Add .KTAC [Kin'iro no Corda 3 (Vita), Atelier Ryza (PS4)]

This commit is contained in:
bnnm 2021-01-24 20:17:07 +01:00
parent 628e98116b
commit d71c280729
8 changed files with 528 additions and 0 deletions

View File

@ -251,6 +251,7 @@ static const char* extension_list[] = {
"kovs", //fake extension/header id for .kvs
"kns",
"kraw",
"ktac",
"ktsl2asbin",
"ktss", //fake extension/header id for .kns
"kvs",
@ -1332,6 +1333,7 @@ static const meta_info meta_info_list[] = {
{meta_DSP_WIIADPCM, "Exient WIIADPCM header"},
{meta_DSP_CWAC, "CRI CWAC header"},
{meta_COMPRESSWAVE, "CompressWave .cwav header"},
{meta_KTAC, "Koei Tecmo KTAC header"},
};
void get_vgmstream_coding_description(VGMSTREAM* vgmstream, char* out, size_t out_size) {

View File

@ -884,6 +884,10 @@
RelativePath=".\meta\kraw.c"
>
</File>
<File
RelativePath=".\meta\ktac.c"
>
</File>
<File
RelativePath=".\meta\ktsc.c"
>

View File

@ -202,6 +202,7 @@
<ClCompile Include="meta\imuse.c" />
<ClCompile Include="meta\ios_psnd.c" />
<ClCompile Include="meta\kat.c" />
<ClCompile Include="meta\ktac.c" />
<ClCompile Include="meta\ktsc.c" />
<ClCompile Include="meta\ktsr.c" />
<ClCompile Include="meta\ktss.c" />

View File

@ -1825,6 +1825,9 @@
<ClCompile Include="meta\mogg.c">
<Filter>meta\Source Files</Filter>
</ClCompile>
<ClCompile Include="meta\ktac.c">
<Filter>meta\Source Files</Filter>
</ClCompile>
<ClCompile Include="meta\ktsc.c">
<Filter>meta\Source Files</Filter>
</ClCompile>

514
src/meta/ktac.c Normal file
View File

@ -0,0 +1,514 @@
#include "meta.h"
#include "../coding/coding.h"
typedef struct {
int channels;
int sample_rate;
int loop_flag;
int32_t num_samples;
int32_t loop_start;
int32_t loop_end;
uint32_t file_size;
uint32_t stream_offset;
uint32_t stream_size;
uint32_t table_offset;
uint32_t table_entries;
int type;
int encoder_delay;
int end_padding;
int frame_samples;
} ktac_header_t;
static int make_m4a_header(uint8_t* buf, int buf_len, ktac_header_t* ktac, STREAMFILE* sf);
/* KTAC - Koei Tecmo custom AAC [Kin'iro no Corda 3 (Vita), Shingeki no Kyojin: Shichi kara no Dasshutsu (3DS), Dynasty Warriors (PS4)] */
VGMSTREAM* init_vgmstream_ktac(STREAMFILE* sf) {
VGMSTREAM* vgmstream = NULL;
ktac_header_t ktac = {0};
/* checks */
/* .ktac: header id */
if (!check_extensions(sf,"ktac"))
goto fail;
if (!is_id32be(0x00,sf, "KTAC"))
goto fail;
/* 0x04: version? (always 1) */
ktac.file_size = read_u32le(0x08,sf);
if (ktac.file_size != get_streamfile_size(sf))
goto fail;
ktac.stream_offset = read_u32le(0x0c,sf);
ktac.stream_size = read_u32le(0x10,sf);
ktac.type = read_u32le(0x14,sf);
ktac.sample_rate = read_u32le(0x18,sf);
ktac.num_samples = read_u32le(0x1c,sf); /* full samples */
ktac.channels = read_u16le(0x20,sf);
ktac.frame_samples = read_u16le(0x22,sf);
ktac.encoder_delay = read_u16le(0x24,sf);
ktac.end_padding = read_u16le(0x26,sf);
ktac.loop_start = read_u32le(0x28,sf);
ktac.loop_end = read_u32le(0x2c,sf);
/* 0x30: ? (big, related to loops) */
/* 0x34: ? (always null) */
ktac.table_offset = read_u32le(0x38,sf);
ktac.table_entries= read_u32le(0x3c,sf);
ktac.loop_flag = (ktac.loop_end > 0);
/* type 1 files crash during sample_copy, wrong fake header/esds?
* (0=AoT, KnC3 bgm, 1=KnC3 1ch voices, 2=DW4, Atelier Ryza) */
if (ktac.type == 1)
goto fail;
/* build the VGMSTREAM */
vgmstream = allocate_vgmstream(ktac.channels, ktac.loop_flag);
if (!vgmstream) goto fail;
vgmstream->meta_type = meta_KTAC;
vgmstream->sample_rate = ktac.sample_rate;
vgmstream->num_samples = ktac.num_samples - ktac.encoder_delay - ktac.end_padding;
vgmstream->loop_start_sample = ktac.loop_start * ktac.frame_samples - ktac.encoder_delay;
vgmstream->loop_end_sample = ktac.loop_end * ktac.frame_samples - ktac.encoder_delay;
/* KTAC uses AAC, but not type found in .aac (that has headered frames, like mp3) but raw
* packets + frame size table (similar to .mp4/m4a). We make a fake M4A header to feed FFmpeg */
#ifdef VGM_USE_FFMPEG
{
ffmpeg_codec_data* ffmpeg_data = NULL;
int bytes;
uint8_t* buf = NULL;
int buf_len = 0x400 + ktac.table_entries * 0x4;
if (buf_len > 0x100000) /* ??? */
goto fail;
buf = malloc(buf_len);
if (!buf) goto fail;
bytes = make_m4a_header(buf, buf_len, &ktac, sf);
ffmpeg_data = init_ffmpeg_header_offset(sf, buf, bytes, ktac.stream_offset, ktac.stream_size);
free(buf);
if (!ffmpeg_data) goto fail;
vgmstream->codec_data = ffmpeg_data;
vgmstream->coding_type = coding_FFmpeg;
vgmstream->layout_type = layout_none;
/* not part of fake header since it's kinda complex to add (iTunes string comment) */
ffmpeg_set_skip_samples(ffmpeg_data, ktac.encoder_delay);
}
#else
goto fail;
#endif
return vgmstream;
fail:
close_vgmstream(vgmstream);
return NULL;
}
/* *********************************************************** */
/* Helpers for M4A headers, an insane soup of chunks (AKA "atoms").
* Needs *A LOT* of atoms and fields so this is more elaborate than usual.
* - https://developer.apple.com/library/archive/documentation/QuickTime/QTFF/QTFFPreface/qtffPreface.html
*/
/* generic additions */
typedef struct {
uint8_t* out;
int bytes;
} m4a_state_t;
typedef struct {
STREAMFILE* sf;
ktac_header_t* ktac; /* config */
uint8_t* out; /* current position */
int bytes; /* written bytes */
m4a_state_t chunks; /* chunks offsets are absolute, save position until we know header size */
} m4a_header_t;
static void add_u32b(m4a_header_t* h, uint32_t value) {
put_u32be(h->out, value);
h->out += 0x04;
h->bytes += 0x04;
}
static void add_u24b(m4a_header_t* h, uint32_t value) {
put_u16be(h->out + 0x00, (value >> 8u) & 0xFFFF);
put_u8 (h->out + 0x02, (value >> 0u) & 0xFF);
h->out += 0x03;
h->bytes += 0x03;
}
static void add_u16b(m4a_header_t* h, uint16_t value) {
put_u16be(h->out, value);
h->out += 0x02;
h->bytes += 0x02;
}
static void add_u8(m4a_header_t* h, uint32_t value) {
put_u8(h->out, value);
h->out += 0x01;
h->bytes += 0x01;
}
static void add_name(m4a_header_t* h, const char* name) {
memcpy(h->out, name, 0x4);
h->out += 0x04;
h->bytes += 0x04;
}
static void add_atom(m4a_header_t* h, const char* name, uint32_t size) {
add_u32b(h, size);
add_name(h, name);
}
/* register + write final size for atoms of variable/complex size */
static void save_atom(m4a_header_t* h, m4a_state_t* s) {
s->out = h->out;
s->bytes = h->bytes;
}
static void load_atom(m4a_header_t* h, m4a_state_t* s) {
put_u32be(s->out, h->bytes - s->bytes);
}
/* common atoms */
static void add_ftyp(m4a_header_t* h) {
add_atom(h, "ftyp", 0x18);
add_name(h, "M4A "); /* major brand */
add_u32b(h, 512); /* minor version */
add_name(h, "isom"); /* compatible brands */
add_name(h, "iso2"); /* compatible brands */
}
static void add_free(m4a_header_t* h) {
add_atom(h, "free", 0x08);
}
static void add_mdat(m4a_header_t* h) {
add_atom(h, "mdat", 0x08 + h->ktac->stream_size);
}
/* variable atoms */
static void add_stco(m4a_header_t* h) {
add_atom(h, "stco", 0x10 + 1 * 0x04);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 1); /* Number of entries */
/* there may be an entry per frame, but only first seems needed */
save_atom(h, &h->chunks);
add_u32b(h, 0); /* Absolute offset N */
}
static void add_stsz(m4a_header_t* h) {
int i;
add_atom(h, "stsz", 0x14 + h->ktac->table_entries * 0x04);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 0); /* Sample size (CBR) */
add_u32b(h, h->ktac->table_entries); /* Number of entries (VBR) */
for (i = 0; i < h->ktac->table_entries; i++) {
uint32_t size = read_u32le(h->ktac->table_offset + i*0x04, h->sf);
add_u32b(h, size); /* Sample N */
}
}
static void add_stsc(m4a_header_t* h) {
add_atom(h, "stsc", 0x1c);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 1); /* Number of entries */
add_u32b(h, 1); /* First chunk */
add_u32b(h, h->ktac->table_entries); /* Samples per chunk */
add_u32b(h, 1); /* Sample description ID */
}
static void add_stts(m4a_header_t* h) {
add_atom(h, "stts", 0x18);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 1); /* Number of entries */
add_u32b(h, h->ktac->table_entries); /* Sample count */
add_u32b(h, h->ktac->frame_samples); /* Sample duration */
}
/* from mpeg4audio.c (also see ff_mp4_read_dec_config_descr) */
static const int m4a_sample_rates[16] = {
96000, 88200, 64000, 48000, 44100, 32000, 24000, 22050, 16000, 12000, 11025, 8000, 7350
};
static const uint8_t m4a_channels[14] = {
0,
1, // mono (1/0)
2, // stereo (2/0)
3, // 3/0
4, // 3/1
5, // 3/2
6, // 3/2.1
8, // 5/2.1
//0,
//0,
//0,
//7, // 3/3.1
//8, // 3/2/2.1
//24 // 3/3/3 - 5/2/3 - 3/0/0.2
};
static void add_esds(m4a_header_t* h) {
uint16_t config = 0;
/* ES_descriptor (TLV format see ISO 14496-1) and DecSpecificInfoTag define actual decoding
- config (channels/rate/etc), other atoms with the same stuff is just info
* - http://ecee.colorado.edu/~ecen5653/ecen5653/papers/ISO%2014496-1%202004.PDF */
{
uint8_t object_type = 0x02; /* 0x00=none, 0x01=AAC main, 0x02=AAC LC */
uint8_t sr_index = 0;
uint8_t ch_index = 0;
uint8_t unknown = 0;
int i;
for (i = 0; i < 16; i++) {
if (m4a_sample_rates[i] == h->ktac->sample_rate) {
sr_index = i;
break;
}
}
for (i = 0; i < 8; i++) {
if (m4a_channels[i] == h->ktac->channels) {
ch_index = i;
break;
}
}
config |= (object_type & 0x1F) << 11; /* 5b */
config |= (sr_index & 0x0F) << 7; /* 4b */
config |= (ch_index & 0x0F) << 3; /* 4b */
config |= (unknown & 0x07) << 0; /* 3b */
}
add_atom(h, "esds", 0x33);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u8 (h, 0x03); /* ES_DescrTag */
add_u32b(h, 0x80808022); /* size 0x22 */
add_u16b(h, 0x0000); /* stream Id */
add_u8 (h, 0x00); /* flags */
add_u8 (h, 0x04); /* DecoderConfigDescrTag */
add_u32b(h, 0x80808014); /* size 0x14 */
add_u8 (h, 0x40); /* object type (0x40=audio) */
add_u8 (h, 0x15); /* stream type (6b: 0x5=audio) + upstream (1b) + reserved (1b: const 1) */
add_u24b(h, 0x000000); /* buffer size */
add_u32b(h, 0); /* max bitrate (256000?)*/
add_u32b(h, 0); /* average bitrate (256000?) */
add_u8 (h, 0x05); /* DecSpecificInfoTag */
add_u32b(h, 0x80808002); /* size 0x02 */
add_u16b(h, config); /* actual decoder info */
add_u8 (h, 0x06); /* SLConfigDescrTag */
add_u32b(h, 0x80808001); /* size 0x01 */
add_u8 (h, 0x02); /* predefined (2=default) */
}
static void add_mp4a(m4a_header_t* h) {
add_atom(h, "mp4a", 0x57);
add_u32b(h, 0); /* ? */
add_u32b(h, 1); /* Data reference index */
add_u32b(h, 0); /* Reserved */
add_u32b(h, 0); /* Reserved 2 */
add_u16b(h, h->ktac->channels); /* Channel count */
add_u16b(h, 16); /* Sample size */
add_u32b(h, 0); /* Pre-defined */
add_u16b(h, h->ktac->sample_rate); /* Sample rate */
add_u16b(h, 0); /* ? */
add_esds(h); /* elementary stream descriptor */
}
static void add_stsd(m4a_header_t* h) {
add_atom(h, "stsd", 0x67);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 1); /* Number of entries */
add_mp4a(h);
}
static void add_stbl(m4a_header_t* h) {
m4a_state_t s;
save_atom(h, &s);
add_atom(h, "stbl", 0x00);
add_stsd(h); /* Sample description */
add_stts(h); /* Time-to-sample */
add_stsc(h); /* Sample-to-chunk */
add_stsz(h); /* Sample size */
add_stco(h); /* Chunk offset */
load_atom(h, &s);
}
static void add_dinf(m4a_header_t* h) {
add_atom(h, "dinf", 0x24);
add_atom(h, "dref", 0x1c);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 1); /* Number of entries */
add_atom(h, "url ", 0x0c);
add_u32b(h, 1); /* Version (1 byte) + Flags (3 byte) */
}
static void add_smhd(m4a_header_t* h) {
add_atom(h, "smhd", 0x10);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u16b(h, 0); /* Balance */
add_u16b(h, 0); /* Reserved */
}
static void add_minf(m4a_header_t* h) {
m4a_state_t s;
save_atom(h, &s);
add_atom(h, "minf", 0x00);
add_smhd(h);
add_dinf(h);
add_stbl(h);
load_atom(h, &s);
}
static void add_hdlr(m4a_header_t* h) {
add_atom(h, "hdlr", 0x22);
add_u32b(h, 0); /* version (1 byte) + flags (3 byte) */
add_u32b(h, 0); /* Component type */
add_name(h, "soun"); /* Component subtype */
add_u32b(h, 0); /* Component manufacturer */
add_u32b(h, 0); /* Component flags */
add_u32b(h, 0); /* Component flags mask */
add_u16b(h, 0); /* Component name */
}
static void add_mdhd(m4a_header_t* h) {
add_atom(h, "mdhd", 0x20);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 0); /* Creation time */
add_u32b(h, 0); /* Modification time */
add_u32b(h, h->ktac->sample_rate); /* Time scale */
add_u32b(h, h->ktac->num_samples); /* Duration */
add_u16b(h, 0); /* Language (0xC455=eng?) */
add_u16b(h, 0); /* Quality */
}
static void add_mdia(m4a_header_t* h) {
m4a_state_t s;
save_atom(h, &s);
add_atom(h, "mdia", 0x00);
add_mdhd(h);
add_hdlr(h);
add_minf(h);
load_atom(h, &s);
}
static void add_tkhd(m4a_header_t* h) {
add_atom(h, "tkhd", 0x5C);
add_u32b(h, 0x00000001); /* Version (1 byte) + Flags (3 byte), 1=track enabled */
add_u32b(h, 0); /* Creation time */
add_u32b(h, 0); /* Modification time */
add_u32b(h, 1); /* Track ID */
add_u32b(h, 0); /* Reserved 1 */
add_u32b(h, h->ktac->num_samples); /* Duration */
add_u32b(h, 0); /* Reserved 1 */
add_u32b(h, 0); /* Reserved 2 */
add_u16b(h, 0); /* Layer */
add_u16b(h, 0); /* Alternate group (1?) */
add_u16b(h, 0x0100); /* Volume */
add_u16b(h, 0); /* Reserved */
add_u32b(h, 0x00010000); /* matrix_A */
add_u32b(h, 0); /* matrix_B */
add_u32b(h, 0); /* matrix_U */
add_u32b(h, 0); /* matrix_C */
add_u32b(h, 0x00010000); /* matrix_D */
add_u32b(h, 0); /* matrix_V */
add_u32b(h, 0); /* matrix_X */
add_u32b(h, 0); /* matrix_Y */
add_u32b(h, 0x40000000); /* matrix_W */
add_u32b(h, 0); /* Width */
add_u32b(h, 0); /* Height */
}
static void add_trak(m4a_header_t* h) {
m4a_state_t s;
save_atom(h, &s);
add_atom(h, "trak", 0x00);
add_tkhd(h);
add_mdia(h);
load_atom(h, &s);
}
static void add_mvhd(m4a_header_t* h) {
add_atom(h, "mvhd", 0x6c);
add_u32b(h, 0); /* Version (1 byte) + Flags (3 byte) */
add_u32b(h, 0); /* Creation time */
add_u32b(h, 0); /* Modification time */
add_u32b(h, h->ktac->sample_rate); /* Time scale */
add_u32b(h, h->ktac->num_samples); /* Duration */
add_u32b(h, 0x00010000); /* Preferred rate */
add_u16b(h, 0x0100); /* Preferred volume */
add_u32b(h, 0); /* Reserved 1 */
add_u32b(h, 0); /* Reserved 2 */
add_u16b(h, 0); /* Reserved 3 */
add_u32b(h, 0x00010000); /* matrix_A */
add_u32b(h, 0); /* matrix_B */
add_u32b(h, 0); /* matrix_U */
add_u32b(h, 0); /* matrix_C */
add_u32b(h, 0x00010000); /* matrix_D */
add_u32b(h, 0); /* matrix_V */
add_u32b(h, 0); /* matrix_X */
add_u32b(h, 0); /* matrix_Y */
add_u32b(h, 0x40000000); /* matrix_W */
add_u32b(h, 0); /* Preview time */
add_u32b(h, 0); /* Preview duration */
add_u32b(h, 0); /* Poster time */
add_u32b(h, 0); /* Selection time */
add_u32b(h, 0); /* Selection duration */
add_u32b(h, 0); /* Current time */
add_u32b(h, 2); /* Next track ID */
}
static void add_moov(m4a_header_t* h) {
m4a_state_t s;
save_atom(h, &s);
add_atom(h, "moov", 0x00);
add_mvhd(h);
add_trak(h);
//add_udta(h);
load_atom(h, &s);
}
/* *** */
static int make_m4a_header(uint8_t* buf, int buf_len, ktac_header_t* ktac, STREAMFILE* sf) {
m4a_header_t h = {0};
if (buf_len < 0x300 + ktac->table_entries * 0x4) /* approx */
goto fail;
h.sf = sf;
h.ktac = ktac;
h.out = buf;
add_ftyp(&h);
add_free(&h);
add_moov(&h);
add_mdat(&h);
/* define absolute chunk offset after all calcs */
put_u32be(h.chunks.out, h.bytes);
return h.bytes;
fail:
return 0;
}

View File

@ -938,4 +938,6 @@ VGMSTREAM* init_vgmstream_acx(STREAMFILE* sf);
VGMSTREAM* init_vgmstream_compresswave(STREAMFILE* sf);
VGMSTREAM* init_vgmstream_ktac(STREAMFILE* sf);
#endif /*_META_H*/

View File

@ -518,6 +518,7 @@ VGMSTREAM* (*init_vgmstream_functions[])(STREAMFILE* sf) = {
init_vgmstream_ifs,
init_vgmstream_acx,
init_vgmstream_compresswave,
init_vgmstream_ktac,
/* lowest priority metas (should go after all metas, and TXTH should go before raw formats) */
init_vgmstream_txth, /* proper parsers should supersede TXTH, once added */

View File

@ -751,6 +751,7 @@ typedef enum {
meta_DSP_WIIADPCM,
meta_DSP_CWAC,
meta_COMPRESSWAVE,
meta_KTAC,
} meta_t;
/* standard WAVEFORMATEXTENSIBLE speaker positions */