shurshun/count-in

What it does

Count-In analyzes an audio file to find its tempo and first downbeat, then places a drum count-in so the final click lands exactly one beat before the song's "one". The same analysis also supports making a backing track by attenuating or removing guitar and vocals, with separate level controls for each.

It targets musicians and practice/backing-track workflows rather than general audio editing. Count-in length is one or two bars in 4/4, 3/4 or 6/8, with even, build-up, or user-supplied MIDI patterns. Guitar/vocal separation runs through a neural network in one pass. Input formats are mp3, FLAC, ALAC and WAV; output can be MP3 320 kbps, FLAC, ALAC or WAV, and tags plus cover art are preserved. Processing is offline; the network is used only for updates.

The interface is in English and Russian and follows the system language, with a manual switch in Settings.

Quick start

Download the current build from Releases. Windows 10/11 x64 uses Count-In-windows-x64.zip; macOS 12+ on Apple Silicon uses Count-In-macos-arm64.zip. The builds are unsigned, so each system warns on first launch: on Windows, run Count-In.exe and choose "More info" → "Run anyway" if SmartScreen appears; on macOS, drag Count-In.app to Applications and use System Settings → Privacy & Security → "Open Anyway".

Then work on one track at a time:

1. Drop a track onto the window or open it (⌘O); analysis runs in the background.
2. Check the orange "1" marker on the waveform. Drag it to the correct beat if needed; it snaps to detected beats, and holding ⌥/Alt places it freely. The wheel zooms, dragging pans.
3. Choose pattern, count-in length and meter. Blue marks show where clicks will sound.
4. For a backing track, lower the guitar and vocal sliders, or use the speaker button beside each to remove that part in one click.
5. Preview with Space or the preview button; guitar/vocal sliders change the sound during playback.
6. Save (⌘S). By default the file is written next to the source and the file manager opens with it selected; this can be disabled in Settings.

Output names follow the action, for example Song_countin.mp3 for a count-in only, or Song_backing_guitar-off_vocals-6dB.mp3 for a backing track. There is also a command line for batch or scripted use, e.g. uv run countin song.mp3 --style buildup --bars 2, with flags for meter, BPM, first beat, pre-roll handling, volume and output format.

Constraints

• Builds are distributed only for Windows 10/11 x64 and macOS 12+ Apple Silicon; there is no listed Intel Mac or Linux build.
• Guitar/vocal separation is the slow part: about 2.5 minutes for a 4-minute song on Apple Silicon, and noticeably longer on Windows, where only the CPU is used. Jobs can be cancelled and show stage, percent and time remaining.
• The app checks GitHub for updates at startup and downloads about 800 MB in the background, verifying SHA-256 against the release. Self-update requires a writable folder; running from Downloads on macOS or a protected folder instead opens the download page.
• Licensing is MIT for the project, but bundled third-party components keep their own terms: BS-RoFormer SW weights are described as the author's terms with no explicit license, while FFmpeg is an LGPL build, Qt for Python is LGPL and PyTorch is BSD.

HotRepo's take

On the 2026-11-15 Python new-project snapshot, Count-In recorded 106 stars and 6 stars gained over a window of roughly 2.5 hours, with no forks. The sampled star curve rises from 100 to 106 within that morning, so the signal is a short, early burst rather than a sustained ranking history; the repository has no recorded GitHub daily rank. What stands out technically is the pairing of beat/downbeat detection with a single-pass guitar/vocal separator inside a desktop workflow, plus an explicit CLI and release verifier. The unsigned prebuilt binaries and the self-updater are the main things to weigh before installing.