What an archive will actually accept

First published 20 October 2026

Part 2 of 5 · The culture series — A listener, not a form · What an archive will actually accept · Off the record, restricted, and the two bags · Your family’s dish is a lesson · A community keeps its own archive

An archive does not accept “a recording of Grandma.” It accepts a recording together with a few plain, readable companions: a transcript that says who is speaking and where in the recording they said it, dates written so a machine can sort them, a note of who holds the rights, and a list of checksums so that in fifty years someone can tell the files are whole. Each companion has a name and a public specification, and none is exotic. What is hard is adding them afterward, from memory, when the person who knew has gone. Record with them in mind from the first conversation and a deposit becomes a hand-over instead of a rescue.

What a transcript looks like when an archive can read it

Here is an invented fragment in WebVTT, a W3C format for text tied to time in a recording:

WEBVTT

00:12:40.000 --> 00:12:47.500
<v Narrator>We sold the farm in, oh, around 1957.

00:12:48.000 --> 00:12:51.000
<v Interviewer>Was that before the flood?

Each cue says when, and the <v> tag says who. The W3C specification (Candidate Recommendation Draft, 20 May 2026) covers captions and subtitles and, more generally, “any form of metadata that is time-aligned with audio or video content”; its first example is the captioning of an interview. Aviary, the platform that OHMS (the Oral History Metadata Synchronizer, made at the University of Kentucky’s Nunn Center) has been integrated into, says WebVTT “is the preferred transcript format and has the highest likelihood of success when being imported into Aviary.” It also takes SRT, plain text, OHMS XML and Gentle JSON, and its indexes, the chaptered table of contents of a long interview, can be WebVTT, OHMS XML or AES-60 XML.

Scholars who want more ask for TEI. Its Guidelines (P5 version 4.12.0, 28 July 2026) give spoken text a chapter of its own, “Transcriptions of Speech,” where each turn is an utterance marked <u who="#mar">. Researchers who code interviews by theme work in ATLAS.ti, MAXQDA or NVivo, and the REFI-QDA Project file, ending in .qdpx, carries sources, codes, memos and cases between programs that implement it, though its authors are candid that “users may experience some modifications” on import. Their home page argues that “a common standard maximises the possibility of using project data in the future and for archiving purposes.”

How it is packed, dated and labelled

BagIt (RFC 8493, October 2018) is “a set of hierarchical file layout conventions designed to support storage and transfer of arbitrary digital content”: a folder of your files plus a manifest, such as manifest-sha512.txt, listing a checksum for each. The RFC names the Library of Congress and the Rockefeller Archive Center among its users. RO-Crate, now at 1.3, adds the description: put a file called ro-crate-metadata.json beside the files and the folder can be considered a crate, a method of “aggregating and describing data for distribution, re-use, publishing, preservation and archiving.” The Library of Congress’s PREMIS records what has happened to a file since, and its data dictionary asks every event it records to carry an identifier, a type and a date and time, so “converted to WAV, checked, transcribed” becomes a dated line instead of a rumour.

Dates are the family’s quiet trouble: people say “around 1957,” “the fifties,” “spring.” The Library of Congress’s Extended Date/Time Format writes them down without pretending: 1957~ for approximately 1957, 195X for a year with its last digit unspecified, 1957-21 for spring 1957. The hedge survives into the archive.

Rights come last, and they are the archive’s to apply. RightsStatements.org offers 12 standardized statements in three groups: in copyright, no copyright, and other, for works whose status is unclear. They are, the site says, “not intended to be used by individuals to license their own creations.” So you do not pick one. You bring the facts the archive needs to pick it: who made the recording, who spoke, who holds the rights.

This week

Look up the archive you might one day deposit with, whether a regional archive, a university’s oral-history program or a museum, and find its page on depositing oral history. It is usually a short page, and the answers differ. The University of Pittsburgh’s Archives & Special Collections recommends WAV or BWF as master files, with MP3 as the usual use copy. Somerset’s heritage centre in England asks for two copies of each recording, WAV and MP3, and accepts transcripts “in hard copy and/or electronically.” Ask yours which transcript format it prefers, and ask before you record.

Whichever archive you choose, four habits cost nothing: keep the original recording exactly as it came off the device; say each speaker’s name at the start; write down the recording date and, whenever someone gives a date, how sure they sounded; and note who holds the rights. If you tell your story to Maddie on formaddie.com, she keeps every word of both speakers, and each chapter comes with the complete transcript. That is the raw material every format above is made from. This piece does not claim she hands you any of those files today; it describes the archive’s side of the door.

the previous piece is about why a life story needs a listener rather than a form, and the next takes up what “off the record” and “restricted” should mean in a recording, and why a research copy and a family copy are kept apart. When the archive is your community’s own, a community keeps its own archive picks the thread up.

Where would you want your family’s recording to live in fifty years?

Where this connects

Sources

published by Jason Jeyanandan through his cognitive companion, one unit