douyin-video-summary

Douyin Video Summary

Summarize Douyin videos: extract audio → transcribe locally → AI summary.

Prerequisites

Install these tools (macOS example):

brew install whisper-cpp ffmpeg
# Download whisper.cpp GGML model (small recommended for speed/quality balance)
curl -L -o models/ggml-small.bin "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.bin"

Workflow

When a Douyin link is received:

Step 1: Extract Video ID

Parse the Douyin URL to get the video ID. Douyin share links come in two formats:

• Short link: https://v.douyin.com/xxxxx/ → follow redirect to get video ID
• Direct link: https://www.douyin.com/video/7604713801732365681

# Follow redirect to get final URL, extract numeric video ID
curl -sL -o /dev/null -w '%{url_effective}' 'https://v.douyin.com/xxxxx/' | grep -oE '[0-9]{15,}'

Step 2: Get Audio via Browser

Douyin blocks direct downloads (yt-dlp, aria2c all get 403). Use the browser to intercept the audio URL:

1. Open the Douyin video page in the browser
2. Inject JS to intercept network requests before navigation:

window.__audioUrls = [];
const origOpen = XMLHttpRequest.prototype.open;
XMLHttpRequest.prototype.open = function(method, url) {
  if (url && (url.includes('.mp3') || url.includes('.m4a') || url.includes('mime_type=audio'))) {
    window.__audioUrls.push(url);
  }
  return origOpen.apply(this, arguments);
};

1. Navigate to the video page, click play to trigger audio loading
2. Retrieve intercepted URLs: window.__audioUrls
3. Download with curl (Referer header required):

curl -H "Referer: https://www.douyin.com/" -o audio.mp4 "<audio_url>"

Important: aria2c will 403 on Douyin CDN URLs. Always use curl with the Referer header.

Step 3: Convert to WAV

ffmpeg -i audio.mp4 -ar 16000 -ac 1 -c:a pcm_s16le audio.wav

Step 4: Transcribe with whisper.cpp

whisper-cli -m /path/to/ggml-small.bin -l zh -f audio.wav -otxt -of output