audio-edit

原文涵盖的主题

• Audio editing (ffmpeg)
• Normalize loudness
• Even out inconsistent levels (loud speaker vs quiet audience)
• Denoise (hiss / hum)
• Add background music
• Gotchas

Loudness normalize, denoise, and background-music mixing. These are plain ffmpeg one-liners — no CLI command wraps them. Every recipe stream-copies the video (-c:v copy, fast + lossless) and re-encodes only the audio to aac. Pick the operation: Audio too quiet / loud / inconsistent across clips → normalize. Background hiss / hum / fan noise → denoise. Lay a music bed under narration → add music (use --duck-style sidechain so music drops under speech). Cutting um/uh hesitations → use the filler-removal skill, not this one. -16 LUFS / -1.5 dBTP is a good general/web target. The single pass above is fine for most clips. For precise targets (or batch consistency), do two-pass: run once with loudnorm=...:printformat=json, read the measured values, then pass them back as measuredI/measuredTP/measuredLRA/measuredthresh on a second run. The most common "fix the audio" complaint on talks/panels/interviews isn't noise — it's level swings: the presenter is loud and near the mic, audience questions are faint, peaks nearly clip. Compress to tighten the dynamic range, then normalize: acompressor pulls the loud parts down; makeup lifts everything so the quiet parts come up — net result is consist

这里只展示节选，请以完整原文和当前要求为准。