Audio file import, captions from any clip and speech models you bring yourself

Drop an audio file on Review and it becomes a clip — a WAV, MP3, M4A, FLAC, OGG or AIFF file, or the audio track of an MP4, MOV or WebM, is run through your model, your languages and your corrections like a dictation, then waits in the queue to be reviewed. Use the Import audio button or drop files anywhere on the window; several at once transcribe one after another, and the file's name becomes the clip's title.

Every clip exports its own captions — a clip's new Export menu copies SRT or VTT with its timings, and Plan dub in Studio hands the corrected transcript to the dub planner.

Bring your own speech model — Settings - Models has Import model file… for a .bin you fetched yourself: a slow connection, a proxy, or a build the shelf does not list. The file is checked to really be a Whisper model before it joins the shelf, and it becomes the profile's model when the one named is not installed.

Live translation can be switched on from Dictate — the captions toggle sits beside Stream now. If the pair you picked has no pack yet, flipping it takes you to Settings - Speech to take one rather than failing quietly.

Your music pauses itself while you dictate — the moment the overlay appears, whatever Now Playing is running is paused, so a chorus cannot join the transcript. On macOS it is a directed pause — a player you already paused is never started again; on Windows the Stop key goes out for the same reason.

Context keeps a pack per project — import more than one folder, mark which is active for dictation, and rename or delete a project from its card. Scanning a folder again refreshes its terms and keeps the pins and exclusions you set, and a term search sits above the list for the packs that outgrew a glance.

A paired phone can sync this profile — while a captions pairing session is live, the phone pulls your corrections, personal vocabulary, context pack terms and learning samples over the same link, and pushes its own changes back. The newest edit wins, a deletion on either side deletes on both, and a phrase changed on both devices at once is named as a conflict rather than silently merged. What crosses is sealed end to end under the session PIN — see Privacy.

Starting a dictation no longer waits on the device list — with no microphone pinned, capture opens straight on the system default and the picker fills in behind it. And until the microphone is really open, the overlay says Opening the microphone… instead of claiming to listen into dead air.

Free remembers twice the corrections — the free tier keeps 100 learned rules now, up from 50. Pro is unchanged: all of them, plus scoping, merging and export.

A finished model download stays finished — the progress bar holds at the end until the install is confirmed, so a row no longer flashes back to Download before it shows as installed.

Studio and Voice Clone leave the rail while the feature set is finished — the same treatment Actions got: the rail shows what is ready. The dub planner still opens from a clip's Export menu, and Voice Clone — cloning your own voice from a few saved dictations, then speaking new text on this device — is built into this release ahead of its space opening.