菜单
Post-processing for generated music. Default path is FFmpeg loudness normalization (fast, CPU-only). Escape hatches: AI vocal denoise for artifacts, stem separation for surgical edits. Both escape-hatch paths reuse tooling from the claude-video skill.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/AgriciDaniel/claude-music --skill claude-music-enhance命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | claude-music-enhance |
| description | Post-processing for generated music. Default path is FFmpeg loudness normalization (fast, CPU-only). Escape hatches: AI vocal denoise for artifacts, stem separation for surgical edits. Both escape-hatch paths reuse tooling from the claude-video skill. |
| when_to_use | Use when the user says enhance, normalize, denoise, master, polish, loudness, stem separation, split vocals, or wants to clean AI artifacts. |
| allowed-tools | ["Bash","Read"] |
Loudness-normalize to the platform target in one ffmpeg call. This is the right answer 90% of the time — the generated audio is already clean; it just needs the level right.
# Two-pass loudnorm to -14 LUFS (Spotify/YouTube/Apple/TikTok all target this).
bash ~/.claude/skills/claude-music/scripts/music_export.sh spotify input.flac output.flac
See references/post-processing.md for the full platform × LUFS table and
the measure-then-apply two-pass recipe if you need manual control.
Use only when you hear artifacts in vocals (hissing, AI mush, robot tones). FFmpeg normalization won't help; you need DeepFilterNet3.
source ~/.video-skill/bin/activate
python3 ~/.claude/skills/claude-video/scripts/audio_enhance.py denoise \
input.flac --output output_clean.flac
Requires claude-video skill + DeepFilterNet3 weights. <1 GB VRAM.
Use only when you need to fix vocals/drums/bass/other separately, then remix. Not needed for normal enhancement.
source ~/.video-skill/bin/activate
python3 ~/.claude/skills/claude-video/scripts/audio_enhance.py separate \
input.flac --model htdemucs_ft --output-dir ./stems/
Produces vocals.wav, drums.wav, bass.wav, other.wav. ~7 GB VRAM.
Is the generated audio too quiet or too loud?
→ use the default path (loudnorm)
Do you hear specific vocal artifacts?
→ escape hatch 1 (denoise) THEN default path (loudnorm)
Do you need to fix one stem (e.g., reduce drums, retune vocal) before final mix?
→ escape hatch 2 (separate) → edit stem → remix → default path (loudnorm)
For platform-specific export (codec/bitrate/sample-rate), use
claude-music-export — it wraps music_export.sh with the right flags per
target (Spotify FLAC, YouTube MP3 320k, TikTok M4A 256k, etc.).
scripts/music_export.sh — default loudnorm + format conversion path.~/.claude/skills/claude-video/scripts/audio_enhance.py — AI denoise and
stem separation (external skill; must be installed separately).