mcpField
T
transcribe

transcribe

github.com/bholmesdevCreative & media17957 installsUpdated 2w ago
Repository

About transcribe

transcribe is a skill that runs Whisper on a user-provided media file, cleans the resulting subtitles, and saves them as an SRT file beside the source. It is intended for people preparing readable subtitles while retaining the speaker’s meaning and approximate timing.

Key features

  1. 01Run Whisper on a media file
  2. 02Create an SRT beside the source
  3. 03Clean text for subtitle readability
  4. 04Preserve meaning and approximate timing
  5. 05Reveal the finished file in Finder

Use cases

  1. 01Preparing readable subtitles from media
  2. 02Cleaning Whisper-generated subtitle text

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/bholmesdev/skills --skill transcribe

After installing, type/transcribein Claude Code to confirm it loaded.

Details

From the project README

transcribe

Input

Use the media file path the user provided. If it's unclear which file to transcribe, ask the user for the local media file path and stop.

Instructions

When this skill is invoked, do the work immediately.

  1. Run Whisper on the input file and write an .srt file next to the source media file.
  2. Review the generated subtitles and make them clean and subtitle-ready rather than overly literal.
  3. Reveal the final .srt file in Finder.
  4. Report the output .srt path.

Subtitle cleanup requirements

You must optimize the subtitles for readability while preserving meaning and approximate timing.

  • Remove obvious filler words when they do not add meaning, such as um, uh, like, and similar verbal padding.
  • Remove obvious false starts, accidental repetitions, and disfluent fragments when they make captions harder to read.
  • Fix punctuation, capitalization, and obvious formatting problems.
  • Preserve intentional wording, important emphasis, and technical terms.
  • Do not rewrite the speaker's meaning or turn the transcript into a summary.
  • Prefer concise, natural subtitle text over raw verbatim speech.

Command

Run Whisper with the provided media file:

INPUT="<media file path>"
whisper "$INPUT" --output_format srt --output_dir "$(dirname "$INPUT")"

The generated subtitle path is:

SRT_PATH="$(dirname "$INPUT")/$(basename "${INPUT%.*}").srt"

After cleaning the subtitle text, reveal the file in Finder:

open -R "$SRT_PATH"

Output

Report the final .srt path:

$SRT_PATH

If Whisper fails, report the error briefly and stop.

Comments

Markdown supported