Yamaha Style Studio

Browser app to design, edit, and export Yamaha arranger styles (.sty) from Live Audio Styles (.aus) and MIDI.

Files never leave your device — all parse / compile runs client-side.

Quick start

npm install
npm run dev

1. Drop a .aus (Audio Phraser) or open an existing .sty
2. Optionally drop .mid files onto timeline lanes (Bass / Chord / Pad / Phrase)
3. Preview, set name / tempo, pick your target keyboard
4. Compile .sty → Download → USB → Style → User

Supported targets

Model	SFF	Live Audio (.aus)
Genos2, Genos	SFF2	yes
PSR-SX920 / SX720	SFF2	yes
PSR-SX900 / SX700	SFF2	yes
PSR-S975 / S970	SFF2	yes
Tyros5	SFF2	yes
PSR-SX600	SFF2	no

The target model is picked in the export panel and drives validation. PSR-SX600 does not play Audio Styles — compiling an .aus-derived style for it is refused up front rather than letting you discover it from “Data not loaded properly” on the instrument. The model list lives in src/lib/models/models.ts; add a model there and the picker, the gate and the size advisory all follow.

File format — what the real files contain

Both dialects below were dumped byte-for-byte from real files. fixtures/knjraus.aus (an Audio Phraser export) and working demo sty.sty (a loadable Live Audio style) are the project's ground truth and are asserted against in npm run test:binary.

knjraus.aus                       working demo sty.sty
MThd (fmt 0, div 960)             MThd (fmt 0, div 1920)
MTrk (126 B)                      MTrk
CASM   204                        CASM   2122
  CSEG                              CSEG ×3   (sections are grouped!)
    Sdec  15 sections                 Sdec + N × Ctab(27) + N × Cntt(2)
    Ctb2  47 B (rhythm only)        OTSc   8580
AASM   159                          FNRc   141
  ASEG ▸ Adec + Atab + AMix         AASM   318
AFil   508459                       AFil   1016974
  ADSg ▸ AWav ▸ WAVE ▸ Afmt/Adat
  AInf ▸ BPnt (beat grid)

Notes that matter when you touch the writers:

• Ctb2 is 47 bytes, Ctab is 27. Both appear inside files marked SFF2. Audio Phraser writes a single Ctb2 (the rhythm channel); a real loadable style uses Ctab. We generate Ctab by default because that layout is verified field-by-field against 14 varied real tables — see the header of sff2Writer.ts for the byte map.
• Cntt (2 B per channel) carries the bass-on flag, not Ctab byte 21.
• Sections are split across several CSEG groups. Reading only the first one loses most of the section list.
• Section content starts at measure 2. Measure 1 is the SFF setup block (SFF2 / SInt markers, System On sysex, initial CCs). Both reference files put their first section marker exactly one bar in.
• A real style's declared MTrk length can overrun the start of CASM — in the reference style by 67 bytes. findEffectiveSmfEnd() prefers the CASM offset. This is not corruption; do not "fix" it.
• AInf ▸ BPnt is a table of (sample-frame, beat) pairs — the exact loop grid. We read it in preference to guessing tempo from the loop duration.

AUS → STY layout (default)

SMF Format 0 (SFF2 + SInt [+ optional timeline MIDI only])
→ CASM (from your .aus only)
→ OTSc (One Touch Setting — required for Style Editor)
→ FNRc (preserved when the source has one)
→ AASM / AFil (Live Audio from your .aus only)

Style Editor crash (Unexpected error / SRJRRR L11): caused by Live Audio .sty files missing OTSc after CASM. The exporter always injects one.

Source	Used?
Your .aus CASM	Yes
Your .aus AASM / AFil / AWav	Yes
Your .aus FNRc	Yes, when present
Timeline MIDI you assigned	Only if present
Demo / template MIDI channels	Never

Why not the hex-editor forum method?

The psrtutorial recipe (convert .aus → .sty, then splice AASM + AFil from your .aus into a donor style's MThd…SInt header) was a genuinely useful discovery — it identified AASM + AFil as the chunks that carry the audio. As a procedure it has three problems this app avoids:

1. The donor's CASM comes with it. You inherit another style's channel tables, note limits, chord rules and section list. Nothing describes your audio.
2. The chunk order in that post is wrong. It lists OTSc last; real loadable styles put it between CASM and AASM, and the Style Editor crashes when it is missing there.
3. “Copy MThd until SInt” drops the section markers, because SInt sits at tick 0 and every marker comes after it. You get a style with no section map.

This app takes CASM, AASM/AFil and FNRc from your .aus only, rebuilds the conductor around them, and injects the OTSc the .aus lacks.

Keyboard export validation

Export validates structure before download, then checks the result against your target model. Prefer Require source CASM (on by default).

If the keyboard says “Data not loaded properly”:

• Re-export a complete .aus from Audio Phraser (CASM + AASM/AFil)
• Confirm the build log: CASM from AUS + AASM/AFil audio
• Confirm the target model in the export panel actually supports Audio Styles
• Copy the full file to a FAT32 USB stick; load from User memory

Scripts

Command	Purpose
npm run dev	Vite dev server
npm run build	Typecheck + production build
npm run typecheck	tsc -b
npm test	Both suites below
npm run test:binary	Golden-byte tests against the real .aus / .sty fixtures
npm run test:export	Structure + round-trip tests

The test scripts import the app's real .ts modules through a small loader (scripts/ts-loader.mjs) built on Node's own type-stripping — no toolchain download, and they run offline. Requires Node 22.18+.

Known gaps

• The exported OTSc is still a borrowed template when your .aus has none — it carries another style's voices. Generating OTS from your chosen sounds is not done yet.
• One MIDI part per channel, not per (channel × section): you cannot yet give Bass one pattern in Main A and a different one in Main B.
• No CASM editor — note transposition rules, note limits and retrigger behaviour can only be copied from the source, not edited.
• Voice names in gmPrograms.ts are GM-accurate but the “XG” bank numbers are not real PSR-SX/Genos voice addresses. No MegaVoice / Super Articulation handling.

Targets & limits

• Preview uses GM SoundFonts (smplr); needs network once to fetch fonts
• Sections: timeline section pills; CASM Sdec aligned when lifted from AUS

License

See LICENSE.