Transcript publication framing

Transcript publication framing

Why publish transcripts: the spoken corpus counts as much as the written one.

What a transcript is: cues with a timecode, a speaker and a text, read by the reader plugin.

Pseudonymisation before indexing: no real name enters the search index.

The dictionary of pseudonyms is never written into the published site.

Roles kept instead of pseudonyms when the configuration asks for it.

Publication is off by default: publish_transcripts must be requested explicitly.

Detected personal data outside the dictionary raises a finding for review.

Timecodes are anchors: a cue is citable from any page.

Cues are grouped by consecutive speaker in the rendered page.

The decisions extracted from a meeting point at their cue.

Twin resources: a deck, its notes and its transcript form one page.

Grouping criteria: folder, date and textual overlap.

One entry in the search index for the three files.

The original file stays downloadable, whatever the conversion did.

Conversion at publication, cached by fingerprint.

The extracted text feeds recognition and similarity alike.

The position of a passage, page or slide, is kept for the citation.

What the reader sees without JavaScript: the first page, the text, the notes.

The viewer loads on demand and never in the initial bundle.

Staleness: a transcript older than the threshold is reported like a note.

Open questions: keep_roles per source, the size of the cache.

Next steps: the privacy framing note, the review process.

Questions