How to listen to research papers
Listening is not a way to skim. Here is the part of reading a paper it genuinely replaces, and how different kinds of paper hold up.
In short: Listening does not replace skimming a paper. It replaces the proper read, once you have already decided the paper is worth an hour.
You have thirty papers saved and time for about four of them. Everyone in this position eventually has the same thought: what if I could just listen to them on the walk in?
It is a good idea, and it works, but not for the reason most people expect and not at the stage of reading where most people try it.
Nobody reads a paper front to back
The advice every supervisor eventually gives is to read a paper in passes rather than straight through.
The first pass is a five-minute skim. Title, abstract, headings, the figures, the conclusion. You are not learning the paper, you are deciding whether it deserves an hour. Most papers do not.
The second pass is the real read. You go through the argument properly, follow the method, and come out able to describe what the authors did and whether you believe them.
The third pass, which most papers never earn, is working through it in detail — checking the maths, deciding whether you could reproduce it.
Listening replaces the second pass, not the first
This is the thing worth being clear about, because it is the opposite of how read-aloud apps are usually sold.
Audio is linear. You cannot jump your eyes to the conclusion, glance at a figure, and back out in ninety seconds. Skimming is a visual skill and no voice will give it to you. If you try to triage a pile of papers by listening, you will spend eight minutes on each one and still not know which are worth keeping.
Where listening genuinely wins is the second pass, and the re-read. Once you have decided a paper matters, listening lets you take it somewhere you could not take a screen — a walk, a commute, the washing up — and gives you the whole argument at a pace you cannot rush. Many people find they absorb a familiar paper better on the second time through by ear than by eye, because they stop skipping the paragraphs they think they already understand.
So: skim on screen, listen for the read.
Why this usually sounds terrible
If you have tried this before and given up, it was probably not your fault.
A PDF is closer to a photograph of a page than to a document. It records where to draw each letter, and nothing else. There is no marker saying "this is a paragraph" or "this is the page number". Those distinctions live in your eyes, not in the file.
So a tool that simply reads out the text finds no reason to treat a footer differently from a sentence, and you get this:
...the resulting distribution is therefore stable. 4. Preprint. Under review. Figure 2: Overview of the pipeline.
Four separate things collided there. To you it is noise. To the software it was all just text, in order. Multiply that by every page and the paper becomes unlistenable — the title repeated forty times, a bare number interrupting every sentence that runs across a page break, and axis labels read out as though they were prose.
What ONarrator does instead
Before it reads anything, ONarrator looks at each page the way you do — as a shape, not as a stream of letters. It works out which parts of the page are the writing and which parts are furniture.
The furniture stays out of the narration: the running header, the footer, the page number, the figure captions, the tables, the displayed equations, and the reference list at the back. What is left is the paper, and that is what you hear.
Each of those is a switch, so if you actually want the captions read you can have them, and you can decide differently for one paper than for the rest of your library.
One detail worth knowing: the very first time you open a paper it may read everything, because it has not finished looking at the pages yet. Give it a moment and reopen it. From then on it remembers.
The other thing worth knowing is what happens when it is unsure. It reads. A page it cannot confidently make sense of is narrated in full, rather than skipped. Hearing a stray page number is a small annoyance; silently losing two paragraphs of someone's method is not.
How different papers hold up
This varies more than anyone advertising a read-aloud app will tell you.
- A single-column preprint is the good case. arXiv-style papers read cleanly from the abstract to the conclusion.
- A two-column conference paper is mostly fine, but the order can occasionally jump between columns. It is the one failure you will still hear.
- A maths-heavy paper skips the big displayed equations rather than reading them as gibberish. Symbols inside a sentence still sound like symbols, and no amount of clever handling fixes a sentence that was written to be looked at.
- A survey or review is where the difference is most obvious. Fifteen pages of citations simply do not get read to you.
- A thesis or a government report stops repeating its own title at the top of every page, which on a two-hundred-page document is the whole ballgame.
A routine that works
Add the paper. Give it a few seconds to look at the pages. Then generate the audio for the whole thing before you leave the house, so it plays without a connection and without pausing to think — this matters on the Underground and it matters on a plane.
If you would rather have the paper in your podcast app, save it out as an audiobook file and it will turn up there with proper chapters.
Then listen at a speed that suits you. Most people land somewhere between 1.1× and 1.4× for non-fiction. Move in small steps; the range where it stays comfortable is narrower than you would guess.
If the paper is dense, read along with it. The word being spoken is highlighted as you go, so you can follow on screen and still get the pacing of somebody reading it to you. Tap any sentence and it starts from there.
What it will not do
It reads you the paper. It does not summarise it, explain the figures, or tell you what the contribution was. Several tools in this space will happily do that, and if that is what you want, they are the right choice.
This one gives you the actual words the authors wrote, in order, without the page numbers. Working out what they mean is still your job — which, for a paper you are going to cite, is rather the point.
If you are weighing this against the alternatives, the comparison page lays out who does what. And if it is books rather than papers you are trying to get through, that works much the same way.