What problem does it solve? Audio already placed in a HyperFrames composition often sounds wrong together: a music bed fights the voiceover, tracks are uneven, or a voice sounds boomy or harsh. This Skill mixes those placed tracks by writing effect chains, volume and effect automation envelopes, ducking, and voiceover carves directly onto the audio elements as HTML attributes, so preview and render produce the same processed mix. ## Core Features & Use Cases - Voiceover carve: Analyzes the voice tracks, finds the bands speech occupies, and writes peaking-filter dips plus a level-match gain envelope onto the music bed via scripts/carve.mjs, so the bed keeps its character while the voice stays intelligible. - Effect chains and automation: Applies EQ, compressor, limiter, gate, saturation, delay, reverb, chorus, phaser, and bitcrush through data-fx-chain, with breakpoint automation lanes on volume or effect parameters via data-automation, including submix buses with <hf-audio-group>. - Diagnosis of unheard audio: Provides measurement recipes (FFmpeg band energy, noise floor, loudness, level-over-time) in references/diagnosis.md to identify defects by comparing against the clean original or the pauses rather than guessing. - Use Case: Given a composition with narration and a background music track, run the carve script to automatically detect the bed and voices, then write the carve attributes so the music ducks only in the speech bands. ## Quick Start Ask the AI to carve the music bed under the voiceover in your HyperFrames composition by running the carve script against your index.html file.
