bg-music

Background music makes a podcast clip feel like a TikTok, not a recording. This skill picks a random mp3 from the local ./songs/ library, loops it to match the clip's length, and mixes it under the speaker with a hard volume attenuation. No ducking sidechain — the bed is just quiet enough to live underneath broadcast-leveled speech.

Invoke

.claude/skills/bg-music/bg-music.sh <input> <out> [category=ALL SONGS] [volume=0.17] [fade=0.6]

• input — finished, loudnorm'd mp4
• out — output mp4 with music mixed in
• category — subfolder of ./songs/ to pick from (e.g. Energetic, Story, Inspiring). Defaults to ALL SONGS, which recurses across every mood folder under ./songs/.
• volume — linear gain for the bg track, default 0.17. Lower for more speech-forward clips.
• fade — seconds of fade-in so the bed ramps up instead of blasting at t=0 (it would otherwise be exposed under the speech-free title card). Default 0.6.

How

• Picks a random .mp3/.MP3/.wav from $SHORTS_ROOT/songs/<category>/.
• Loops it with -stream_loop -1 so any track length covers the clip.
• amix with duration=first so output matches the video duration exactly.
• Bg volume hard-set; output limited with alimiter=limit=0.97 to avoid clip.
• Video is stream-copied.

Caveats

• Re-encodes audio only.
• Idempotent via <out>.bgmeta (input mtime + category + volume + fade + chosen track basename).
• The chosen track is logged to stderr so you can reproduce / blacklist.