Documentary paper edits. Keep every quote traceable.

Build a paper edit from reviewed recordings

A paper edit is a proposed story assembled from selected passages before the film is cut. Polyglot can transcribe authorized interview recordings on your Apple Silicon Mac, let you review their wording alongside playback, and export timestamped text. Arrange excerpts in a separate document or spreadsheet. The result is an editorial plan that a person can follow back to the footage.

Use this workflow when a producer or story editor needs to compare voices, test a sequence and hand over traceable selections. Polyglot does not cut video when you rearrange text, generate an edit decision list, sync separate production audio or create a finished documentary. If you need text edits to change the video automatically, a dedicated text-based video editor is a different tool category; Descript documents that behavior.

Give every source its own identity and clock

Before transcription, agree which approved recording the editor will use. Keep the source filename, interview or session ID, recording version and access conditions in a private source manifest. If an audio extract or proxy has a trimmed lead-in, identify that difference before anyone relies on its timestamps. Do not overwrite an original to make the times look convenient.

For each source, review the speech in Polyglot and export Text transcript with Include timestamps enabled. The timestamped TXT export contains cue starts, not word-level edit points or an in/out list. Its hours, minutes, seconds and milliseconds are positions in that project’s timeline. Keep unshifted source timing for this task and verify a recognizable early and late passage against the editor’s copy.

Camera timecode, a separately recorded sound file and an assembled sequence may use different origins. Label your reference as recording-relative elapsed time; do not present it as camera timecode. An editor must establish any necessary correspondence in the editing system. Polyglot does not recover embedded camera timecode or calculate a multi-camera sync map.

Keep source order separate from story order

Preserve an unchanged reviewed transcript for each recording. Copy proposed excerpts into a separate selects sheet rather than rearranging the source transcript or changing its cue timing. Give each selection a stable ID, source ID, approximate start and end checked in playback, verbatim wording, surrounding question or context, proposed story position and review status.

A cue start helps you find the passage. Choose the actual beginning and ending by listening; the next cue start is not a dependable endpoint for an editorial selection. If one sentence crosses several cues, keep its full meaning and record the chosen range yourself. The sheet below is an invented working example, not an app export or a customer project.

Fictional paper edit v01
Clock: source elapsed time

Story 01 / Select B03
Source: interview-B-v1.mov
Range: 00:12:04–00:12:16
“We reopened in the spring.”
Context: after the closure

Story 02 / Select A07
Source: interview-A-v1.mov
Range: 00:03:10–00:03:24
“At first, we could not
afford the repairs.”
Context: before reopening

Order is a proposal.
Both ranges need playback.
No final-film times yet.

Review the joins, not just the selected words

Listen to the question, the selected answer and the surrounding passage before approving a quote. A later sentence may correct a date or narrow a claim. Keep that information with the selection even if it will not appear on screen. Mark paraphrases and proposed narration separately so they cannot be mistaken for recorded words.

Now read the proposed sequence across its joins. In the example, placing the repair problem after the reopening could sound like a new problem unless the film establishes the change in time. Reordering is an editorial choice; the transcript alone cannot establish that it is fair. Do not splice distant fragments into a seamless quotation without reviewing their combined meaning and audible transitions.

For oral-history material, the Oral History Association specifically recommends contextualizing excerpts, retaining the narrator’s perspective and pointing back to the full interview. Apply the project’s agreed review and access conditions. Permission to record or read an interview does not itself settle permission for a public excerpt, a new distribution channel or a translation. Hold restricted material out of shared working documents.

Make a handoff another editor can follow

Send the approved source manifest, reviewed source transcripts and versioned selects sheet through the project’s agreed channel. Keep source IDs attached when moving rows. Identify unresolved words, translation decisions and missing permissions explicitly; a fluent-looking draft is not approval. Polyglot does not identify speakers automatically, so check any names or roles you add manually.

Ask the receiving editor to locate every selected range in the agreed media, confirm its first and last audible words, and listen around both boundaries. Check that two excerpts with the same elapsed time have not been confused across recordings. If a source is replaced, shortened or relabeled, recheck affected selections before using the sheet. Keep story order provisional until the editor accepts the references.

Only after assembly should you prepare subtitles for the actual film. Source transcripts do not share the finished sequence’s timing, and rearranging their text does not produce a synchronized caption track. Review speech, meaningful sound and readability against that final cut, then deliver its own checked SRT or WebVTT. The documentary subtitle workflow below covers that later stage.

Plan the local transcription step

Polyglot runs on Apple Silicon with macOS 14.4 or later. Speech recognition supports 25 European languages; translation targets are separate. Check your source language and review difficult or mixed-language passages with someone who understands them. Media and transcript processing stays local after model setup; see the privacy page for downloads, licensing, updates and analytics boundaries.

New AI transcription jobs are unlimited during the seven-day trial and require Pro afterward. Editing and exporting completed tracks remain free. This page describes a manual production workflow supported by transcript exports, not a tested integration with a video editor or an automated story, rights or fact-checking service.

Keep exploring.