An all-in-one drum-charting tool for Paradiddle (.rlrr), with Clone Hero (.chart) support.
STANDALONE .EXE VERSION: https://limewire.com/d/UV9Zm#DHqxKgEtmn
REPO FOR .EXE VERSION: https://github.com/sherifican/ParaKit---Releases
Take a song, isolate the drums, turn them into a playable drum chart, refine it in a visual MIDI editor, practice it with falling notes, and export it — all in one app.
ParaKit always has been and always will be free of charge, and will never host ads. This repository now makes the full v4.x source code open under the GPLv3 license, so anyone can run it from source, learn from it, fix it, or build their own version.
Version in this release:
4.4.57.99-10• Runtime: Python 3.12 (required)
ParaKit is mid–transition between two generations, and it matters for what you can do here:
| ParaKit v4.x (this release) | ParaKit v5 (future) | |
|---|---|---|
| UI framework | Tkinter / TTK | PySide6 / Qt |
| Status | The complete, stable, shipping app | Early rebuild, barely started — not in this release |
| Themeable with UI Studio? | No | Yes (UI Studio is built for v5) |
UI Studio — the visual UI/layout designer — is built for the v5 (PySide6) rebuild and is not compatible with this v4.x (Tkinter) app. It is not included in this release because it can't run without the v5 code, which isn't ready yet. UI Studio and v5 will arrive in a later follow-up. You cannot use UI Studio to re-theme or edit this v4.x app — but once it ships, you'll be able to use it to design for v5 or build your own custom ParaKit from source.
See docs/ROADMAP.md for the v5 / UI Studio / GPU-build plan.
The falling-note Practice experience has been rebuilt from scratch as a fast, self-contained web app — a big step up from the in-app mini-game and the practice_v2/ alpha. Notes fall down 8 lanes in time with the music while you play along on a USB MIDI drum kit or your keyboard.
- Rock-solid timing — a sample-accurate Web Audio clock, no drift.
- One file, zero setup — open it in any modern browser; includes a built-in demo plus a synth that makes any chart audible even with no audio file.
- Latency calibration, mid-song mix/stem switching, touch support, a results screen with a timing histogram, and full keyboard + MIDI play.
▶ Try it: download Practice Window v2 - Web Edition/parakit-practice.html and open it in any modern browser (Chrome or Edge for USB-MIDI).
This web edition is the de-facto final version of Practice Mode v2 for now. An updated native (
.py) version folded back into the app is planned for a later release — but you don't need to wait: the web edition is complete and ready to play today.
| Folder / file | What it is |
|---|---|
ParaKit v4.0.py |
The full ParaKit v4.x app (single file) |
Run ParaKit v4.0.bat |
Windows double-click launcher |
requirements.txt |
Python dependencies for the main app |
extractor/ |
RLRR Extractor — converts .rlrr charts back into .mid MIDI (readme) |
practice_v2/ |
Practice Window v2 — standalone falling-note practice mini-game (alpha, readme) |
Practice Window v2 - Web Edition/ |
Practice Mode v2 — Web Edition — the new self-contained browser rebuild (recommended; see the section above) |
docs/ |
Building from source, troubleshooting, roadmap |
LICENSE |
GNU GPL v3 |
Practice v1 vs v2: the stable Practice mode is v1, built into the main app.
practice_v2/is an in-development alpha — included so you can build on it. See its readme.
- MIDI editor — visual note placement and refinement
- Audio → MIDI detection — automatic drum transcription with three engines: Spectral (traditional), ML / ONNX (neural net), and Hybrid (combined), plus genre presets (Pop / Rock / Metal / Funk)
- Stem splitter — isolate a drums-only track from any song (Demucs)
- MusicXML → MIDI — convert sheet music into a chart
- YouTube → FLAC — download lossless audio to chart from
- Asset management — metadata, album art, preview clips
- Preview & Practice — falling notes synced to the audio, keyboard or USB MIDI kit
- Song Tester — verify sync before export
- Export — Paradiddle (
.rlrr) and Clone Hero (.chart)
ParaKit targets Python 3.12 specifically. Get it from
python.org (check "Add to PATH" / use the py launcher).
py -3.12 -m pip install -r requirements.txt
(See requirements.txt — note the Stem Splitter pulls in demucs + torch, a large
~2–3 GB download you can skip if you won't split stems.)
ParaKit shells out to several tools that are not Python packages:
- FFmpeg (
ffmpeg/ffplay/ffprobe) — audio conversion /pydub - yt-dlp (+ deno, its JS signature runtime) — YouTube → FLAC downloads
- ADB (+
AdbWinApi.dll,AdbWinUsbApi.dll) — "push to Quest" / device transfer
These are distributed separately as the Requirements.zip bundle (≈174 MB — too large to
include in the Git repo). Download it here → Requirements.zip (LimeWire)
I have also uploaded the Jarredou model alongside the Reqs on the LimeWire page since the original repo for it is down. There are mirrors on HuggingFace but I don't wanna force people do dig thru the giant repo + read thru the 600+ page report.
Extract it, then place the files next to ParaKit v4.0.py, or keep them in the included
Requirements\ subfolder beside it. They're kept out of the Git tree on purpose — large
binaries with their own licenses, well over GitHub's per-file size limit. Leave yt-dlp's
auto-update on so it stays current with YouTube changes.
py -3.12 "ParaKit v4.0.py"
…or just double-click Run ParaKit v4.0.bat.
- Stem Splitter — isolate drums from the backing track
- Audio → MIDI — transcribe the drums-only stem to MIDI
- MIDI editor — clean up and refine the chart
- Preview / Practice — watch it as falling notes
- Song Tester — confirm sync
- Assets — set metadata, album art, preview clip
- Export — Paradiddle
.rlrror Clone Hero.chart
The stem splitter uses GPU acceleration on NVIDIA GTX 10-series through RTX 40-series. On the newer RTX 50-series (Blackwell — 5070/5080/5090) the stock PyTorch build does not include support for those GPUs, so splitting falls back to CPU (it still works, just slower). AMD / Intel GPUs are CPU-only as well (Demucs needs CUDA).
A working GPU fix exists (CUDA 12.8 / cu128 PyTorch + a save-path tweak) — see
docs/TROUBLESHOOTING.md. We're also preparing a separate,
creator-verified RTX 50-series build with GPU acceleration configured out of the box,
to ship as a follow-up. The CPU fallback always stays in the code regardless — the point is
that the feature works on every machine, even if a bit slower.
Because this is the full source, you can add features, remove them, rearrange the UI, or
make your own personal ParaKit. See docs/BUILDING.md for running from
source and compiling a standalone .exe.
ParaKit is released under the GNU General Public License v3.0 (see LICENSE).
In short: you're free to use, study, modify, and share it — but if you distribute a modified
version, you must also make your source available under the same license. This keeps ParaKit
and everything built from it free and open.
Bundled third-party tools (FFmpeg, yt-dlp, ADB) and Python dependencies each carry their own licenses.
ParaKit — free forever, no ads.















