video_transcript_whisper

A CLI tool that transcribes the audio from a video file to text using whisper.cpp — no cloud API needed, everything runs locally.

How it works

1. Extracts audio from the input video and converts it to 16 kHz mono WAV (the format whisper.cpp requires) via bundled ffmpeg.
2. On first run, automatically downloads and compiles whisper.cpp and fetches the chosen Whisper model (default: medium.en).
3. Runs transcription locally and prints the result to stdout, or saves it to a file.

Installation

npm install -g video_transcript_whisper

Or run directly with npx:

npx video_transcript_whisper -i video.mp4

Usage

video_transcript_whisper [options]

Options:
  -i, --input-video-path <path>   Path to the video file to transcribe (required)
  -o, --output-path <path>        Path to save the transcription (default: print to stdout)
  -v, --verbose                   Enable verbose logging (default: false)
  -h, --help                      Display help

Examples

Print transcription to stdout:

video_transcript_whisper -i lecture.mp4

Save transcription to a file:

video_transcript_whisper -i interview.mp4 -o transcript.txt

Watch download/compilation progress on first run:

video_transcript_whisper -i demo.mp4 -v

First run

The first time the tool runs it will:

1. Compile whisper.cpp v1.5.5 into ./output/whisper.cpp/ (requires a C++ compiler).
2. Download the medium.en Whisper model (~1.5 GB) into the same directory.

Subsequent runs skip both steps and go straight to transcription.

Supported video/audio formats

Any format supported by ffmpeg (MP4, MOV, MKV, AVI, MP3, WAV, …).

Requirements

• Node.js 18+
• A C++ compiler (gcc / clang) for the one-time whisper.cpp build step
• make

Development

# Install dependencies
npm install

# Type-check
npm run typecheck

# Build
npm run build

# Run from source
node dist/cli.js -i video.mp4

License

MIT