metaphrasislisted

# metaphrasis A compact encoding of audio into things a text-and-image model can actually read: energy becomes a block-character sparkline, tempo and key and loudness become numbers, timbre becomes words, and structure becomes a strip chart. **This is measurement, not hearing.** It reliably catches loud, busy, restless, badly-matched and wrongly-structured material. It cannot tell you a track is beautiful, and it cannot hear an artifact or an AI shimmer. Say so rather than letting a grade stand in for a judgement. ## Requirements ffmpeg and ffprobe on PATH; Python with librosa, numpy, scipy, soundfile. ## Commands ```bash B=~/.claude/skills/metaphrasis/scripts/metaphrasis.py python "$B" scan <dir> # one line per track — start here python "$B" read <file> # strip chart + best cut points python "$B" pick <dir> --for voiceover # shortlist: voiceover|opener|closer|montage|long python "$B" matrix <dir> # which tracks join cleanly with which python "$B" plan <spec.json> # resolve a splice without rendering python "$B" splice <spec.json> # cut and sew python "$B" narration <video> # transcript + pacing + is the bed burying the voice ``` Analysis is cached in a `.metaphrasis/` folder beside the audio, keyed on file size and mtime, so repeat commands are instant. First run on a large library is slow — roughly a few seconds per track. ## Reading a scan line ``` the-verdict-03 2:59 -