Show HN: A CLI tool to transcribe and clean YouTube videos with Whisper and LLMs

A command-line tool to turn any YouTube video into a clean, readable text transcript. It uses OpenAI's Whisper for transcription and the LLM of your choice to automatically clean and reformat the output.

1. Automatic download from YouTube
2. Fast and accurate transcription using OpenAI's Whisper models
3. LLM-powered cleaning that removes filler words, fixes grammar, and organizes content into readable paragraphs
4. Multiple output formats (TXT, SRT, VTT) for any use case
5. Flexible LLM support - use Gemini, ChatGPT, Claude or any other (local) LLM for cleaning

Option 1: Clone and run

Option 2: Install as package

Requirements:

• Python 3.7+
• FFmpeg (for audio processing)
• LLM API key (for cleaning, optional but recommended)

Basic transcription with cleaning:

Create clean subtitles:

High-quality lecture transcription:

Raw transcript (no cleaning):

• --format, -f: Output format (txt, srt, vtt)
• --model, -m: Whisper model (tiny, base, small, medium, large, turbo)
• --llm-model: LLM for cleaning (gemini-2.0-flash-exp, gpt-4o-mini, etc.)
• --cleaning-style: presentation, conversation, or lecture
• --save-raw: Keep both raw and cleaned versions
• --no-clean: Skip AI cleaning

Popular models:

• gemini-2.0-flash-exp (recommended - fast, cheap)
• gpt-4o-mini (OpenAI, fast)
• claude-3-5-sonnet-20241022 (Anthropic, high quality)

Uses Simon Willison's excellent llm package for provider flexibility.

What it does:

• Removes filler words (um, uh, so, like, you know, etc.)
• Fixes grammar and punctuation errors
• Organizes content into logical paragraphs
• Maintains original meaning and context

Cleaning styles:

• presentation: Professional tone, organized paragraphs
• conversation: Natural flow, minimal cleanup
• lecture: Educational format, clear sections for notes

Note: SRT/VTT preserve timing while cleaning text content.