How It Works

How rearranger.ai works

Four simple steps from upload to download

1
Upload your style file
Drag and drop or browse for your .sty (Korg), .sst/.pst (Yamaha) or .kst/.evs (Ketron) file. Select source and target keyboards.
2
Automatic conversion
Our engine normalises channels, remaps drum notes and detects harmonic sections automatically.
3
Review & confirm
Check the detected root keys and drum mappings. Override anything that needs a human touch.
4
Download
Your converted style file is ready. Save a drum map for faster future conversions.
Pipeline

Five stages of conversion

1
Parse & validate
The uploaded file is parsed according to the source keyboard's proprietary MIDI-based format spec. For Yamaha, this covers CASM chunks, SFF1/SFF2 style structures, and OTS data. For Korg, the Pa-format arrangement file is decoded. For Ketron, the AKS container is unwrapped. Validation checks that all required sections (Main A/B, Fill, Intro, Ending) are present and structurally intact before any conversion begins.
2
Drum track detection
All MIDI tracks are analysed using the three-pass drum detection engine (described in detail below) to identify which channels carry percussion data. Non-drum tracks are preserved as-is for root key analysis. Detected drum tracks are queued for remapping.
3
Root key analysis
Each non-drum channel in each style section is analysed independently to determine the root key of the chord voicings. The analyser uses a chromagram-based approach over the MIDI note distribution, combined with the style's known chord root hints where available.
4
Mapping & remapping
With drum tracks identified and root keys detected, the engine applies mappings. For drums: each source MIDI note is looked up in the active drum map and remapped to the target note. For chord channels: NTT tables are recalculated based on the detected root key and the target keyboard's chord engine conventions.
5
Serialise to target format
The remapped MIDI events, section metadata, and NTT data are serialised into the target keyboard's file format. All style sections are written in the correct order with the target's required headers and footers. The output file is validated before it is made available for download.
Drum detection engine

Three-pass drum identification

No single heuristic is reliable across all keyboards. rearranger.ai runs three independent passes and combines their scores to reach a confidence-weighted decision.

Pass 1
Structural analysis
Checks MIDI channel 10 (GM percussion convention), CASM drum flags (Yamaha), and Pa-format track type markers (Korg). This pass is fast and highly accurate for standard-format files.
Pass 2
Statistical analysis
Examines the note distribution of each channel. Drum tracks have a characteristic sparse pitch range (usually 35–81) with high repetition. A Gini coefficient score separates drum from melodic tracks reliably.
Pass 3
Rhythmic fingerprint
Applies a rhythmic density analysis — drum tracks fire on beat subdivisions with low onset-to-onset variance. A short-time Fourier transform of the onset density is matched against known percussion rhythm templates.

Confidence scoring

The three passes produce individual scores (0–1). A weighted average — structural (40%), statistical (35%), rhythmic (25%) — gives the final confidence. Tracks below 60% are flagged for manual review in the UI.

92%
High confidence
Auto-confirmed, no review needed
71%
Medium confidence
Suggested but flagged for review
48%
Low confidence
Highlighted, manual override required
Ready to bridge your styles?
Upload your first style file and watch the engine work in real time.
Create free account Tutorials