Transcribe Oral History Recordings with AI

Before transcribing, ensure your audio file is accessible on your computer. Check that:

• The recording is clear (background noise significantly reduces accuracy)
• You have the speaker's name(s) for attribution
• You know the approximate duration

If your recording is on cassette or other physical media, you'll need to digitize it first using a USB audio interface and free software like Audacity.

Go to otter.ai and create a free account.

1. Click Import or Upload Audio/Video
2. Select your MP3/WAV file from your computer
3. Name the file with the interviewee name and date (e.g., "Smith_John_interview_2024-03-15")
4. Click Upload and wait for processing

What you should see: A processing indicator. Most transcriptions complete at about 1/3 to 1/5 of real-time (a 1-hour interview takes 12–20 minutes to process).

Troubleshooting: If accuracy is very poor, the audio quality may be too low. Check that the recording was made with a decent microphone and at a reasonable volume. Background music, crowd noise, and poor mic placement all reduce accuracy significantly.

When processing is complete, Otter shows you the transcript with speaker labels and timestamps.

1. Read through the transcript while listening to the audio at a few key points (spot-check)
2. Note any sections with poor accuracy (marked with lower confidence or clearly wrong words)
3. Download the transcript: Click Export → Text or DOCX

What you should see: A paragraph-by-paragraph transcript with speaker labels and timestamps. Accuracy is typically 80–90% for clear recordings with distinct speakers.

Open Claude at {{tool:Claude.url}} and start a new conversation. Paste the exported transcript text.

Type this message:

I have an AI-generated transcript of an oral history interview that needs cleanup. Please:
1. Fix obvious transcription errors (wrong words, garbled phrases)
2. Format speaker turns consistently (INTERVIEWER: / [INTERVIEWEE NAME]:)
3. Indicate unclear sections with [unclear] rather than guessing
4. Do NOT change the substance, vocabulary, or speaking style of the interviewee — preserve their voice exactly
5. Remove filler words (um, uh) sparingly — only when they significantly interrupt readability

Transcript: [paste your Otter transcript]

What you should see: A cleaned-up transcript with corrected errors, consistent formatting, and all unclear sections flagged rather than guessed.

Open the cleaned transcript in your word processor. Add a header with:

• Interviewee name
• Interviewer name
• Date and location of interview
• Project/collection name
• Archival restrictions notice if applicable
• "(Transcript generated with AI assistance; reviewed by [your name], [date])"

Save as a PDF and add to your digital repository or attach to the ArchivesSpace record.