What problem does it solve? Audio tracks that sound fine alone often fight each other in a mix — a music bed drowns out a voiceover, a voice sounds boomy or harsh, and fixing it by ear is impossible for an agent that cannot listen. This Skill provides deterministic mixing for HyperFrames compositions: effect chains, volume automation, ducking, and a voiceover carve that cuts only the bands the voice occupies. ## Core Features & Use Cases - Voiceover carve: Analyzes voice tracks and writes peaking-filter dips plus a level-match envelope onto the music bed via data-fx-carve, data-fx-chain, and data-automation, so the voice stays intelligible without flattening the music. - Effect chains and automation: Ships gain, EQ, compressor, limiter, gate, saturate, delay, reverb, chorus, phaser, and bitcrush effects with per-parameter automation lanes, plus an <hf-audio-group> submix bus for treating several tracks at once. - Diagnosis without listening: Provides ffmpeg-based measurement recipes (band energy, noise floor, loudness via ebur128) to identify defects by comparing a file against itself rather than guessing. - Use Case: You have a narrated video composition where the background music fights the narration. Run node scripts/carve.mjs --comp index.html to automatically detect the bed and voices, carve the bed at strength 0.8, and write the mix attributes into the HTML. ## Quick Start Ask the AI to carve the music bed under the narration in your HyperFrames composition by running the carve script against your index.html file.
