使用时注意几个坑：依赖版本、synchronized 锁、ThreadLocal 串数据，提前规避。
visual_lambda.pdf.
。体育直播是该领域的重要参考
functions with explicit generic annotations. For old-style generics,
Silero is a tiny, open-source model (around 2MB) that can quickly determine whether a short chunk of audio contains speech. Turn-taking is a much harder problem than speech detection, but VAD is still a useful primitive, especially for deciding whether audio should be forwarded to more expensive downstream systems.