A lamp-lit corner desk at night with an open notebook, a lit hallway blurred behind it

If the problem is a housemate on the phone, you are solving a masking problem, and the research supports doing something about it. If the problem is that you cannot make yourself start, sound is not the lever. Most pages on this question answer the second one with advice for the first.

The distinction matters because it changes what you should put on. Masking asks for something steady and uninteresting at a volume that competes with the intrusion. Focus audio, whatever that means to you, is a much softer claim with much weaker support. Knowing which problem you have takes about ten seconds and saves you from shopping for the wrong thing.

Here is what the evidence separates cleanly.

Why is a person talking worse than traffic?

Because what disrupts you is variation, not volume.

The effect has a name in the literature, the irrelevant speech effect, and it has been studied since the 1970s. Jones and Morris reviewed it in 1992 and landed on the account that has held up since: disruption is driven by the changing state of the sound, not by whether it is speech as such or by whether it sounds like what you are reading. A stream that keeps shifting captures processing. A stream that does not, does not.

Hughes, Tremblay and Jones sharpened this in 2005 by taking speech apart. Across two experiments using consonant-vowel-consonant syllables, they found that changes in the vowels were the dominant source of disruption, more so than changes in the consonants. That is a technical result with a very ordinary consequence: a conversation, which is mostly moving vowels, is close to the worst possible background sound, and a refrigerator hum, which barely moves at all, is close to the least bad.

It is also task-specific. In 2024, Marsh and colleagues ran two high-powered replications, one with 64 participants and one with 128, and found that changing-state speech disrupted visual-verbal serial recall but not visual-spatial serial recall. Reading and writing are the vulnerable activities. Work that is spatial rather than verbal appears to survive the same background better. If you have ever wondered why you can sketch a diagram in a cafe but cannot read a paragraph there, that is the shape of it.

Does playing something over it actually help?

Yes, and for a narrower reason than people assume.

Park, Kohlrausch and van Leest tested this directly in 2013, comparing serial recall in silence, with speech, with noise alone, and with speech masked by noise at different signal-to-noise ratios. Masking helped: an adaptive masker reduced the error rate by about 9 percent compared with unmasked speech. But the detail that matters is the one that is usually left out. Performance did not differ significantly between the adaptive masker and a plain stationary one once the average sound level was matched.

So the mechanism is not that a continuous sound beats an intermittent one in some general way. It is that making the speech less intelligible makes it less disruptive. What you are doing when you put something on is not soothing yourself, it is raising the floor until the words next door stop resolving into words.

That reframes the practical question. You do not need the right sound. You need enough of a sound, and you need it to be dull. Anything with its own changing vowels, which is to say anything with a vocal, is contributing to the problem it is meant to solve. The same reasoning runs through the post on whether music helps you study, where lyrics turn out to be the variable that decides most of the disagreement in that literature.

What should you actually put on?

Something steady, instrumental, and loud enough to do the job.

Steady is the property the research points at, so the ranking goes roughly: generated noise and rain at the dull end, instrumental music with a repetitive structure in the middle, anything with a voice at the end you do not want. This is the place to say plainly what this app is. Pocket Lofi plays original lofi and chillhop music in five stations, Focus, Rain, Sleep, Chill and Groove. It does not generate white, pink or brown noise. If a noise generator is what your apartment calls for, a generator is what you should get, and the Rain station is the nearest thing here for anyone who came looking for rain.

One more distinction is worth drawing before you set anything, because the two common cases pull in opposite directions. If the intrusion is intermittent, a door, a dog, a laugh from the next room, the job is to raise the floor high enough that those events stop standing out against silence, which usually means a little more volume than feels necessary. If the intrusion is continuous, steady traffic or a fan, you are already living with a masker and adding a second layer mostly adds loudness. In that case turn it down or leave it off. The research above does not test these two cases separately, so that split is reasoning from the masking mechanism rather than a finding.

Loud enough is the part people get wrong in the opposite direction. If you set masking audio at a polite volume it will not mask anything, and you will have added a second sound to the room without removing the first. Set it where the intrusion stops being intelligible, check it during an actual intrusion rather than during a quiet moment, then leave it alone.

For choosing between instrumental options, the study playlist guide goes through the properties that matter. The short version is that repetition, which is an aesthetic weakness in music you are listening to, is the functional property here.

What if the noise is not the real problem?

Then no volume of anything will fix it, and it is worth being honest about that before you spend an evening adjusting sliders.

The University of North Carolina’s Learning Center, writing for students rather than for researchers, treats sound as one item in a longer list: the physical setup, the phone, the tab situation, the time of day, and the habit of having a location the brain associates only with work. Their framing is useful because it refuses to make the sound environment the whole story. Sound is the easiest variable to change, which is exactly why it absorbs attention that belongs elsewhere.

The most common version of this is not being distracted but being unable to start, and that is a different problem with a different fix. Starting is a structure problem: a short, defined block with a stated end is easier to begin than an open evening. The deep work guide covers how those blocks are supposed to be built, and the 25 minute timer is the crudest possible version of the same idea, which is most of why it works.

How to set up one session

Decide which problem you have, then do the matching thing.

If the house is loud, put on something steady before you sit down rather than after you have already lost ten minutes to it, set the volume against the actual intrusion, and pick something without vocals. If the house is quiet and you still cannot work, skip the audio decision entirely and start a timer instead. If you are somewhere in between, which most people are, do both in that order: sound first because it takes five seconds, then the timer, then start.

That sequence is what the Focus station in Pocket Lofi is built for. It is instrumental by construction, it loops without an ending to notice, and the built-in timer means the session has a stated finish before you begin rather than a vague intention to work until you stop. Starting the station before the work, not during it, is the part most people skip.

None of this is a productivity system, and the effects in the studies above are measured in error rates on recall tasks rather than in grades. What the evidence supports is modest and specific: intelligible speech is the worst thing in your environment, and covering it up measurably helps. Everything past that is preference, and preference is a fine reason to pick the thing you actually like.

Sources