Oral History Transcription

Incoming File

To create a transcript for an oral history interview, you will want to start with the access copy of the audio file (see file specifications here), or if the interview is video, create an mp3 file of the audio to use just for this process. This process should only be done once the interview has approval from the interviewee.

Initial Whisper Transcript

First, run the file through the UGA Libraries Whisper (currently a Demo stage). Select the options chosen in the example image below:

Of special note, "Diarization - Speakers" should be set to empty so that Whisper estimates speaker numbers instead of a guess from us. Whisper will run (probably will take at least several minutes) and generate an array of output files. The only file that you need to download is the .json.

Cleaning the JSON

A colleague has created a Python script that cleans up the JSON. While it is not public currently, you can get access to this script from a colleague in the unit. Follow these steps to run the script on your JSON files:

1. Put all JSON files in a directory. Note the path of this directory.
2. In a Terminal window, run this command: python3 /path/to/script.py /path/to/json/directory
3. This script will run quickly and create a .txt file for each .json file. The .txt file is the cleaned version of the transcript that we will work with. The .json files can be discarded.

Quick Pass in Aviary

The Oral History Unit uses the OHMS editor in Aviary to edit transcripts and related interview metadata. To get access to this Aviary database, contact the Head of Oral History. For each interview, you will need to create a new record in the Aviary database and fill in all metadata as follows:

• Title = Title (from Airtable)
• Accession Number = Accession Number (from Airtable)
• Interviewee = Interviewee (from Airtable)
• Interviewer = Interviewer (from Airtable)
• Interview Date = Interview Date (from Airtable)
• Type = "oral history"
• Collection ID = Collection ID (from Airtable)
• Collection Name = Collection Title (from Airtable)
• Embed Code = Get this code from CollectiveAccess once the item is live. Open the source of the item's public object page, search for .mp3 or .mp4 (depending on whether it's video or audio), and copy the entire source URL. You can test it by pasting it into a browser and see if it opens just a player with the file.
• Format = "oral history"
• Rights Statement = "Resources may be used under the guidelines described by the U.S. Copyright Office in Section 107, Title 17, United States Code (Fair use). Parties interested in production or commercial use of the resources should contact the Russell Library for a fee schedule."

These are the only fields required. Save the metadata. Now, if you open the Transcript in the editor (Transcript, then Edit Transcript), you can make a quick pass and make these initial changes:

• Replace the speaker placeholder fields with the actual interviewer and interviewee last names. The first time any speaker is listed, use their full name.
• Check for any glaring issues with a quick scan through the text.

Full Pass and Authentication by Oral History Unit

From here, we will pass the transcript off to the Oral History Unit, and they will do a full listen and correct all text in the transcript. Once they have completed their work, we will have an authenticated transcript that we can attach to our CollectiveAccess record and preserve along with the media (a close copy in the Digital Hub and a preservation copy written to the metadata AIP when it's created down the road).