gettr-transcribe-summarize

⚠Review·Scanned 2/18/2026

This skill downloads audio from GETTR posts, transcribes it locally with mlx_whisper and ffmpeg, and writes outputs to ./out/gettr-transcribe-summarize/<slug>/. It runs shell tools (bash, ffmpeg, mlx_whisper), performs network fetches to https://gettr.com or user-supplied URLs, and references huggingface-cli login.

from clawhub.ai·v6da3536·22.2 KB·0 installs
Scanned from 1.0.4 at 6da3536 · Transparency log ↗
$ vett add clawhub.ai/kevin37li/gettr-transcribe-summarizeReview findings below

Gettr Transcript & Summary

Download audio from a GETTR post, transcribe it locally with MLX Whisper on Apple Silicon, and produce a clean summary or timestamped outline.

What it does

  • Extracts audio from a GETTR post (via og:video → 16kHz mono WAV)
  • Transcribes locally with MLX Whisper (no API keys required)
  • Outputs VTT with timestamps for precise outline generation
  • Summarizes into bullets or a timestamped outline

Quick start

# For /post/ URLs: extract the og:video URL via script
python3 scripts/extract_gettr_og_video.py "<GETTR_POST_URL>"

# For /streaming/ URLs: use browser automation directly (see SKILL.md Step 1)
# The extraction script is unreliable for streaming URLs

# Run download + transcription (slug is the last path segment of the URL)
bash scripts/run_pipeline.sh "<VIDEO_URL>" "<SLUG>"

Outputs to ./out/gettr-transcribe-summarize/<slug>/:

  • audio.wav – extracted audio
  • audio.vtt – timestamped transcript

Prerequisites

  • mlx_whisper (pip install mlx-whisper)
  • ffmpeg (brew install ffmpeg)

Features

  • Auto-detects non-video posts (image/text) with helpful error messages
  • Retries network requests with exponential backoff
  • Transcribes in original language (auto-detected)
  • Prevents hallucination propagation with --condition_on_previous_text False