AGSC (File Format)
AGSC is the sound effect format for Metroid Prime and Metroid Prime 2: Echoes. Each AGSC file contains a group of sound effects. The first two Metroid Prime games utilize the MusyX audio engine created by Factor5 and as a result, AGSC files are essentially just embedded MusyX files.
The audio codec used in AGSC is the standard GameCube DSP-ADPCM codec, but MusyX itself also offers uncompressed PCM as an option.
|To do: |
Better descriptions for how SoundMacros work and a description for what each command does.
|This file format is almost completely documented |
The proj and sdir chunks have a couple unknowns left.
The AGSC format is essentially four data chunks combined into one resource, each of which is a standard MusyX file. Of the four data chunks (pool, proj, samp, and sdir), there's one for sound engine scripts, one for sound properties, one for actual ADPCM sound data, and one for sound metadata. The main difference between Prime 1 and 2 is the header, and some slight changes in the way the four chunks are organized. In Metroid Prime, each chunk begins with its own size value; in Metroid Prime 2, every chunk instead has its size listed at the beginning of the file, at the end of the header. In addition, in Metroid Prime, the third chunk is samp, and the fourth is sdir; in Metroid Prime 2, it's the other way around.
|0x0||D||Audio Directory. Always "Audio/". Zero-terminated.|
|0x0 + D||N||Audio Group Name. Zero-terminated.|
|0x0 + D + N||End of header|
Metroid Prime 2
|0x0||4||Unknown; always 1|
|0x4||D||Audio Group Name. Zero-terminated.|
|0x4 + D||2||Group ID; 0xFFFF if unspecified|
|0x6 + D||4||Pool size|
|0xA + D||4||Project size|
|0xE + D||4||Sample directory size|
|0x12 + D||4||Sample size|
|0x16 + D||End of header|
The Pool chunk contains sub-chunk tables for SoundMacros, ADSR, keymaps, and layers, if applicable. It starts with a 16-byte header before the different data tables begin.
|0x0||4||SoundMacros Offset (always 0x10)|
|0x10||End of entry|
After this are four tables of objects. Each object is identified with a 16-bit ObjectID.
Factor5 designed ObjectIDs to used in a polymorphic manner. The top 2 bits of the ID are used to differentiate between SoundMacros, Keymaps, and Layers. If the ID passes a mask of 0x4000, the object is a Keymap. If the ID passes a mask of 0x8000, the object is a Layer. Otherwise, the object is assumed to be a SoundMacro. Tables don't require this type of polymorphism due to the context in which they are accessed.
The first Pool table denotes MusyX's SoundMacros, small scripts that apply various effects on the sounds in the game. Each macro is composed of a header followed by a number of commands; each command specifies its type through a single-byte command ID, then specifies the parameters of that particular command, which vary.
The header of each SoundMacro is eight bytes, and is structured as follows:
|0x0||4||Chunk Size (note: includes the size value itself)|
On the commands, each 4 bytes were originally little endian, but have been swapped to big endian in the AGSC files (despite not being longs). To read the data as originally formatted, every four bytes needs to be byte-swapped. Each command is 8 bytes, and is structured as follows:
|0x0||1||Command ID (varies; there are 79 known commands in the MusyX audio engine.)|
|0x1||7||Command arguments (varies between commands)|
|0x8||End of entry|
The SoundMacro will continue with commands until it terminates when the END command is executed. The command ID for END is 0 and has null command arguments; the next SoundMacro begins after reading it.
These are the possible commands:
|0x2||SPLITKEY||Keynumber||SoundMacro ID||SoundMacro step|
|0x3||SPLITVEL||Velocity||SoundMacro ID||SoundMacro step|
|0x6||GOTO||SoundMacro ID||SoundMacro step|
|0x8||PLAYMACRO||Addnote||SoundMacro ID||SoundMacro step||Priority||MaxVoices|
|0xA||SPLITMOD||Mod value||SoundMacro ID||SoundMacro step|
|0xC||SETADSR||Table-ID (ADSR)||DLS mode|
|0xD||SCALEVOLUME||Scale||Add||Table-ID (Curve)||Org vol|
|0xE||PANNING||Pan position||Time ms||Width|
|0xF||ENVELOPE||Scale||Add||Table-ID (Curve)||Org vol|
|0x13||SPLITRND||RND||SoundMacro ID||SoundMacro step|
|0x14||FADE-IN||Scale||Add||Table-ID (Curve)||ms switch||Ticks/Millisec.|
|0x15||SPANNING||Pan position||Time ms||Width|
|0x17||RNDNOTE||Note Lo||Detune||Note Hi||Fixed/Free||Abs/Rel|
|0x18||ADDNOTE||Add||Detune||org key||ms switch||Ticks/Millisec.|
|0x1B||PORTAMENTO||Port. State||Port. Type||ms switch||Ticks/Millisec.|
|0x1C||VIBRATO||Level note||Level fine||Modwheel flag||ms switch||Ticks/Millisec.|
|0x20||SETPITCHADSR||Table-ID (ADSR)||Note range||Detune range|
|0x21||SCALEVOLUME DLS||Scale||Org vol|
|0x22||MOD2VIBRANGE||Key range||Cent range|
|0x23||SETUP TREMOLO||Tremolo scale||Modw. add scale|
|0x25||GOSUB||SoundMacro ID||SoundMacro step|
|0x28||TRAP_EVENT||Event||SoundMacro ID||SoundMacro step|
|0x33||PITCHWHEELR||Range up||Range down|
|0x39||AGECNTVEL||AGE Base||AGE Scale|
|0x40||VOL_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x41||PAN_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x42||PitchW_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x43||ModW_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x44||PEDAL_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x45||PORTA_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x46||REVERB_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x47||SPAN_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x48||DOPPLER_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x49||TREMOLO_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x4A||PREA_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x4B||PREB_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x4C||POSTB_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x4D||AUXAFX_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x4E||AUXBFX_SELECT||MIDI Contr.||Scaling percentage||Combine||is var.||Fine scaling|
|0x50||SETUP_LFO||LFO Nr.||Period in ms|
|0x5A||SRCMODE_SELECT||SRC type||Type 0 SRC filter|
|0x60||ADD_VARS||Var/Ctrl||A =||Var/Ctrl||B +||Var/Ctrl||C|
|0x61||SUB_VARS||Var/Ctrl||A =||Var/Ctrl||B -||Var/Ctrl||C|
|0x62||MUL_VARS||Var/Ctrl||A =||Var/Ctrl||B *||Var/Ctrl||C|
|0x63||DIV_VARS||Var/Ctrl||A =||Var/Ctrl||B /||Var/Ctrl||C|
|0x64||ADDI_VARS||Var/Ctrl||A =||Var/Ctrl||B +||Immediate|
|0x70||IF_EQUAL||Ctrl||A ==||Ctrl||B||Not||SoundMacro Step|
|0x71||IF_LESS||Ctrl||A <||Ctrl||B||Not||SoundMacro Step|
After the last soundmacro, the table terminated by a value of 0xFFFF.
Tables have two functions: for defining curves for volume scaling, or to be used as ADSR envelopes.
The tables continue until 0xffffffff terminator is reached.
|Chunk Size||ADSR/Curve data|
When the size of the table data is exactly 8, it may represent ADSR envelopes with this structure:
Note: All fields of the envelope are little endian.
|0x0||2||Attack time; in milliseconds|
|0x2||2||Decay time; in milliseconds|
|0x4||2||Sustain; percentage mapped between [0x0,0x1000]|
|0x6||2||Release time; in milliseconds|
|0x8||End of ADSR|
MusyX can also express more advanced envelopes using a modified DLS representation. This representation includes scaling coefficients to respond to played note and velocity (so slamming down a key harder plays longer).
The attack and decay members are expressed in time-cents. This may be converted to seconds using:
2timecents / (1200.0 * 65536.0)
The attack and decay scale members are expressed as 0.1% increments in 16.16 fixed-point.
This may be converted to a normalized factor using:
scale / (1000.0 * 65536.0)
Note: All fields of the envelope are little endian.
|0x0||4||Attack time; in time-cents|
|0x4||4||Decay time; in time-cents|
|0x8||2||Sustain; percentage mapped between [0x0,0x1000]|
|0xA||2||Release; in milliseconds|
|0xC||4||Velocity to Attack Scale; 0.1% increments as 16.16 fixed-point|
|0x10||4||Key to Decay Scale; 0.1% increments as 16.16 fixed-point|
|0x14||End of DLS ADSR|
To express a volume curve, the table data is simply an arbitrarily-sized table of
uint8_t values (although typically in MIDI range [0,127])
Keymaps are swappable, fixed-length tables mapping 128 MIDI keys to sound-producing objects.
The keymaps continue until 0xffffffff terminator is reached.
|0x0||4||Chunk Size; (usually 0x1032)|
|Chunk Size||128 Keymap entries|
|0x8||Padded to 8 bytes|
Layers are one-to-many, ranged keyboard mappings to sound-producing objects.
The layers continue until 0xffffffff terminator is reached.
|Chunk Size||Layer data|
Within the layer data, there is a u32 count of layer range structs:
|0x7||1||Surround Pan; (0: extreme forward, 64: center, 127: extreme rearward)|
|0x8||1||Pan; (0: extreme left, 64: center, 127: extreme right)|
|0xC||Padded to 12 bytes|
The entire Pool chunk is terminated by a value of 0xFFFF.
The Project properties chunk contains values for the sounds, including priority, polyphony, volume, etc.
Structurally, the Project is the root of the Audio Group tree, defining one or more Song Groups or SFX Groups
|0x0||4||Group end offset (points to next group in project)|
|0x4||2||Group Type; 0 for SongGroup (for use with CSNG), 1 for SFXGroup.|
|0x8||4||SoundMacro ID table offset|
|0xC||4||Sample ID table offset|
|0x10||4||Tables table offset|
|0x14||4||Keymaps table offset|
|0x18||4||Layers table offset|
|0x1C||4||Normal page table (SongGroup) / SFX table offset (SFXGroup)|
|0x20||4||Drum page table offset (SongGroup)|
|0x24||4||MIDI Setup table offset (SongGroup)|
|0x20||End of group header|
After the header are a number of data tables.
SoundMacro ID Table
This is a ranged-table of shorts; there's no count value, so it's terminated with a value of 0xFFFF. It's a list of SoundMacro IDs present in the file. Contiguous ranges are expressed by IDs with most-significant bit set (0x8000). The range begins on the marked ID and incrementally reaches the next ID in the list, including that ID. All other IDs are singular.
Sample ID / Table / Keymap / Layer Tables
These function the same way as the SoundMacro ID table, but indexes other types of entities instead.
Note: Keymap and Layer IDs in these tables have their top 2 bits (indicating their type) masked off. Keymaps must be OR'd with 0x4000 and Layers must be OR'd with 0x8000 in order to reconstruct the actual IDs.
Normal / Drum Page Entry
Used to map General MIDI program numbers (instruments) to sound entities (macros, keymaps, layers)
|0x2||1||Priority; voices are limited, so priority is used to play more important sounds over others|
|0x3||1||Max number of voices|
|0x4||1||GM Program Number|
Note: The drum table is accessed when the MIDI channel is 10, otherwise the normal table is accessed.
Used to map auto-generated
#define IDs (used by game code) to sound entities (macros, keymaps, layers)
This table begins with a 16-bit count value, then 16 bits of padding. Each entry in the table is 10 bytes.
|0x0||2||DefineID; referenced by game code|
|0x4||1||Priority; voices are limited, so priority is used to play more important sounds over others|
|0x5||1||Max number of voices|
|0x6||1||Definite Velocity; volume (usually 7F)|
|0x8||1||Definite Key; The default pitch - usually 0x3C (MIDI C4)|
MIDI Setup Entry
Table of fixed-length tables to map all 16 MIDI channels to program numbers (in-turn resolving to sound entities via the page table).
Multiple MIDI Setups may be created to support Song data requiring totally different banks of instruments.
Each MIDI Setup starts with a u16 MIDI-Setup-ID, then 16-bits padding, then 16 entries of the following structure (one for each channel):
MIDI setups continue until the group end offset is reached.
The Sample chunk is all the sound data encoded using the standard Gamecube DSP ADPCM codec. It can be decoded the same way as a DSP file. Each sound's size is padded to 32 bytes before the next sound's data begins.
The Sample Directory chunk (chunk 4 in Metroid Prime, chunk 3 in Metroid Prime 2) is made up of two sets of tables. The structure of both these tables is identical between both games.
The first metadata table has one entry per sound, and is terminated with 0xFFFFFFFF; since there's no known sound count anywhere in the file, the only way to read this correctly is to read until you reach the terminator value. Each entry is 0x20 bytes long.
|0x2||2||Padding; always 0|
|0x4||4||Sound start offset, relative to the start of the ADPCM chunk|
|0xC||1||Base Note; Corresponds to the MIDI note played in the sample, at the native sample-rate (which MusyX obtains from the INST chunk of .aiff files or SMPL chunk of .wav files, along with looping info). To play at a specified pitch in cents, set the playback sample rate using this formula: |
|0xD||1||Padding; always 0|
|0x10||1||Audio format |
|0x11||3||Number of samples|
|0x14||4||Loop start sample|
|0x18||4||Loop length, in samples. To get the loop end sample, add this to the start sample and subtract 1.|
|0x1C||4||Table B entry offset, relative to the start of the sound metadata chunk|
|0x20||End of entry|
These are accessed through the offsets in table A's entries; note that it might not match the sound count, because the same entry in this table can be used with multiple sounds. Each entry is 0x28 bytes long.
|0x0||2||Unknown; always 8|
|0x2||1||Initial predictor/scale (matches first frame header)|
|0x3||1||Loop predictor/scale (matches loop start frame header)|
|0x4||2||Loop context sample history 2|
|0x6||2||Loop context sample history 1|
|0x8||2 × 16||Decode coefficients|
|0x28||End of entry|
- Prime Audio Decoder by Aruki will dump all sound effects contained in a given AGSC file.