Douyin Video Transcript Extractor

Problem

Douyin videos often contain spoken content that is hard to capture as plain text. This skill turns v.douyin.com / douyin.com video links into editable transcripts for archiving, search, and further processing.

How It Works

The skill is built around speech transcription: it uses yt-dlp to obtain audio, ffmpeg for format handling, and faster-whisper to convert speech to text. The first run checks dependencies and caches paths, while later runs start faster. It supports --lang auto/zh/yue/en/ja/ko, --model tiny/base/small/medium, --timestamps, --output, and --cookies. It corrects common Whisper homophone errors using context and includes some built-in correction for Buddhist terminology. When a request returns 403, it falls back through audio streams, format lists, and Playback API formats.

Boundaries

Most public videos can be processed without extra setup; restricted or login-gated content may require a cookie file. Higher-accuracy models use more time and resources. For Cantonese, dialects, noisy audio, or technical terms, use a larger model and review the output manually. Transcripts may still contain subtle speech-to-text errors, so they should not be treated as exact quotations.