<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>视频逆向 on Lynx 的技术博客</title><link>https://blog.lynxflow.co/tags/%E8%A7%86%E9%A2%91%E9%80%86%E5%90%91/</link><description>Recent content in 视频逆向 on Lynx 的技术博客</description><generator>Hugo -- gohugo.io</generator><language>zh-cn</language><lastBuildDate>Fri, 25 Sep 2026 20:45:00 +0800</lastBuildDate><atom:link href="https://blog.lynxflow.co/tags/%E8%A7%86%E9%A2%91%E9%80%86%E5%90%91/index.xml" rel="self" type="application/rss+xml"/><item><title>把头部电影解说拆到声学层:649支样本逆向出的结构配方</title><link>https://blog.lynxflow.co/posts/movie-explainer-structure-recipe-649-samples/</link><pubDate>Fri, 25 Sep 2026 20:45:00 +0800</pubDate><guid>https://blog.lynxflow.co/posts/movie-explainer-structure-recipe-649-samples/</guid><description>&lt;img src="https://blog.lynxflow.co/images/movie-explainer-structure-recipe-649-samples-cover.webp" alt="Featured image of post 把头部电影解说拆到声学层:649支样本逆向出的结构配方" /&gt; 结论先讲:我把越哥说电影、木鱼水心等 7 个头部电影解说频道共 652 支视频做了全量逆向——元数据全收,声学层跑了 649 支。配音占比 26% [23-30%],配音/原声切换周期 38 秒 [33-42],七个频道的置信区间全部交叠。 标题策略分「情绪流」和「片名流」两条路,但底层结构配方,全行业是同一套。</description></item></channel></rss>