liuchong110

Soundcheck — AI skill for Claude Code

AI community

Measure, correct and verify the audio of AI video: 7 AI audio tells (loudness, true peak, sibilance, mud, take drift), deterministic ffmpeg chain, independent verification.

How to install Soundcheck

This entry records only its repository, not the path inside it, so there is no exact command to give. Open liuchong110/soundcheck and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Soundcheck does

Measure, correct and verify the audio of AI video: 7 AI audio tells (loudness, true peak, sibilance, mud, take drift), deterministic ffmpeg chain, independent verification. Agent Skill for Claude Code, Codex, Cursor and any SKILL.md host.

Alternatives in AI

Launch An AI Feature — Workflow recipe — take an AI/LLM feature from a probabilistic-aware PRD to a launch-ready model card by chaini 1.3k ★
VibeMeter — A project to measure costs for Cursor and other AI provider 379 ★
Prismor — Self-hosted runtime control plane for AI agents 334 ★

README

soundcheck

English · [简体中文](README.zh-CN.md)

**Your AI video sounds AI. Here are the 7 tells - and the fix.**

`48/100 -> 100/100`, measured on a shipped test fixture, verified with an independent meter.

![soundcheck: 64/100 to 98/100 on a 15 second clip](docs/demo-ab.gif)

https://github.com/user-attachments/assets/dfad655d-6312-4777-bcbf-6c460b0c4f09

npx skills add liuchong110/soundcheck
python scripts/soundcheck.py fix clip.mp4 --target youtube   # video in, video out
python scripts/soundcheck.py demo          # no assets needed
python scripts/verify.py                   # 20/20 self-test

The problem

AI video pipelines got good at pictures and stayed bad at sound:

TTS takes are rendered separately and glued together at different levels
the music bed never ducks under the voice
the mix is pushed into the ceiling to sound "loud" and clips
the synthetic voice keeps its 6-10 kHz hiss and 100-300 Hz boom

None of that is taste. It is measurable, and it is fixable.

What it does

$ python scripts/soundcheck.py demo

soundcheck - target youtube (streaming)
  duration 9.0s  48000 Hz  2 ch

  loudness         -6.9   target -14.0    FAIL
  true_peak         0.7   target -1.0     FAIL
  sibilance        0.72   target 0.45     FAIL
  mud              0.71   target 0.4      FAIL
  drift             2.7   target 3.0      ok
  noise_floor       n/a   target -45.0    ok    (no quiet passage)
  phase             1.0   target 0.2      ok

  SCORE 48/100 - sounds AI

$ python scripts/soundcheck.py fix demo_ai_voice.wav --target youtube
...
  loudness        -14.0   target -14.0    ok
  true_peak        -2.0   target -1.0     ok
  sibilance        0.14   target 0.45     ok
  mud              0.31   target 0.4      ok

  SCORE 100/100 - broadcast-ready
  verification: all measured tells within target