chad-nethery

Karaoke Maker — Development skill for Claude Code

Development community

Turn any song into a karaoke track: vocal removal, real lyrics, word-level highlight video.

How to install Karaoke Maker

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

What Karaoke Maker does

Turn any song into a karaoke track: vocal removal, real lyrics, word-level highlight video. Local, GPU-accelerated. Claude Code skill + CLI.

Alternatives in Development

Codex Claude Academic Skills — 本仓库包含三个面向学术科研人员的Skills，覆盖从文献阅读、论文写作到科学计算的完整研究工作流 3.2k ★
Working Memory — This is your working memory 192 ★
Jarvis Assistant Vocal — Assistant vocal local en francais : Claude ou Ollama (offline), domotique Hue, OBS, agenda, navigateur, appels 189 ★

README

karaoke-maker

Turn any song into a karaoke track, locally, with one command.

python scripts/karaoke.py "https://youtu.be/..."        # or a local mp3/flac/wav/m4a/mp4

You get a folder with:

karaoke.mp4 – 1080p video: title card, two-line lyrics with a word-by-word highlight sweep, countdown dots before long instrumental gaps, progress bar, blurred cover-art background
instrumental.mp3 – the song with the vocals removed
vocals.wav – the isolated vocal
lyrics.lrc – word-synced lyrics for karaoke apps and music players
lyrics.txt – the lyrics that were used, so you can correct them and re-run

Built as a [Claude Code](https://claude.com/claude-code) skill (drop it in `~/.claude/skills/` and say *"make a karaoke version of this song"*), but the script is a plain CLI and works on its own.

How it works

Step	Tool	Time (RTX 5060 Ti, 4-min song)
Fetch	yt-dlp or your file → 44.1 kHz WAV	seconds
Separate	Demucs `htdemucs` → vocals + instrumental	5 s
Lyrics	LRCLIB lookup (free, keyless) → fallback: Whisper transcript	1 s
Timing	stable-ts Whisper `large-v3-turbo` on the isolated vocal, chunked along its own silences, then the real lyrics are matched word-by-word onto that timing	10 s
Render	ffmpeg + libass (`\kf` karaoke sweep), x264	60-90 s

Everything runs on your machine. Nothing is uploaded anywhere; the only network calls are the optional download of the song and a lyrics lookup by artist/title.

The hybrid lyric step is what makes the result sing-able: transcription alone mishears words and breaks lines like captions, and database lyrics alone have no timing. Matching one onto the other gives exact text with the songwriter's line breaks and Whisper's word timing. When the database has nothing (or has the studio version of a live cut, detected by du