video-stt_skill

Overview

This skill extracts audio from video URLs and produces speech-to-text transcripts using local Whisper models or cloud STT APIs. It supports downloading videos, extracting audio, and saving output as plain text, SRT, VTT, or JSON with timestamps. It is designed for YouTube videos, podcasts, lectures, and any online video with speakable audio.

How this skill works

Given a video URL the tool downloads the media (yt-dlp), extracts audio (ffmpeg), and runs a speech-to-text engine either locally (Whisper models) or via cloud APIs (OpenAI, Azure, Google). It can run different local model sizes for speed vs. quality and supports multiple output formats including subtitle files and timestamped JSON. Environment variables configure API keys and the script can be invoked from the command line or called from Python.

When to use it

• You need a quick transcript of a YouTube video or podcast episode.
• You want subtitle files (.srt/.vtt) for a video you own or are allowed to transcribe.
• You prefer using a local Whisper model to avoid sending audio to the cloud.
• You need timestamped JSON for downstream processing or indexing.
• You want to batch-archive and transcribe videos for an archive or backup.

Best practices

• Choose model size based on tradeoff: tiny/base for speed, medium/large for better accuracy.
• Provide high-quality audio or use noise reduction to improve transcription accuracy.
• Use cloud APIs when you need faster processing or higher-accuracy commercial models.
• Set API keys as environment variables (e.g., OPENAI_API_KEY) rather than embedding them in scripts.
• Save transcripts in multiple formats if you need both readable text and timed subtitles.

Example use cases

• Transcribe a YouTube lecture to searchable text for research notes.
• Generate SRT subtitles for a podcast episode before publishing.
• Extract and index speech from archived videos for a media library.
• Run local Whisper on sensitive videos to keep audio processing on-premise.
• Convert interview recordings into timestamped JSON for QA and analysis.