness is not a clean bill of health for calculation. What a calculation appears to object to is not language but variation: a recording of a room keeps changing even after the voices are stripped from it. That is the strongest argument for a synthesized sound over a recorded one during problem sets, because a generated noise can hold still in a way that a recording of anywhere cannot.
Writing and drafting
Writing has its own measurement, and it is the least forgiving of the four. Keus van de Poll and colleagues had participants write five-minute essays under five levels of speech intelligibility, measured as speech transmission index. Writing fluency dropped sharply, and pauses longer than five seconds became more frequent, once the index went above 0.23 — a value low enough that the speech is hard to follow. Their own framing is that realistic work drops off even at low intelligibility.
The practical consequence for anyone drafting an essay next to a conversation is that half-covering the voices is not enough. Masking has to push intelligibility a long way down, not merely below the point where the words are effortless, and if it cannot get there, the fix is the room rather than the volume.
The disruption fades, and a break brings it back
None of the results above is permanent, which changes how a long session should be planned. In an earlier set of experiments Banbury and Berry tested whether background noise can be adapted to at all, using memory for prose. Background speech was habituated to after twenty minutes of exposure, and neither the meaning of the speech nor its repetition changed how much habituation appeared. Office noise with no speech in it was habituated to as well.
The third experiment is the one with teeth. A five-minute period of quiet was enough to partly restore the disruption that had already been habituated to, while merely changing the voice was not. Twenty minutes of adjustment, undone by a five-minute break — that is an awkward shape for anyone studying in blocks with breaks between them, and it means the first stretch after every break is the expensive one.
Two things follow. Longer blocks get more of the adapted state than short ones, so a sound that has to be tolerated rather than enjoyed argues for fewer, longer sittings. And if a break is taken in the same sound rather than in quiet, the restart may cost less, though the experiment tested a quiet break rather than a continuous one and cannot settle that.
Why background noise for studying helps some people and not others
Söderlund and colleagues played the same white noise to children with ADHD and to a control group. Recall improved in the first group and got worse in the second. One study, one age group, one sound — but it is the cleanest demonstration available that this question has no single answer at the population level, and the reason a general recommendation can only ever be a starting point. Two people can follow identical advice, get opposite results, and both be reporting their own experience accurately.
The authors explain the split through stochastic resonance, the idea that a moderate amount of noise can push a signal over a threshold it would otherwise miss, with the amount required depending on dopamine levels. That is their interpretation rather than a measured quantity, and it predicts a different optimum for different people rather than a single best setting for everyone. Whether it holds or not, the measured result stands: direction of effect reversed between two groups listening to the same sound.
What actually predicts the cost is change, not loudness
Three of the findings above point the same way. Unintelligible speech disrupted recall. Office noise with the speech removed still disrupted arithmetic. Music disrupted recall whether or not the listener liked it. The variable those share is not loudness, and it is not meaning: it is that the sound keeps changing. A signal that varies over time gives attention something to track; a flat, steady one gives it nothing.
That is what masking buys and what it costs. Covering a conversation with steady noise does not take sound out of the room — it replaces a changing, word-bearing sound with an unchanging, wordless one. Against a quiet room that is a straight loss. Against a room with a talker in it, it is usually a win, and the busier the room the better the trade looks.
How loud should study sounds be
No study has established an optimal volume for studying, so a precise decibel recommendation for study sounds is quoting a number nobody measured. Two things are known. In the creativity experiments, the conditions were room levels of roughly 50, 70 and 85 decibels: the benefit appeared at the middle one and was gone at the loudest, where the authors report less information being processed. Those are levels in a room, not a figure to dial into headphones.
Second, for hearing safety the National Institute on Deafness and Other Communication Disorders puts sounds at or below 70 A-weighted decibels, even after long exposure, as unlikely to cause hearing loss, with long or repeated exposure at or above 85 dBA capable of causing it. Normal conversation sits at 60 to 70 dBA.
Together they give a rule rather than a target: set the sound at the lowest level that stops the room from intruding, and keep it clearly below conversation level. If the sound itself is noticeable during verbal work, it is too loud. Changing the volume mid-session is its own interruption, so set it before starting.
Finding the best noise for focus by testing it on yourself
Because the group-level effects are small and documented in both directions, a short personal trial is worth more than any recommendation. It is a single-subject experiment and should be read as one: it reports what happens to you, on that task, at that desk.
Pick one task you repeat often and can count — pages read, problems finished, cards recalled correctly. Hold the task, the time of day and the session length constant and change only the sound. Spend the first sessions on three broad options: silence, a steady broadband sound, and something textured such as rainfall. Then run the best two against each other again, because a first pass catches day-to-day variation along with the sound. The study noise tool keeps the conditions comparable: the same five colours, a session timer, and no track changes in the middle.
Write down the count and a one-to-ten focus rating at the end of each session rather than at the end of the week, by which time memory of the early sessions has rearranged itself around whatever conclusion has formed. And expect small differences. If one condition looks dramatically better, the room, the task or the day is the more likely explanation than the audio.
Two details in the design matter more than they look. Test the task you actually have trouble with: the findings above diverge by task, so a trial run on easy reading says little about problem sets. And give each condition at least a full block, because twenty minutes is roughly where adaptation showed up in the habituation experiments and a five-minute sample is measuring the adjustment rather than the state.
Three things to avoid
Words, spoken or sung, during verbal work. This is the one recommendation the evidence supports without hedging. Intelligible speech and lyrics were the most disruptive conditions in the reading meta-analysis, and unattended speech disrupted recall even when it could not be understood.
Changing the sound mid-session. Every switch is a change to notice, and change is the thing that costs. Choose before starting and leave it alone.
Using sound to paper over a bad room. Masking works on speech at moderate levels. Against traffic or construction it means raising the level until the sound is the problem, and the writing study suggests partly covered speech is close to the worst case rather than a safe middle. Moving the desk, closing a door or using closed-back headphones treats the cause.
References
- Vasilev, M. R., Kirkby, J. A. & Angele, B. (2018). Auditory Distraction During Reading: A Bayesian Meta-Analysis of a Continuing Controversy. Perspectives on Psychological Science 13(5), 567-597 (65 experiments; noise, speech and music each left reading slightly worse than quiet; speech and lyrics worst)
- Salamé, P. & Baddeley, A. (1982). Disruption of short-term memory by unattended speech: Implications for the structure of working memory. Journal of Verbal Learning and Verbal Behavior 21(2), 150-164 (unattended and unintelligible speech both disrupted recall of visually presented lists)
- Perham, N. & Vizard, J. (2011). Can preference for background music mediate the irrelevant sound effect? Applied Cognitive Psychology 25(4), 625-631 (serial recall was poorer under both liked and disliked music than under quiet; preference made no difference)
- Banbury, S. & Berry, D. C. (1998). Disruption of office-related tasks by speech and office noise. British Journal of Psychology 89(3), 499-517 (office noise with speech disrupted memory for prose and mental arithmetic; with the speech removed it still disrupted arithmetic, and the disruption did not depend on the meaning of the speech)
- Banbury, S. & Berry, D. C. (1997). Habituation and dishabituation to speech and office noise. Journal of Experimental Psychology: Applied 3(3), 181-195 (background speech was habituated to after 20 minutes regardless of its meaning or repetition; office noise without speech was too; a 5-minute period of quiet partly restored the disruption, while a change of voice did not)
- Keus van de Poll, M., Ljung, R., Odelius, J. & Sörqvist, P. (2014). Disruption of writing by background speech: The role of speech transmission index. Applied Acoustics 81, 15-18 (five-minute essays across five speech transmission index conditions; writing fluency dropped and pauses longer than five seconds increased above an index of 0.23)
- Mehta, R., Zhu, R. & Cheema, A. (2012). Is Noise Always Bad? Exploring the Effects of Ambient Noise on Creative Cognition. Journal of Consumer Research 39(4), 784-799 (five experiments; a moderate 70 dB room level beat a quiet 50 dB one on creative tasks, and a loud 85 dB level hurt)
- Söderlund, G., Sikström, S. & Smart, A. (2007). Listen to the noise: noise is beneficial for cognitive performance in ADHD. Journal of Child Psychology and Psychiatry 48(8), 840-847 (white noise improved recall in the group with ADHD and worsened it in the control group)
- Noise-Induced Hearing Loss — National Institute on Deafness and Other Communication Disorders (at or below 70 dBA unlikely to harm even after long exposure; 85 dBA and above can cause hearing loss with long or repeated exposure; normal conversation 60-70 dBA)
Frequently Asked Questions
What noise is best for studying, if you have to pick one?
For most study tasks the defensible pick is a steady broadband sound — white, pink or brown noise — and the reason is not that it helps. It is that it holds still. No sound tested in the reading literature beat quiet, so the only job left for a study sound is to cost less than the room you are in. A sound that does not change gives attention nothing to follow; a playlist, a café recording or a conversation does.
Does background noise for studying actually improve performance?
Usually not. Across 65 experiments on reading, noise, speech and music each left performance slightly worse than quiet. The documented exception is creative idea generation, which improved at a moderate room level in a five-experiment study and got worse when the level was raised further. Treat a study sound as cover for a worse distraction rather than as a performance aid.
Is wordless noise safe for problem sets and mental arithmetic?
Less safe than it sounds. One experiment stripped the speech out of an office noise recording and found it still disrupted mental arithmetic, even though prose memory survived that version. What a calculation seems to need is not wordlessness but steadiness, which makes a synthesized noise a better bet for problem sets than a recording of a room.
Why is music with lyrics worse than instrumental music?
Sung words carry speech, and speech enters the verbal part of working memory whether or not you attend to it. In the reading meta-analysis, intelligible speech and lyrics were the most disruptive conditions tested. Instrumental music is the lesser risk rather than a safe one: a serial recall study found liked and disliked music both impaired performance compared with quiet, so enjoying the track does not protect the task.
Why do study sounds help some people and not others?
The difference is documented rather than imagined. One study played the same white noise to children with ADHD and to a control group, and recall improved in the first group while it got worse in the second. Because group-level effects in this literature are small and run in both directions, a short trial on a task you can count is worth more than any general recommendation, including this one.