Transcribe

AI transcription skill for Claude Code - Transform audio/video recordings into transcripts with speaker diarization, AI-powered summaries, and visual infographics.

This skill provides a complete pipeline for processing recordings:

- Transcription - Convert audio/video to VTT format with speaker identification (OpenAI Whisper)
- Summarization - Generate structured markdown summaries (OpenAI GPT-5.1)
- Infographics - Create visual summaries from text (Google Gemini)
- All-in-one - Process video → transcript → summary → infographic in one command

See skills/transcribe/SKILL.md for complete usage guide.

Use in Claude Code

This is a Claude Code skill. Install it from the marketplace:

``bash
/plugin marketplace add krasnoperov/claude-plugins
/plugin install transcribe@krasnoperov-plugins
`

Once installed, use the /transcribe skill in your conversations:

`
/transcribe transcribe meeting.mp4 to VTT with speaker diarization
/transcribe summarize this transcript into key points
/transcribe create an infographic from this summary
`

Command Line Usage

You can also use this package directly via npx:

`bash
export OPENAI_API_KEY="your-openai-key"
export GOOGLE_AI_STUDIO_KEY="your-google-key"

Transcribe audio/video

npx -y @krasnoperov/transcribe@latest transcribe meeting.mp4 -o transcript.vtt

Generate summary

npx -y @krasnoperov/transcribe@latest summarize transcript.vtt -o summary.md

Create infographic

npx -y @krasnoperov/transcribe@latest infographic summary.md -o visual.png

All-in-one pipeline

npx -y @krasnoperov/transcribe@latest process recording.mp4 --output-dir ./output
`

Get your API keys:
- OpenAI
- Google AI Studio

Core Operations

`
transcribe Audio/Video → VTT transcript with speakers
summarize Text/VTT → Markdown summary
infographic Text → Visual infographic image
process All-in-one: video → transcript → summary → infographic
`

These operations can be used individually or chained together.

Examples

See skills/transcribe/examples/ directory:

1. 01-basic-workflow.sh - Step-by-step transcription pipeline
2. 02-all-in-one.sh - Single command processing

$3

`bash
npx -y @krasnoperov/transcribe@latest transcribe podcast.mp3 \
--language es \
--model gpt-4o-transcribe-diarize \
-o podcast.vtt
`

$3

`bash
npx -y @krasnoperov/transcribe@latest transcribe meeting.mp4 \
--model gemini-3 \
-o meeting.vtt
`

Gemini 3 offers excellent transcription with built-in speaker diarization and can handle very long audio files (up to ~8 hours).

Output (VTT with speaker tags):
`
WEBVTT

00:00:00.000 --> 00:00:02.450
Welcome to the podcast...

00:00:02.850 --> 00:00:08.200
Thanks for having me...
`

$3

`bash
npx -y @krasnoperov/transcribe@latest summarize transcript.vtt \
--prompt "Focus on action items and decisions" \
-o summary.md
`

$3

`bash
npx -y @krasnoperov/transcribe@latest infographic summary.md \
--style "modern minimal corporate" \
-o infographic.png
`

Options

$3


`
--model Transcription model:
OpenAI: gpt-4o-transcribe-diarize (default), gpt-4o-transcribe, whisper-1
Google: gemini-3
--language Language code (en, es, ru, de, etc.)
-o, --output Output VTT file
`

$3

`
--prompt Custom summarization instructions
-o, --output Output markdown file
`

$3

`
--style Style instructions for visual
--reference Reference image for style
-o, --output Output image file
`

$3

`
--output-dir Output directory for all files
--language Language for transcription
--model Transcription model
--style Style for infographic
`

Requirements

- Node.js >= 18.0.0
- ffmpeg (for audio extraction)

`bash


macOS

brew install ffmpeg

Ubuntu/Debian

sudo apt install ffmpeg
`

Development

`bash
npm run build # Build TypeScript
npm run typecheck # Type checking
npm run test # Run tests
npm run dev # Dev mode with type stripping
``

License

MIT License - Copyright (c) 2025 Aleksei Krasnoperov

See LICENSE file for details.