NewTypeCola/VOKU-Archive

VOKU Archive

VOKU Archive brings 66,010 pages of student-produced campus radio cue sheets and scripts from 1988–2019 into a form that people can read, search, and study. VOKU (Voice of Kyonggi University) is a student-run university broadcasting station whose work also includes video production and stage and live event production and operation. This collection covers the radio materials within that wider activity.

A former station member made this digital archive independently so that documents created by earlier members would remain readable instead of being left in storage. Alongside redaction, pseudonym tokens were inserted to preserve, where possible, repeated participation by the same producer and collaboration within episodes. The tokens are part of the prepared public images; they were not names written in the historical scripts.

This is a selected surviving archive, not a reconstruction of every broadcast during those years. It contains preparation documents, which do not establish that everything written was performed on air. Scope, exclusions, and known exceptions are collected in the release notes.

What is provided

Material	Content and use
Final WebP images	66,010 pages showing the prepared public version of the documents. Use these to check what is actually displayed.
Fresh OCR	Machine-recognized Korean text, lines, confidence values, and coordinates read from those final images. Useful for search, navigation, and subsequent processing; not a human-corrected transcription.
Public token annotations	91,776 records on 25,162 pages, connecting displayed pseudonym tokens to their placement boxes and associated recorded name-removal boxes.
Metadata and schemas	Page, episode, and program-unit identifiers; Korean and editorial English titles; image hashes; file locations; field definitions; and validation records.

The same corpus is supplied as 32 yearly TAR files for ordinary files and 54 Parquet shards for page-level access. The TAR files contain 5,330 episode OCR JSON files and 5,330 corresponding annotation JSON files. These are two access forms for the same pages, not separate corpora. Detailed counts and checks.

A page from the archive

여행에 미치다 / Crazy About Travel, 2017, first semester, source episode bundle 02, page 01. The English title is an editorial translation. This preview is a byte-identical copy of images/2017/voku_2017_1_crazy_about_travel_02_page01.webp inside the 2017 TAR, not the workflow's synthetic-identifier demo.

Read the program heading, planned broadcast information, production-role line, and music rundown as parts of the document. The role line contains the inserted tokens 물빛수첩, 물빛결빛, and 노을노래. Its annotation page has three token records. The blurred approval area illustrates why token annotations are not a map of every redacted region. How to read a token record.

Open one episode

Download archives/voku_2017.tar. With a working directory containing archives/, the following standard-library Python reads one episode directly from the TAR; no conversion or extraction is needed.

import json
import tarfile

document_id = "voku_2017_1_crazy_about_travel_02"
with tarfile.open("archives/voku_2017.tar", "r:") as archive:
    ocr = json.load(archive.extractfile(f"ocr/2017/{document_id}.json"))
    annotation = json.load(
        archive.extractfile(f"annotation/2017/{document_id}.json")
    )
    page = next(p for p in ocr["pages"] if p["page"] == 1)
    marks = next(p for p in annotation["pages"]
                 if p["page_id"] == page["page_id"])
    print(page["page_id"], page["width"], page["height"])
    print(page["text"])
    print(marks["annotations"])
    webp_bytes = archive.extractfile(page["image"]).read()

The selected page is 1191 × 1684 pixels. A WebP-capable image viewer can open it after ordinary TAR extraction. The data guide explains the JSON fields and provides a short Parquet example.

Find and interpret data

• Page index: locate each image, its yearly archive and Parquet shard, and its two episode JSON files.
• Episode index: group pages into source bundles and find episode JSON hashes.
• Data guide: identifiers, title policy, OCR coordinates, and token annotation semantics.
• Release notes: counts, recorded validation, preserved exceptions, and the single corpus split.
• Rights and credits: CC BY 4.0 scope, attribution as VOKU Archive, and the creator's separate credit.

Production history

The separate VOKU Archive Workflow repository on GitHub documents the judgments, review, corrections, and production history behind the images. Its synthetic-identifier demo illustrates processing and is not this dataset's example data. The dataset can be used through the documents above without first reading that history.

Reuse

VOKU Archive is made available under CC BY 4.0 within the scope described in Rights and credits. The designated attribution name is VOKU Archive. The unmodified license text is in LICENSE.