# Verbatim

Verbatim is a self-hosted media utility that generates English closed captions by transcribing the English dub audio track of your own media with faster-whisper on a GPU. It is open source and self-hosted, installable on Unraid.

Canonical page: https://ourlocalhost.com/app/3445/verbatim

## Description

Verbatim is a self-hosted media utility that generates English closed captions by transcribing the English dub audio track of your own media with faster-whisper on a GPU. It targets users who need accurate captions for dubbed anime or other releases where existing English subtitle files are translations of the original Japanese audio instead of matching the dub. It runs locally in one container, helps select the dub audio stream, and can optionally use a Gemini API key to repair misheard words, disabled by default.

## Platform availability

Available on 1 of 10 platforms: Unraid.

## Health score

39/100 (Low), based on 3 of 4 factors.
- Maintained: 65 (last commit 103 days ago)
- Popular: 0 (0 GitHub stars)
- Easy to install: 35 (1 install platform, install notes, Docker image)

## Links

- GitHub: https://github.com/wastedpadre/verbatim
- Unraid: https://ca.unraid.net/apps/verbatim-0olbt2w0vghlj0?q=Verbatim

## Common questions

### How does Verbatim choose the audio track for dual-audio releases?

It scores available streams using language tags, titles, and channel count, then shows which stream it chose.

### How does Verbatim improve character name accuracy?

It extracts correctly spelled names, places, and terms from the English sub track, feeds that vocabulary to the decoder, and fuzzy-corrects near-misses afterward.

### How does Verbatim reduce Whisper hallucinations during music or silent scenes?

It disables condition_on_previous_text, uses VAD to gate non-speech, and runs a cleanup pass that drops known artifacts and collapses repetition loops.

### How are the generated caption lines formatted?

Cues are rebuilt from word-level timings to keep them to 42 characters per line, two lines maximum, 20 characters per second, no overlaps, and clause-boundary breaks.

### Can I fix a repeated misspelled name without editing every cue?

Yes. The caption editor uses find-and-replace with glossary chips, so you can replace an incorrect spelling across the job in one click.

---
Source: Our Local Host (https://ourlocalhost.com/app/3445/verbatim), the self-hosted and open-source app directory.
