skate-3-rust-engine/tools/audio-file-inspect

Audio file inspection

Readers for Skate 3's audio formats, run against your own disc. They check the layouts the native audio runtime (crates/skate-audio) and the audio setup (tools/asset_pipeline/audio_export.py) rely on, and let you look inside a bank, an emitter file or the grain data. All read-only, all headless.

Script	What it does
aems_survey.py	Surveys every ABKC bank (.abk) and MOIR project (.csi) in audiofiles.big: header invariants, records per bank, capacities, program opcode census, export kinds, voice objects, curve types, and a failure list. --dump BANK.abk prints one bank's records, instance layout and program.
bank_layout_check.py	Checks that each bank's S10A slot order matches the stream scan setup uses (so exported WAV index = slot index), and whether the rebase / interface lists lie after the sample data (they do, so the runtime needs the whole .abk). --extract DIR also writes every .abk / .csi (plus csi_order.txt) for the tools below.
decode_bank_samples.py	Decodes chosen banks' samples to WAVs the way setup does (<out>/<stem>/NNNN.wav, slot order), for banks the install does not export. Uses vgmstream-cli.
multichannel_census.py	Which sample-group entries of the banks' player objects use multichannel (2/4/6-channel) samples, and their azimuth bytes.
splc_fields.py	Census of SPLC (.bnk) record, group and member fields: value ranges per 4-byte offset (as f32 and u32). With a bank and record id, dumps that record raw.
ems_dump.py	Dumps the world emitter files (.ems): records per file, the most used sounds (64-bit name ids resolved against the archive's member names), and with --records every record's position, extent, scalars and gains.
grain_survey.py	Surveys the granular rolling-bed data: per .grain member the header, seek table and EAAC stream header; per grain class the vault tuning (from the install's converted database). --rms adds a loudness / spectral-centroid profile per member. Writes JSON.

Inputs

• Your extracted disc: --disc DIR (the folder holding data/), default $SKATE3_DISC or .local/skate3-disc.
• Extracted banks for decode_bank_samples.py / multichannel_census.py: run bank_layout_check.py --extract .local/audio-file-inspect/banks first (the default location).
• .bnk files for splc_fields.py: extract them with tools/world-stream-inspect/big_list.py <disc>/data/audio/audiofiles.big "\.bnk$" --out DIR.
• grain_survey.py vault part: assets/private/stock/skater-collections.json from setup (--vault).

Outputs go to .local/audio-file-inspect/ (gitignored). They are decoded game data: keep them local, never commit them.

Usage

py -3.13 tools/audio-file-inspect/aems_survey.py
py -3.13 tools/audio-file-inspect/aems_survey.py --dump SomeBank.abk
py -3.13 tools/audio-file-inspect/bank_layout_check.py --extract .local/audio-file-inspect/banks
py -3.13 tools/audio-file-inspect/decode_bank_samples.py SomeBank OtherBank
py -3.13 tools/audio-file-inspect/multichannel_census.py
py -3.13 tools/audio-file-inspect/splc_fields.py .local/bnk
py -3.13 tools/audio-file-inspect/ems_dump.py sfx_ --records
py -3.13 tools/audio-file-inspect/grain_survey.py --rms

Example output

bank_layout_check.py (made-up counts):

120 banks; slot order mismatches 0; banks whose rebase/interface lists lie after the sample data (so the whole file is needed, not just the resident part): 120

ems_dump.py (made-up names):

data/audio/sfx_example.ems: 42 records; top: [('water_loop', 12), ('?0123456789ABCDEF', 3)]
records 42, sound ids resolved 39

Requirements

• Python 3.13; numpy only for grain_survey.py --rms.
• The repository's setup modules (tools/owned_game, tools/asset_pipeline/audio_formats.py, audio_export.py).
• vgmstream-cli for decode_bank_samples.py and grain_survey.py --rms: setup downloads it to data/tools/vgmstream-cli/; or pass --vgmstream / set VGMSTREAM.