ghosted

De-identify transcript batches with one local app:

• ghost() opens a local Shiny app for .vtt, .docx, and .txt transcripts.
• The app asks for input/output directories, known interviewer names, known participant names, and other terms to redact.
• It scans the input directory, reviews likely names detected with local rule-based matching, and writes redacted transcripts to the output directory.
• It also includes options for output format, filename suffix, redaction token, blank lines between turns, console reports, common-name redaction, and the completion notice.

The app runs locally on your computer and does not use AI or remote services.

Installation

You can install the development version of ghosted like so:

# via remotes
remotes::install_github("abiraahmi/ghosted")

# or with pak
# pak::pak("abiraahmi/ghosted")

Use

library(ghosted)

ghost()

In the app:

1. Enter the input directory containing .vtt, .docx, or .txt transcripts.
2. Enter an output directory, or leave it blank to write outputs next to the input files.
3. Choose processing options, such as output format and redaction token.
4. List known interviewer and participant names, plus any other terms to redact.
5. Click Scan directory.
6. Review likely names and mark them as interviewer, participant, or other.
7. Click Redact transcripts.

When you run ghost() again in the same R session, the app preloads the directories, options, names, and likely-name selections from the previous run.

Notes

• Leading speaker tokens at line start, such as Name: or <v Name>, are normalized to Interviewer or Participant.
• Names elsewhere in the transcript body are replaced with the selected redaction token.
• For DOCX/TXT to VTT output, cues are written without timestamps.
• Install officer for DOCX reads and writes.