Splitting one recording into several notes

Back-to-back calls do not always give you a gap to stop the recorder in. You come back an hour later and one file holds the standup, the customer call, and the design review — with one summary on top trying to be about all three at once, which means it is useful for none of them.

Splitting takes that file apart. You mark where one meeting ends and the next begins, and each stretch becomes its own note: its own transcript, its own summary, its own title. The merged original is filed away rather than deleted, so nothing you had is lost if you cut in the wrong place.

Nothing is re-transcribed. The split works from the transcript Oris already made for that recording, which is why it takes seconds rather than minutes. (The one exception is an old recording whose transcript has been pruned — see When it isn’t available below.)

From the Oris app

Every finished recording in the app’s Recordings history offers Split… in its row’s … menu (or on a right-click) for as long as its audio and transcript are still on disk. Choose it and a sheet opens with the recording as a bar across the top and its whole transcript underneath. The points Oris found are marked on the bar and between the sentences where they fall, and they start switched on, so if the suggestion is right the whole job is one more click on Split into N.

The bar and the transcript are two views of the same cuts, so whatever you do on one shows on the other. On the bar, click a mark to switch it off, or click anywhere else to add a cut of your own. A cut always lands on the nearest sentence boundary, never in the middle of what someone was saying, and hovering over the bar shows the last words before that point and the first words after it, so you can see what you are about to separate before you click. Each click on the bar also scrolls the transcript to that cut, and the stretch of transcript you are reading is shaded on the bar.

In the transcript, each cut is a line between two sentences. The ones Oris found have a checkbox and say why they were found, such as a long silence or someone saying “thanks everyone”, while yours have a button to remove them, and hovering between any two sentences offers a cut right there. The bar shows how long each piece will be, and the button counts them.

While the split runs, the row’s own status line says which piece is being summarized, and when it finishes each piece appears as a row of its own with a small split pill, dated from where it began inside the recording. The original row says how many pieces it became. Where those pieces go depends on where the recording went:

• Obsidian — one note per piece, written beside the original. A box next to Split into N names the original note and archives it to the _archive/ folder of its own vault, as the plugin does. Archiving moves the whole note, so the box starts unticked when that note also holds another recording, such as a daily note with two meetings in it, and if another recording lands in the note while the window is open, the pieces are still made but the original stays where it is.
• Apple Notes and Markdown folders — each piece is delivered as a new note to the same destination the recording used, even a folder that normally appends to one note, and the original note is left where it is.

If a split stops part-way, the pieces that landed keep their notes and rows, the original row says how many of them there were, and the original note is left as it was.

From Obsidian

The Obsidian plugin’s command palette has the same thing as Oris: Split a recording…. Pick the .m4a, and Oris opens a window with the transcript on the left and the resulting segments on the right.

Oris proposes the cuts, you decide them. Before the window opens it reads the transcript looking for two things: long silences, and the phrases people actually say when a meeting is ending — “alright, thanks everyone”, “I’ve gotta hop”, “let’s wrap up”, “see you next week”. Each one it finds shows up as a faint suggested marker, ranked by how confident it is. Take the ones that look right, dismiss the ones that do not with the × beside them, and add your own: play the audio, and press m to drop a marker wherever you are. Clicking any transcript line seeks the playback there, so finding the seam is usually a matter of skimming to roughly the right minute and listening.

Every marker splits the recording in two, so N markers give you N+1 segments, listed on the right with their time range and how many transcript lines fall inside. For each one you choose either New note — giving it a filename and a folder — or Discard, which is what the five minutes of nobody-talking between two calls deserves.

The original note is handled for you. Oris looks through your vault for the note that recording landed in and shows it at the top of the window. On Apply, that note is moved into <vault>/_archive/ with a header linking to every note the split produced, and the original is removed from where it was — so you end up with the new notes rather than the new notes plus the stale merged one. If Oris picked the wrong note, Change… points it at another; Don’t archive leaves your original exactly where it is.

Each segment is then summarized separately, by whichever provider you have set up under Settings → Summarization — on-device by default. That is where the time goes on a multi-way split: a handful of seconds per segment rather than a re-transcription.

What each note looks like

The same shape a live recording produces — a summary up top, the full transcript in a collapsible block below — plus a small frontmatter block tying it back to where it came from:

---
title: "Q2 Planning Sync"
source_audio: "/Users/you/Recordings/2026-09-15.m4a"
source_range: 00:00-24:10
generated_by: oris-resplit
---

Timestamps inside each note start at zero. A meeting that began 47 minutes into the original file reads 00:00 at its own first line, not 47:12, because the note is about that meeting rather than about its position in a file you are not going to open again.

When it isn’t available

Splitting needs the recording’s transcript sidecar, which audio retention eventually sweeps along with the audio. In the app Split… leaves the row’s menu once either is gone, and it also stands down on a recording that has already been split, on the pieces themselves, on imports, on an Apple Notes or folder recording whose note hasn’t arrived yet, and on a recording too short to hold anything to split, such as one where nobody spoke or only one sentence was said. The Obsidian command can still split a recording whose audio survives but whose transcript has been pruned, by re-transcribing it first, which takes as long as the original transcription did — Oris says so before it starts.

From the command line

The same three steps the window drives are available directly, which is occasionally handy for a recording you want to split the same way every time, or for working out why a split did something surprising. In a source checkout they are bin/oris-resplit; in the installed app the same script lives at /Applications/Oris.app/Contents/Resources/oris/bin/oris-resplit.

bin/oris-resplit candidates <audio.m4a>     # propose split points, as JSON
bin/oris-resplit transcript <audio.m4a>     # make sure the transcript sidecar exists
bin/oris-resplit apply <manifest.json>      # apply a plan

candidates prints the suggested markers with a score each; apply takes a manifest describing the segments and where each one should land. The window builds that manifest for you, so the hand-written form is a diagnostic tool rather than the intended path.