Changelog
What's new in Speaker Cut.
Every release, and what changed. Speaker Cut checks for updates on launch and nudges you when there's a new one — here's what you're getting.
1.7.3LatestSep 25, 2026
- FixedThe Export Package now rebuilds the Final Cut file before it packs it. It had been reusing whatever that export was the last time you built it, so a package could ship a timeline made by an older version of Speaker Cut — including one without the auto-mix lanes 1.7.2 added. Exporting a package for a saved project you haven’t reopened can’t rebuild it, so the READ ME now says plainly when the file is older than your latest cut.
- FixedFinal Cut could refuse the deliverable outright (“the XML document could not be imported”) when it carried auto-mix. The lanes were written in an order Final Cut’s format doesn’t allow. Every timeline Speaker Cut writes is now checked against Final Cut’s own format definition before release.
1.7.2Sep 25, 2026
- FixedThe Export Package now carries auto-mix. The one-file Final Cut export — the one the package tells you to import — was written with the multicam’s own audio and none of the keyframed mic lanes, so the bleed reduction you set up never reached the timeline. It now arrives the way the single cut export always has: each person’s mic on its own lane, volume keyframes that open only while they’re talking, the program mix muted. The highlight and hook reels in the same file keep the program mix.
1.7.1Sep 25, 2026
- FixedA muted guest track no longer hijacks the edit. Some recorders — StreamYard’s per-guest exports among them — mute a guest’s track completely whenever they aren’t speaking. Speaker Cut measures every mic against its own quiet level, and on a track like that the “quiet level” is pure silence, so the small amount of everyone else’s voice bleeding onto it looked like the loudest thing in the room. The quiet guest took the screen and the people actually talking got cut away from — and muted by auto-mix. Speaker Cut now recognises a muted track and measures it properly. On a real three-person episode the guest who spoke 9% of the time had been holding 45% of the screen; he now holds 11%, and the host who does half the talking went from 29% to 51%.
- FixedAuto-mix no longer ducks someone mid-sentence on those same recordings. A mic now stays open for 93% of its owner’s speech, up from 67%.
1.7.0Sep 17, 2026
- ImprovedAuto-mix rebuilt — it now knows whose voice it is. On an in-person shoot every mic hears everyone, so judging each mic on its own let bleed through. Auto-mix now decides across your mics: a lane opens only while its owner is the one talking, and a gate can’t be carried into someone else’s speech. On a three-person table shoot, the share of the episode with exactly one mic audible went from 28% to 76%, with less of each person’s own speech ducked than before.
- NewDeep clean bleed. One click on a multicam project and Speaker Cut checks every quiet stretch two more ways: whether two mics carry the same words at the same moment, and whether the audio on a lane is actually that person’s voice. Anything that’s somebody else gets muted. Same shoot: one mic audible 91% of the episode. Takes a couple of minutes, once per project. Speaker Cut One / Pro.
- NewSee auto-mix before you cut. The waveform for each mic now shows the raw track dimmed with what auto-mix leaves in gold on top, so you can watch the bleed reduction slider work instead of exporting to find out.
- ImprovedCleaner keyframes in Final Cut. Every speech event is written the way you’d draw it — muted, ease in, hold, ease out, muted — with the mute pinned flat between events. A 75-minute episode went from 30,000 keyframes to about 6,000, so the timeline stays easy to work with.
- NewSpeaker Cut remembers your hosts’ voices. Each show profile can learn what its hosts sound like, and from then on they’re named automatically — in Highlight Cut before the AI even reads the transcript, and in multicam projects, where it offers to name the lanes for you. Only your listed hosts are learned, never guests, and only a voice fingerprint is stored on your Mac — no audio.
- NewSearch the transcript. A search box on the Transcript tab with match counts and next / previous (⌘F, ↵, ⇧↵). Punctuation and split words don’t hide results, and in Highlight Cut jumping to a match drops the playhead there so Space plays it.
- FixedReopening a saved multicam project after working on a Highlight Cut left the transcript in single-video editing mode.
- FixedBlank waveforms when the window wasn’t in front while a project loaded.
1.5.0Sep 17, 2026
- NewHighlight Cut — highlights from one video. No multicam, no separate mics: choose a single recording (a Zoom call, a StreamYard export, a phone video) and Speaker Cut tells the voices apart, names who’s who from the conversation, writes the transcript, and finds your highlights and hooks. Export them as Final Cut timelines like any other project. Speaker Cut One / Pro.
- NewYour show profile names the hosts. Pick a show when you add the video and Speaker Cut listens for its hosts, spells their names the way you saved them, sets the head-count to your hosts plus a guest, and takes the guest’s name from the session name. The show’s tone, audience and past ratings then guide the notes and highlights.
- NewEdit speakers like a doc. Click a speaker’s name on any line to reassign it, and Speaker Cut offers to fix the other lines that sound like the same person. Click a word and press Enter to split a line in two. Your edits build on each other and are never re-rolled away. A preview plays the video at the playhead, and Space plays a few seconds from any word.
- NewNot sure about a voice? Give it more samples. Each voice shows several clean samples to listen to. Mark who’s speaking in a few of them and Speaker Cut re-separates the voices around your picks in seconds.
- NewThe AI explains its picks. Why these picks sums up what it looked for and what it left out; every highlight says why these lines; and after you send a note, What changed tells you how it read your request (expand, tighten, rework or replace) and what it did.
- ImprovedHighlights that tell a complete story. Each highlight now covers the setup, the turn and the payoff (or an idea, where it came from, and why it matters), drawn from the whole episode, and every piece starts and ends on a whole sentence. Pieces can run up to 2 minutes and a highlight up to 3.
- ImprovedNotes that do what you asked. “Expand this” keeps your lines and builds around them. Ask for lines from a specific person (“add three lines from the guest”) and Speaker Cut checks each one is really that person talking, retries if not, and tells you plainly if it came up short.
- ImprovedClearer highlight lengths. A highlight now shows its real run-time across all of its moments (“1:24 long · 3 clips”), and every piece shows its own length.
- ChangedTimestamps no longer jump Final Cut’s playhead. The jump relied on Final Cut being in front with the right timeline open and could misfire, so we removed it from the transcript and highlights.
- FixedOccasional save errors when several changes were saved at the same moment.
1.3.2Sep 6, 2026
- NewBuild a multicam from clips. Drop your cameras and audio recorders straight into Speaker Cut — it listens, syncs them by audio, and assembles the multicam for you, then cuts. Handles ATEM shoots that split into multiple clips per angle, external recorders, and a program feed as the sync reference. No Final Cut round-trip first.
- NewAuto-mix is back — rebuilt as editable keyframes. Each speaker’s mic lands on its own lane, in sync with the cuts, with volume keyframes that duck it whenever they’re not talking — bleed and room tone drop out, every reaction stays in. Nothing is baked: the automation is right there in Final Cut to nudge by hand, and it applies instantly with no rendering. Pick a floor from Gentle to Gate (full mute). Speaker Cut One / Pro.
- ChangedLevel-matching stays paused while we rework it; auto-mix no longer depends on it.
1.3.1Sep 6, 2026
- FixedFinal Cut import error on 44.1 kHz projects. Exports from projects with 44.1 kHz audio hit a “DTD validation failed” error on import (the sequence audio rate was written as “44k” instead of “44.1k”). Fixed — the FCPXML now imports cleanly with no hand-editing.
1.3.0Aug 28, 2026
- NewTwo-shots now get used. When two people trade lines, the cut holds the two-shot instead of ping-ponging between their close-ups.
- NewB-cams for variety. Give someone more than one camera and the extras become B-cams, punched in sparingly while their main camera carries the rest — with a new B-cam variety control and a primary-camera picker on each lane.
- ImprovedSharper speaker tracking. The cut follows whoever actually takes the floor, so it no longer sticks on the wrong person during overlapping or quick exchanges.
- FixedNo more stray empty frame. Fixed an occasional single blank frame in the cut timeline.
- ChangedLevel-matching & auto-mix are paused. We’re reworking them to handle separately-recorded mics properly; they will return in a later update. Your cut and Final Cut export are unaffected.
1.2.0Aug 25, 2026
- NewTell the AI what you’re looking for. The Highlights & Hooks generator now has a “What are you looking for?” box — type something like “the strongest business advice,” “funniest moments,” or “anything a little spicy,” and the picks reflect your request. It steers the first pass and every re-roll.
- NewStitched highlights. A highlight can now weave together moments from different parts of the episode into one coherent clip — the setup early, the payoff later — instead of only a single unbroken timestamp. Stitched clips are marked with a ⧉, and export as one clean, gapless cut.
- NewRe-roll with a note. Re-rolling a single highlight or hook now opens a box where you tell the AI what to change — “expand this,” “make it punchier,” “focus on the pricing story” — and it revises that exact clip instead of jumping to something unrelated. Leave it blank for a fresh, different pick.
1.1.12Aug 24, 2026
- ImprovedThe in-app “Send feedback” reporter now lets you flag your note as a Bug, Feature, Kudos, or Question — so we can triage and act on it faster.
1.1.11Aug 24, 2026
- ImprovedClearer guidance when you drop a file that isn't a multicam — Speaker Cut now tells you exactly how to export a multicam from Final Cut instead of a cryptic error.
1.1.10Aug 24, 2026
- ImprovedOne/Pro features (Match levels, Auto-mix, Transcript, AI notes) now show as a frosted preview when you're on the free app — so you can see exactly what you'll unlock, instead of a blank space.
1.1.9Aug 24, 2026
- NewReport an issue. Hit a snag? There's now a "Report an issue" link in the footer — send us a quick note and we get the context we need to fix it fast. Never your media, transcript, or AI output.
1.1.8Aug 23, 2026
- NewAuto-mix — reduce mic bleed (Speaker Cut One & Pro). For same-room interviews and podcasts, Speaker Cut can now duck each person's mic when they're not talking, killing bleed and room tone while keeping every reaction. It's driven by who's actually speaking (not a noise gate), so nothing gets chopped. Pick Gentle / Balanced / Aggressive / Gate, or dial the exact amount. Joins the Match levels tool as part of One & Pro; the cut & Final Cut export stay free.
- NewClipping warning. The Levels panel now flags any mic that was recorded too hot (peaking near 0 dBFS), so you can catch a distorted source before you cut.
1.1.6Aug 23, 2026
- FixedGenerating a cut no longer pops open a second Speaker Cut window ("Couldn't read that project"). Speaker Cut also no longer takes over as the default app for Final Cut XML files — that stays with Final Cut Pro, where it belongs.
1.1.5Aug 23, 2026
- FixedClicking a transcript timestamp now just moves the Final Cut playhead — it no longer accidentally cuts the timeline or switches the multicam angle. (When a multicam clip was selected, Final Cut was reading the timecode digits as angle-switch shortcuts.)
1.1.4Aug 23, 2026
- FixedMulticams built from compound or synced clips — like StreamYard, Riverside & Zoom per-speaker recordings — no longer show "no audio" on every lane. Speaker Cut now reads the audio from inside those clips, with the correct sync, so your people are detected automatically.
- ImprovedSpeaker Cut now reads each lane's actual streams: an external audio recorder is offered as a mic but not as a camera to cut to, and a silent camera is offered as a shot but not a mic — so the Video/Audio choices and warnings match your real footage.
1.1.3Aug 22, 2026
- NewYour version now shows in the app footer, with a "What's new" link straight to this page — so you always know what you're running and what changed.
- FixedReliability improvements to update checks.
1.1.2Aug 22, 2026
- NewOptional, anonymous diagnostics so we can spot and fix problems faster — often before you'd even report them. Events and errors only; never your media, transcript, AI output, or license key. Turn it off anytime in Settings.
- FixedReliability improvements to license delivery.
1.1.0Aug 21, 2026
- NewThe full episode transcript — plus AI show notes, highlights & cold-open hooks — is now part of Speaker Cut One and Pro.
- NewExternal-recorder support: assign each speaker's audio lane separately from their video lane, so Speaker Cut can detect a speaker on one track and cut to a different camera.
- ImprovedA proper "drag to Applications" installer, fully signed & notarized by Apple.
- ImprovedThe free app still does the multicam speaker-cut and Final Cut export — always free, no account.
1.0.0Initial release
- NewAuto-cut your Final Cut Pro multicam to whoever's talking — drop in the FCPXML, get back a finished timeline.
- NewOn-device speaker detection, loudness-matched audio, and a one-click export package. Your footage never leaves your Mac.