AI 論文ノート: 読み取った内容を検索可能な保管庫に保存する

A thesis has three phases, and the middle one is the longest and the least discussed. The beginning is exciting because everything is new. The end is terrifying and therefore focused. The middle is two years of reading, fieldwork, false starts, and supervision meetings, and it runs on a memory that was never designed to hold two hundred arguments in order.

The standard advice is to take better notes. That advice is correct and almost useless, because the reason people stop taking literature notes is not laziness. It is friction. You finish a dense paper at the end of a long day, you have a real reaction to it, and writing that reaction as three tidy paragraphs feels like starting a second piece of work. So you highlight instead, promise yourself you will write it up, and move on.

Highlighting is not note taking. A highlight records that a sentence seemed important to a version of you who no longer exists. It does not record why. Six months later you open the PDF, find eleven yellow blocks, and have to reconstruct your own reasoning from scratch. The valuable thing was never the author's sentence. It was your reaction to it.

So what makes the reaction cheap enough to actually capture? Speaking it. You can talk through what a paper argues and where it fails in ninety seconds, at a moment when typing three paragraphs feels impossible. The catch has always been that spoken notes come out as unreadable sludge, full of restarts and filler, so nobody wants to reread them. On-device cleanup removes that catch, and once the catch is gone the whole loop changes.

Yaps is a voice assistant that runs on your own device. For postgraduate research it does five concrete things, and the important part is that all of them work without sending your unpublished thinking anywhere.

It turns speech into clean literature notes. Press the Yaps hotkey (on desktop the key is Fn, on Android it is the dictation button on the Yaps keyboard) and talk the way you actually think about a paper, restarts and asides included. Recognition runs on your device across about 25 languages, auto-detected from your speech, so if you read and think in Spanish, German, or Portuguese, your notes do not have to be in English. Because it works across every app on your system, you can dictate straight into your PDF annotation pane, your draft chapter, or a fresh note. The dictation feature page covers the mechanics.

It cleans up the ramble on-device. Cleanup is on by default and runs locally. It strips filler words and self-corrections, fixes punctuation and capitalisation, and turns a spoken list into an actual list. A muttered "so this one is, um, basically arguing that the sample was too, sorry, too small to generalise from, and I think that undercuts the thing in chapter three about, right, about regional variation" comes back as two clean sentences you will be willing to reread. That difference is the whole reason spoken notes become usable.

It keeps everything in a private searchable vault. Notes live in a vault on your own machine, and you can search them and ask across them locally. They are not just walls of text: plain notes, kanban boards, and checklists are all supported, which suits a chapter plan or a list of gaps to chase. Notes export to Markdown and plain text, so nothing is locked in. The voice notes feature goes deeper, and our guide to building an AI memory app you actually own covers the recall side.

It syncs between your phone and your laptop. Vault syncing is a premium feature that pairs your own devices over your local network or an encrypted peer-to-peer link using a QR code, rather than routing through a central server. The reaction you speak on the walk home from the library is on your laptop when you sit down. This matters more than it sounds, because the best thinking about a paper very rarely happens at your desk.

It transcribes audio files offline in Studio. If you have a recording you made yourself, say a supervision meeting you recorded with permission or a fieldwork interview, the Studio editor imports that audio file and transcribes it offline into text or SRT subtitles. That is a file you already own, turned into private text on your own machine. Our guide to offline transcription for qualitative research covers that workflow in detail.

The loop is small, and that is the point. A workflow you follow for three years has to survive tired evenings, so every step in it must be cheap.

Two habits make this loop far more powerful, and both cost nothing. First, always say the author and year out loud inside the note, because that is the string your future self will search for. Second, always say which chapter or research question the paper touches, because that turns a flat pile of notes into something you can assemble.

Here is the whole thing in practice.

Step one: install and open a note. Download Yaps on your Android, Windows, macOS, or Linux device and start the 7-day free trial. Open a note in the vault, or put your cursor wherever you already keep literature notes, because Yaps types into any app on the system. No account is needed to capture a note, only to manage billing.

Step two: say the author, year, and title first. This feels clerical and it is the highest-value habit in the workflow. Search only finds words that exist in your notes, so a note that begins "Okafor, 2023, the paper on regional sampling" is retrievable and a note that begins "this one was really interesting" is not.

Step three: answer three questions out loud. What does it argue. Why does it matter to my thesis. Where do I disagree. Do not perform. Say "the sample size makes the generalisation shaky and I want to push back on that in chapter three" exactly as it occurs to you.

Step four: let cleanup do the tidying. The on-device cleanup step turns the spoken mess into finished prose: filler gone, punctuation in, lists formatted. You skim and correct rather than retype, and none of it needed an internet connection.

Step five: tag the chapter and save. Name the chapter or research question while you are still talking, then save. Over three years this quietly builds a pile per chapter, which is exactly what you want on the day you sit down to draft. See academic writing with your voice for how that pile turns into prose.

That is the question that decides whether a note-taking tool suits a thesis. Not how clever the AI is, but where three years of unpublished thinking rests and who else gets a copy. An unpublished argument, an unreviewed dataset, and a frank opinion about a paper by someone who may review your work later are all things worth keeping on your own machine.

Read the rows where Yaps says no as scope, not as apology. A tool that never uploads your audio cannot leak your unpublished argument, and a tool that does not reach into a citation database was never trying to be your bibliography. The next section says plainly which tools should own those jobs.

This is the section that matters most for research, because a tool that overclaims in an academic workflow can cost you more than time.

Yaps is not a reference manager. It does not store references, does not format citations, does not generate a bibliography, and does not talk to Zotero, Mendeley, or EndNote. Those tools remain the right answer for your bibliography and you should keep using one. Yaps produces clean, structured, paste-ready text that you copy wherever you need it, and that is the honest description of the handoff. There is no sync, no plugin, and no integration.

Yaps does not read or summarise papers you have not read. It captures your thinking about a paper you read yourself. It will not condense a PDF you have not opened, and it will not tell you what a study found. That is a real limitation and it is also deliberate: a summary of a paper you never read is not a literature note, it is a guess you will end up defending in a viva.

Yaps does not search academic databases. It does not query the literature, does not find related work, and does not build your search strategy. Your library discovery service, the subject databases, and your supervisor remain the tools for finding things. Yaps starts working after you have found and read the thing.

Yaps does not write your thesis. It transcribes your words and tidies the delivery. It does not generate arguments, invent sources, or produce paragraphs you did not think of.

Two more honest limits. Yaps does not transcribe live supervision meetings and does not put a bot in a call, so if you want a record of a supervision, record it yourself with everyone's clear permission and then transcribe that file offline in Studio afterwards. And there is no iOS app yet. Yaps runs on Android, Windows, macOS, and Linux today, with iPhone and iPad coming soon, so if your entire reading setup is an iPad, that is the gap to weigh now.

Everything above is about the reading years. The drafting year has a different problem, and voice helps there in a different way.

The blank page in a thesis is rarely a shortage of ideas. It is the gap between knowing roughly what you think and producing a sentence that is good enough to keep. That gap makes people stall for hours over an opening line. Speaking jumps it. Say the paragraph badly, out loud, the way you would explain it to a colleague at lunch, and you will have four hundred rough words in five minutes.

Rough words are not the finished chapter, and nobody should pretend otherwise. They are raw material that you then cut, restructure, and rewrite, which is a completely different and far easier task than staring at nothing. Most people find they can edit for an hour when they could not have started for one. The dictation feature and the wider researcher workflow cover the habit in more depth.

A rough spoken paragraph is not your chapter. It is the thing that makes your chapter possible to edit into existence.

Yaps

Read-aloud closes the same loop from the other end. Yaps can read a selected passage of your draft back to you: on desktop you trigger it with Option and Fn, and on Android you choose "Read aloud with Yaps" from a text selection. Clumsy sentences are extremely hard to see and extremely easy to hear. A clause that loops back on itself, a paragraph with three "however"s, an over-nested sentence that runs out of breath: all of them announce themselves within seconds of being spoken. Academics have proofread by reading aloud for a century, and having the machine do the reading means you notice the prose instead of your own performance of it.

Two caveats on read-aloud. There are 18 voices on desktop (8 that work offline and 10 cloud voices) and 2 on mobile, and the shipped voices are English speakers in practice. If you are drafting a thesis in French, Mandarin, or Arabic, a multilingual cloud reader will serve you better for that specific job, and it is worth saying so rather than pretending otherwise. Dictation itself is multilingual across about 25 auto-detected languages, so capture is not the constraint. Read-aloud is.

Start with the loop, not the tool. For the next ten papers you read, spend ninety seconds speaking your reaction the moment you finish, with the author and year said out loud and the chapter named. That single habit is what makes month twenty survivable, and it works whether you keep the notes in Yaps or anywhere else.

Then make the capture cheap enough to survive a bad week. Yaps is the default starting point for that because it removes the two things that usually kill the habit: the friction of typing when you are tired, and the unease of putting unpublished work on someone else's server. Dictation, cleanup, storage, and search all run on your own device, so the notes work in a basement archive with no signal and stay yours by default. Pair it with a reference manager for your bibliography, because Yaps genuinely does not do that job. If you also want lecture and revision tooling, our roundup of the best AI study apps covers that undergraduate-facing side, and the Yaps for students page is the fastest overview.

The thesis has to be your thinking, start to finish. The only thing worth automating is the part where your thinking gets lost.

What is the best AI app for thesis and dissertation notes?

For capturing and recalling your own thinking, Yaps is a strong pick because dictation, cleanup, storage, and search all run on your own device, so unpublished research notes never travel to a server. You speak your reaction to a paper, on-device cleanup makes it readable, and it lands in a searchable local vault that still answers questions three years later. It runs on Android, Windows, macOS, and Linux. Pair it with a reference manager such as Zotero, Mendeley, or EndNote, because Yaps does not handle citations.

How do I keep track of all the papers I read for a literature review?

Capture a short reaction to each paper immediately after reading it, and store all of those reactions in one searchable place. Record what the paper argues, why it matters to your project, and where you disagree, plus the author surname and year so you can find it later. Speaking that takes about ninety seconds where typing it takes ten minutes, which is why spoken notes survive a three-year project and typed ones often do not. Highlights alone are not enough, because a highlight records that a sentence mattered without recording why.

Can AI write my literature review for me?

No, and you should not try to make it. A literature review is an argument about a field that you are expected to have made yourself, and passing off generated text as your own work is academic misconduct at essentially every institution. Yaps is deliberately built for the other job: recording and organising your own thinking, and helping you draft in your own words. Check your university and department policy on AI use before adopting any tool, and declare use where your institution requires it.

Is Yaps a reference manager like Zotero or Mendeley?

No. Yaps does not store references, format citations, or generate a bibliography, and it has no integration with Zotero, Mendeley, or EndNote. Those tools remain the right answer for your reference library and you should keep using one alongside Yaps. What Yaps produces is clean, paste-ready note text that you copy where you need it, exported to Markdown or plain text when you want it in a document.

Can Yaps summarise a research paper for me?

No. Yaps does not read PDFs, does not summarise papers, and does not search academic databases. It captures and cleans up your own spoken reaction to a paper you have read yourself. That boundary is intentional: a summary of a paper you never opened is not a literature note, and it is the kind of thing that becomes uncomfortable in a viva.

Is dictating literature notes actually faster than typing them?

Most people speak far faster than they type, and the bigger gain is that speaking lowers the barrier to starting at all. A reaction you would never sit down to type at eleven at night is one you will happily say out loud in ninety seconds. The historic problem was that spoken notes read like sludge, which is why Yaps runs cleanup on-device by default to strip filler words and self-corrections and fix punctuation. You end up reviewing a readable note instead of retyping a transcript.

Can I search my research notes from three years ago?

Yes. Notes saved in the Yaps vault are searchable on your own device, so you can look for a half-remembered phrase, an author surname, or a concept and get your own reasoning back. This works best if you build one habit while dictating: say the author name and year out loud inside the note, because search can only find words that are actually in the text. Notes also support checklists and kanban boards, which helps when a pile of reading turns into a chapter plan.

Does Yaps work offline in a library, an archive, or on a train?

Yes. Speech recognition, cleanup, note storage, and local search all run on your device and need no internet connection, which suits basement archives, field sites, and long train journeys. The network is used only for a short list of things such as billing and subscription status, model downloads, optional cloud voices, and optional premium vault sync. Capturing and saving a note never requires you to be online.

Can Yaps transcribe my supervision meetings?

Not live. Yaps does not join calls, does not put a bot in a meeting, and does not do live meeting transcription today. If you record a supervision yourself with everyone's clear and informed permission, the Yaps Studio editor imports that audio file and transcribes it offline into text or SRT subtitles on your own machine, which suits recordings you would rather not upload. For anything live, you will need a different tool and a clear conversation with your supervisor about consent.

How much does Yaps cost, and is there a student free tier?

Yaps offers a 7-day free trial, after which it is a paid subscription. The plans are Pro at $15 a month and Max at $25 a month, and an account is required only for billing. There is no permanently free tier. Core dictation, cleanup, and note capture run on your device throughout, so the subscription pays for the product rather than metering how private your notes are.

Are my unpublished research notes private?

Yes, by design. Your dictated audio is not uploaded to be turned into text, and notes are stored in a vault on your own device rather than a vendor cloud account. That matters when your notes contain unpublished arguments, unreviewed data, or a frank opinion about work by someone who may later review yours. If you turn on premium sync, notes move directly between your own phone and laptop over your local network or an encrypted peer-to-peer link, rather than through a central server.

Does Yaps work on iPhone or iPad?

Not yet. Yaps runs on Android as a full keyboard, and on Windows, macOS, and Linux as a desktop app, plus a Chrome extension for saving web pages into your vault. An iOS version for iPhone and iPad is coming soon and is not publicly available today. If your entire reading and annotation setup lives on an iPad, that is the honest gap to weigh before you commit.

Can Yaps read my draft chapter aloud so I can catch clumsy sentences?

Yes, and it is one of the most effective proofreading habits available. Select a passage and trigger read-aloud with Option and Fn on desktop, or choose "Read aloud with Yaps" from a text selection on Android. Sentences that loop back on themselves or run out of breath are hard to see and easy to hear. There are 18 voices on desktop, 8 of which work offline, and 2 on mobile, though the shipped voices are English speakers in practice, so a multilingual cloud reader is the better fit if you draft in another language.

Does dictation work in languages other than English?

Yes. Yaps dictation recognises about 25 spoken languages and auto-detects them from your speech, so you can capture notes in the language you actually think in without changing a setting. Recognition still runs on your device in every supported language, so the privacy benefit does not disappear when you switch. The honest caveat is read-aloud rather than dictation: the shipped text-to-speech voices are English speakers in practice.