Releases

First version for Omeka Classic, the counterpart of the Oral Archivist 1.6 module for Omeka S.

• Pipeline for audio and video files of items: extract audio (ffmpeg), transcribe with word-level timestamps (OpenAI Whisper), clean up into paragraphs with a language model (the raw transcript is kept), segment into chapters with topics, and an optional summary. It runs in Omeka's background jobs; a failed stage is tried again with a pause, and a retry resumes at the stage that failed, so nothing is billed twice. A MySQL lock per interview tells a running job from one that died (shown as stalled, to retry), and each interview keeps a processing log.
• Providers: OpenAI, or any service with OpenAI's API at the address set in the settings (or oral_archivist.openai_base_url in config.ini), and a free mock provider for trying the workflow. The API key is stored encrypted with a key derived from db.ini, or set in config.ini.
• Queueing from a file's admin page (with the estimated cost), from Batch processing (bulk, with the estimate of the selection), or automatically when files are added (off by default), with a daily cap; files passed over are listed for staff.
• Review queue in the admin menu, with a badge counting interviews waiting, and the review screen: player, editable paragraphs (merge, split, speakers), low-confidence flags with play-snippet, chapters and topics, summary, footnotes anchored to words and linked to notes, URLs, items or files, re-run segmentation, regenerate summary. Nothing is public until staff approve it. Super Users and Admins review; the settings are for Super Users.
• Public display: a synchronized player on item pages (after the metadata, or under the title) and file pages (in place of the file's player, alongside Html5 Media), with chapters, search-to-moment, share and cite, downloads, footnote popups and references; three layouts and five player styles colored by the theme's tokens; the [oral_history] shortcode for simple pages and exhibits; the oralHistoryPlayer() view helper for themes.
• Segment pages at /oral-history: search across every interview's segments, topic pages and segment permalinks, for the items the visitor may see; an optional link in the main navigation.
• Site search: published transcripts join the search text of their file and item, following the item's visibility; unpublishing takes them out. AI summaries are written to an item element of your choice.
• Export and import of curated transcripts as JSON, the same format as the Omeka S module's (its media links import as files).
• Costs: actual cost per stage, the month's spend on the settings page, and a monthly budget that refuses queueing and stops jobs before their next paid stage.
• Staff actions (publish, unpublish, queue, retry, import, re-segmentation, summary, settings) are logged to Audit Trail when it is installed.