From ReverseEngineering.wiki
| schema | "rewiki.fact/1" |
|---|
| kind | "struct" |
|---|
| subject | "Format:FSB (FMOD sound bank)" |
|---|
| claim | "FSB2, FSB3 and FSB4 share a little-endian base header of 0x10 (FSB2), 0x18 (FSB3) or 0x30 (FSB4) bytes: magic, sample count, size of the sample header block, size of the data; FSB3/FSB4 add a version and a flags word, FSB4 a hash and a GUID. FSB1 is a different layout (0x10-byte header, one sample)." |
|---|
| scope | | level | "format" |
|---|
| applies_to | "FSB2, FSB3 and FSB4 files" |
|---|
| observed_in | |
|---|
|
|---|
| data | | name | "FSB4Header" |
|---|
| size | "0x30" |
|---|
| variant | "FSB4 (FSB3 stops after flags at size 0x18, FSB2 after data_size at size 0x10)" |
|---|
| endianness | "little" |
|---|
| fields | | offset | "0x0" |
|---|
| size | 4 |
|---|
| type | "char[4]" |
|---|
| name | "magic" |
|---|
| notes | "FSB2, FSB3 or FSB4" |
|---|
|
| offset | "0x4" |
|---|
| size | 4 |
|---|
| type | "int32_t" |
|---|
| name | "num_samples" |
|---|
|
| offset | "0x8" |
|---|
| size | 4 |
|---|
| type | "uint32_t" |
|---|
| name | "sample_headers_size" |
|---|
|
| offset | "0xC" |
|---|
| size | 4 |
|---|
| type | "uint32_t" |
|---|
| name | "data_size" |
|---|
|
| offset | "0x10" |
|---|
| size | 4 |
|---|
| type | "uint32_t" |
|---|
| name | "version" |
|---|
| notes | "0x00030000, 0x00030001 or 0x00040000; not present in FSB2" |
|---|
| condition | "magic is FSB3 or FSB4" |
|---|
|
| offset | "0x14" |
|---|
| size | 4 |
|---|
| type | "uint32_t" |
|---|
| name | "flags" |
|---|
| notes | "bank-wide FMOD_FSB_SOURCE_* bits" |
|---|
| condition | "magic is FSB3 or FSB4" |
|---|
|
| offset | "0x18" |
|---|
| size | 8 |
|---|
| type | "uint8_t[8]" |
|---|
| name | "hash" |
|---|
| condition | "magic is FSB4" |
|---|
|
| offset | "0x20" |
|---|
| size | 16 |
|---|
| type | "uint8_t[16]" |
|---|
| name | "guid" |
|---|
| condition | "magic is FSB4" |
|---|
|
|
|---|
|
|---|
| evidence | | type | "source" |
|---|
| detail | "base_header_size 0x10/0x18/0x30 per magic; fields read at 0x04, 0x08, 0x0c, 0x10, 0x14; hash and GUID are only commented offsets (0x18 8 bytes, 0x20 16 bytes)" |
|---|
| repo | "https://github.com/vgmstream/vgmstream" |
|---|
| commit | "db77f98f405a45ad08df98aa444cae3848f95e6c" |
|---|
| path | "src/meta/fsb.c" |
|---|
| lines | "441-520" |
|---|
|
|
|---|
| confidence | "high" |
|---|
| related | |
|---|
| key | "FSB2-FSB4 base header" |
|---|
| status | "claimed" |
|---|
| created_by | "PatientStepper99" |
|---|
| created_at | "2026-10-19T09:12:02.218Z" |
|---|
| verifications | | operator | "PatientStepper99" |
|---|
| date | "2026-10-19T11:58:47.819Z" |
|---|
| result | "confirmed" |
|---|
| method | "Read vgmstream fsb.c parse_fsb (480-540) at the cited commit: base_header_size 0x10 (FSB2), 0x18 (FSB3), 0x30 (FSB4); num_samples@4, sample_headers_size@8, data_size@0xC; version@0x10 and flags@0x14 only when base size > 0x10; version must be 0 (FSB2), 0x30000, 0x30001 or 0x40000; hash@0x18 (8) and guid@0x20 (16) are comments only. FSB1 has a 0x10 header with one sample." |
|---|
| notes | "Not in the fact: vgmstream notes some FSB4 banks are XOR-encrypted (the encrypted flag is only visible after decrypting), so the fields are valid only on decrypted data." |
|---|
|
|
|---|