Bulk transcription
Nothing gets lost between the subject and the transcript.
Transcription and translation run side by side as the interview happens. Speakers are separated and labelled automatically. Share a live view with your translator or remote producer without them needing to be on set.
- Four real-time transcription engines covering 20+ languages. Pick the best fit per session.
- 20+ source languages with live translation to a target of your choice.
- Live-share links for up to 25 viewers, each picking their own translation.
- A producer chat on the live share: remote editors and executives message the person in the room, and the subject never sees it.
- Local recording in the browser as a safety net for outages, and a retranscription panel to re-run the recording with another engine afterwards.
Live SessionOne name. Every appearance. Across every file.
Voice prints are matched automatically across your entire project. Rename a speaker once and the change ripples across every interview they appear in, including files you uploaded last month.
- Voice fingerprinting matches speakers automatically across every file.
- Stable identities you can merge, split, or rename project-wide.
- Per-file appearances tracked with timestamps for quick navigation.
- Automatic re-embedding when you correct or reset an identity.
- A change history for the project: every naming, rename, merge and reset, who did it and across how many files.
300 hours of rushes. Find the line you need in seconds.
The classic paper-edit problem: you remembered a quote but not the reel, so it got cut. Upload your library and search by idea rather than exact words. The chat goes further. Ask why something happened across a multi-day shoot and it draws the connections between speakers, surfacing causality and references that would otherwise be lost.
- Three search modes: Words, a fast typo-tolerant literal search; Meaning, which finds passages about the same idea in other words; and Deep scan, which reads every line when the scope is narrow enough.
- Chat retrieves soundbites by idea: describe what you're looking for and it finds the tape, regardless of the speaker's exact words. Every answer cites file, timecode, and speaker.
- Connected moments: while you read one passage, the related passages from other files sit beside it with a sentence on what links them.
- Interview analysis per file: an editorial summary and a breakdown of what the sitting covers, so a producer can triage a day of rushes before listening to any of it.
- Concept extraction pulls recurring themes across hundreds of hours; a project dossier keeps the people, the timeline and the relationships between them in one place.
Check every claim against the internet before it goes to broadcast.
Select any line from a transcript and the fact-checker queries the live web for corroborating evidence, contradicting coverage, and known counter-narratives. Built for journalists and investigative producers who need the public record next to the transcript, not as a separate step.
- Queries the live internet, not a fixed knowledge base. Results reflect what is publicly available right now.
- Surfaces counter-narratives and contradicting reporting the subject did not mention.
- On a multi-day shoot, links the claim to what the same subject said in earlier sessions, so the check covers prior context on tape as well.
- Verdict arrives with source links; corroborating and contradicting items listed separately.
"The bridge inspection reports were suppressed for eighteen months before the blackout."
One inspection report filed eleven months prior, not eighteen. Two others listed as pending.
Three engineers confirmed that safety memos were not forwarded to the oversight committee.
Timeline disputed. Suppression of memos corroborated. Follow up on the eleven-vs-eighteen-month discrepancy.
Transcribe and translate your entire library hands-free.
Drop in a folder of interviews and walk away. PaperCuts queues every file, runs transcription across them in parallel, and translates into your target language automatically. By the time you sit down to review, the whole library is indexed and searchable.
- Video is decoded and transcoded to audio in the browser before anything is sent, so a 4 GB card goes up as tens of megabytes. Uncompressed audio is converted to MP3 the same way.
- A dropped connection retries the upload by itself, without converting again. On a bad link in the field, the small file is what makes it through.
- Your choice of transcription engine per file. Swap any time to suit the source language, and retranscribe a file with a different engine later.
- Rate footage from one to five and mark folders as archive material, so a researcher's read of the tape travels with it.
- A hole left by a recorder that lost an input can be filled from an intact copy of the audio, without touching the selects and notes already built on the transcript.
- Files you trash stay restorable for 90 days.
ACTIVE UPLOADS · 3 FILESShape the story before you cut the picture.
Write the arc as a document: acts as headings, beats as paragraphs, notes in the margin. Drag a beat and it moves with its section. The structure of the film starts taking shape weeks before you sit in the edit bay.
- Multiple named beat sheets per project for parallel structures.
- A document editor, with drag-and-drop for blocks and whole sections, and notes pinned to any beat.
- AI-assisted beat generation from a topic, or a paper cut summarised into a draft.
- Beats stay linked to their source moments. One click back to the tape.
- Export to Word, or put the beat sheet on a Miro board for the room.
BEAT SHEET · ACT 1The Blackout. Power grid fails. Confusion turns to panic as backup systems fail to engage.
The Arrival. Establish setting. The calm before the storm. Key players in the control room.
First Refusal. Marcus refuses to authorize manual override. The schism forms.
The Override. Elena bypasses the chain of command.
The Phone Call. Reyes hears it second-hand.
Build your paper cut without leaving the transcript.
Pull selects directly from any transcript and arrange them into a spreadsheet-style assembly. Timecodes, speaker labels, and section headers travel with every quote.
- Spreadsheet-style editor with drag-and-drop reordering and keyboard shortcuts.
- Play the cut back as a radio cut. Click a word to jump there.
- Versions of the whole cut, and a pass on every row that records which round the bite arrived in.
- Import from Excel, Word, or PDF. Export as a sequence your editor opens directly.
The cut leaves the room with picture already under it.
A paper cut is built on the sound recordist's audio. Hand over the assistant's sync sequences and the export carries the camera angles with it, cut to the same frames as the quotes.
- Recorder files matched by length and folder as well as name.
- Bites with no camera on them export as sound and are counted.
- One package for Avid, Premiere, Final Cut Pro and Resolve: XML, EDL, subtitles, markers, and a conform guide.
A portrait of every character, drawn from their own words.
Character overviews assemble a structured portrait of each speaker from everything they've said on tape, across every interview in the project.
- Four modes: Portraits, Ensemble, Timeline Chronicle, Thematic Assembly.
- Optional Serendipities, Facts & Timelines, and Editor's Note sections per character.
- Per-speaker generation grounded in your project archive. Only what they actually said on tape.
- Export to .docx for sharing with directors, writers, and editors.
Elena is the engineer who saw the failure coming and was told to be quiet. Her account is precise where the others' are evasive. She returns again and again to the night the simulation broke.
In three separate sittings she uses the same phrase, "the numbers stopped making sense", without prompting.
- Joined the project in its second year.
- One of two people who held the override codes.
- Filed the safety memo eleven weeks before the blackout.
Her Day 04 timeline and Marcus's Day 11 account disagree by seven weeks. Worth putting to both.
Every AI decision on the record, and every one of them yours to switch off.
A documentary answers for what it puts on screen. Each AI feature in a project has its own switch, enforced on the server rather than in the interface, and every call is logged: who ran it, what it read, which model answered.
- Per-project switches for character overviews, connected moments, interview analysis, chat, paper cut assist, beat generation and AI voice.
- A noneditorial mode for the reading features: retrieve and quote, never characterise.
- A dashboard of every AI event in the project, by feature and by user, with audio minutes and words processed.
- The speaker change history sits beside it, so a name on screen can be traced to the person who assigned it.