Shape Seeds
Menu

Client project · Audio archive platform

AudioMeta

Every hour of the archive, searchable to the second.

A research and cataloguing workspace for a sound archive holding six decades of oral history, broadcast and field recordings. AudioMeta lets archivists and researchers listen, read, search, tag and assemble material in one place — with the transcript and the playhead moving as one.

AudioMeta waveform and transcript workspace on a graphite monitor in a dark listening studio

Try the synchronised player

OH-1978-0412 · Shipyard workers, reel 3 of 5 Transcript verified
00:14:00
Click any word to jump
Interviewer00:14:00
Narrator00:14:07
Interviewer00:14:19
Narrator00:14:22

Live preview — illustrative recording and transcript. Terms in amber are linked to the archive’s controlled vocabulary.

Client
Regional sound & oral-history archive (under NDA)
My role
Lead product designer — research, IA, interaction, UI system
Users
Archivists, cataloguers, researchers, editors
Platform
Desktop web app on the archive’s own infrastructure

The context

An archive you could only read by listening to all of it.

The archive had digitised decades of reel-to-reel tape, cassettes and broadcast masters. The sound was safe. Finding anything in it was not.

Catalogue records were inherited from card indexes and three generations of spreadsheets. A typical entry held a title, a date and a reel number. To answer a researcher’s question, a cataloguer listened in real time, took notes by hand and replied days later. Valuable passages were found once, then lost again.

  • 38,400recordings in scope
  • 61,000 hof digitised audio
  • 1954–2023date range
  • 14languages and dialects
01

Listening does not scale

Real-time playback was the only way to know what a recording contained.

02

Metadata without meaning

Inconsistent fields, free-text tags and duplicate names made search unreliable.

03

Findings never compound

Each research request started from zero; nothing was saved back to the source.

Listen & read

Read with your ears. Listen with your eyes.

The transcript and the waveform share a single timeline. Wherever you are in one, you are in the other.

Words are aligned to the audio at word level, so clicking a sentence starts playback exactly there, and the active phrase follows the playhead as it moves. Speaker lanes above the waveform show the shape of a conversation before you hear a second of it.

Scan mode

Pitch-corrected playback up to 3×, with silence and tape hiss compressed out. An hour-long interview can be surveyed in under fifteen minutes.

Jump by structure

Move between speaker turns, chapters, tagged moments or search hits with the keyboard — never by dragging a scrubber and hoping.

Correct in place

Transcripts are editable while you listen. Each correction keeps its author and timestamp, so verified text is clearly separated from machine output.

Tag & classify

Machines suggest. Archivists decide.

Tags are anchored to time ranges, drawn from the archive’s controlled vocabulary and always attributed.

Automated analysis proposes topics, places, people and content types. Each suggestion shows its evidence and waits for a person to accept, refine or reject it. Batch tools let a cataloguer apply a decision across hundreds of recordings while seeing exactly what will change.

Taxonomy

  • Work & economy
    • Industry
      • Shipbuilding
      • Steel
    • Industrial action
  • Places
    • Gdańsk
    • Nowa Huta
  • Content type
    • Interview
    • Music

Suggestions for 00:14:02 – 00:15:40

  • Industrial action“strike”, “stopped”, “shift” · 3 mentionsAccepted
  • Working conditions“cold”, “froze”, “overtime”Review
  • December 1970Date reference, cross-checked with catalogueReview
  • WeatherLow relevance to the passageRejected

Patterns

See what the collection is saying.

Any query can be viewed as a distribution — across decades, places, speakers or collections.

Researchers spot where a theme rises, disappears or moves between communities, then click a cell to hear the passages behind it. Curators use the same view to find under-described parts of the collection and plan cataloguing work.

50s60s70s80s90s00s10s20s Labour & industry Housing Migration Faith & ritual Music & dance Childhood
Topic density by decade of recording — illustrative view. Darker cells contain more tagged passages.

Bespoke records

Turn findings into a record that lasts.

Passages from different recordings can be gathered into a curated record — an exhibit, a research dossier, a broadcast package.

Every clip keeps its link to the source, its rights status and its citation. Records can be annotated, versioned and exported as a finding aid, a citation list or a cleared audio package, and they become searchable objects in the archive themselves.

Record · R-2025-017

December at the gates

EAD / XMLPDFWAV + citations
  1. 01 “Somebody chalked the hours on the gate.” OH-1978-0412 · 00:14:02 – 00:15:40
  2. 02 “That was the morning the strike began.” OH-1979-0088 · 00:40:51 – 00:42:10
  3. 03 “They refused overtime first.” BR-1980-1131 · 00:03:30 – 00:04:48

How we got there

Designed beside the people who know the tapes.

  1. Weeks 1–3

    Contextual inquiry

    Sat with cataloguers and reading-room staff, mapped twelve real research requests end to end.

  2. Weeks 4–6

    Data model & IA

    Defined the passage as the core unit, reconciled legacy fields into a single schema and vocabulary.

  3. Weeks 7–14

    Prototype on real audio

    Tested the synchronised player and query builder on the archive’s own recordings, every two weeks.

  4. Weeks 15–24

    System & rollout

    Built a component library with engineering, migrated collections in phases, trained staff on site.

Outcome

From days of listening to minutes of finding.

  • Minutesto locate a known passage, down from a typical half-day of listening
  • 3×more recordings described per cataloguer each month after rollout
  • Oneworkspace replacing a media player, spreadsheets, a text editor and a shared drive

The archive stopped being a store of tapes and became a body of evidence people can question. Research requests are answered with timestamps, not apologies — and every answer makes the next one faster.

Next case study06 / 08

Lelit Distrikt

Personal project · 3D & AI exploration

Exploring life in a simulated society through an Unreal Engine experience operated with CPS Studio.

Read the case study

Let’s work together

What could be easier to use?

Tell me about the product and the part that needs fixing. A few sentences is enough.

Tell me about it