har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge commit '86f4c59bd676672040b89d8fea4c9e3b59bfe7ab'

Browse Source 
* commit '86f4c59bd676672040b89d8fea4c9e3b59bfe7ab':
  twinvq: Split VQF-specific part from common TwinVQ decoder core

Conflicts:
	libavcodec/Makefile
	libavcodec/twinvq.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer 2013-08-06 16:36:25 +02:00
commit 7d03e60c12
4 changed files with 651 additions and 552 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
libavcodec/Makefile
View File

@ -433,7 +433,7 @@ OBJS-$(CONFIG_TSCC_DECODER) += tscc.o msrledec.o
OBJS-$(CONFIG_TSCC2_DECODER) += tscc2.o OBJS-$(CONFIG_TSCC2_DECODER) += tscc2.o
OBJS-$(CONFIG_TTA_DECODER) += tta.o ttadata.o OBJS-$(CONFIG_TTA_DECODER) += tta.o ttadata.o
OBJS-$(CONFIG_TTA_ENCODER) += ttaenc.o ttadata.o OBJS-$(CONFIG_TTA_ENCODER) += ttaenc.o ttadata.o
OBJS-$(CONFIG_TWINVQ_DECODER) += twinvq.o OBJS-$(CONFIG_TWINVQ_DECODER) += twinvqdec.o twinvq.o
OBJS-$(CONFIG_TXD_DECODER) += txd.o s3tc.o OBJS-$(CONFIG_TXD_DECODER) += txd.o s3tc.o
OBJS-$(CONFIG_ULTI_DECODER) += ulti.o OBJS-$(CONFIG_ULTI_DECODER) += ulti.o
OBJS-$(CONFIG_UTVIDEO_DECODER) += utvideodec.o utvideo.o OBJS-$(CONFIG_UTVIDEO_DECODER) += utvideodec.o utvideo.o

565
libavcodec/twinvq.c
View File

@ -25,229 +25,11 @@
#include "libavutil/channel_layout.h" #include "libavutil/channel_layout.h"
#include "libavutil/float_dsp.h" #include "libavutil/float_dsp.h"
#include "avcodec.h" #include "avcodec.h"
#include "get_bits.h"
#include "fft.h" #include "fft.h"
#include "internal.h" #include "internal.h"
#include "lsp.h" #include "lsp.h"
#include "sinewin.h" #include "sinewin.h"
#include "twinvq_data.h" #include "twinvq.h"
enum TwinVQFrameType {
TWINVQ_FT_SHORT = 0, ///< Short frame (divided in n sub-blocks)
TWINVQ_FT_MEDIUM, ///< Medium frame (divided in m<n sub-blocks)
TWINVQ_FT_LONG, ///< Long frame (single sub-block + PPC)
TWINVQ_FT_PPC, ///< Periodic Peak Component (part of the long frame)
};
#define TWINVQ_PPC_SHAPE_CB_SIZE 64
#define TWINVQ_PPC_SHAPE_LEN_MAX 60
#define TWINVQ_SUB_AMP_MAX 4500.0
#define TWINVQ_MULAW_MU 100.0
#define TWINVQ_GAIN_BITS 8
#define TWINVQ_AMP_MAX 13000.0
#define TWINVQ_SUB_GAIN_BITS 5
#define TWINVQ_WINDOW_TYPE_BITS 4
#define TWINVQ_PGAIN_MU 200
#define TWINVQ_LSP_COEFS_MAX 20
#define TWINVQ_LSP_SPLIT_MAX 4
#define TWINVQ_CHANNELS_MAX 2
#define TWINVQ_SUBBLOCKS_MAX 16
#define TWINVQ_BARK_N_COEF_MAX 4
/**
* Parameters and tables that are different for each frame type
*/
struct TwinVQFrameMode {
uint8_t sub; ///< Number subblocks in each frame
const uint16_t *bark_tab;
/** number of distinct bark scale envelope values */
uint8_t bark_env_size;
const int16_t *bark_cb; ///< codebook for the bark scale envelope (BSE)
uint8_t bark_n_coef;///< number of BSE CB coefficients to read
uint8_t bark_n_bit; ///< number of bits of the BSE coefs
//@{
/** main codebooks for spectrum data */
const int16_t *cb0;
const int16_t *cb1;
//@}
uint8_t cb_len_read; ///< number of spectrum coefficients to read
};
typedef struct TwinVQFrameData {
int window_type;
enum TwinVQFrameType ftype;
uint8_t main_coeffs[1024];
uint8_t ppc_coeffs[TWINVQ_PPC_SHAPE_LEN_MAX];
uint8_t gain_bits[TWINVQ_CHANNELS_MAX];
uint8_t sub_gain_bits[TWINVQ_CHANNELS_MAX * TWINVQ_SUBBLOCKS_MAX];
uint8_t bark1[TWINVQ_CHANNELS_MAX][TWINVQ_SUBBLOCKS_MAX][TWINVQ_BARK_N_COEF_MAX];
uint8_t bark_use_hist[TWINVQ_CHANNELS_MAX][TWINVQ_SUBBLOCKS_MAX];
uint8_t lpc_idx1[TWINVQ_CHANNELS_MAX];
uint8_t lpc_idx2[TWINVQ_CHANNELS_MAX][TWINVQ_LSP_SPLIT_MAX];
uint8_t lpc_hist_idx[TWINVQ_CHANNELS_MAX];
int p_coef[TWINVQ_CHANNELS_MAX];
int g_coef[TWINVQ_CHANNELS_MAX];
} TwinVQFrameData;
/**
* Parameters and tables that are different for every combination of
* bitrate/sample rate
*/
typedef struct TwinVQModeTab {
struct TwinVQFrameMode fmode[3]; ///< frame type-dependant parameters
uint16_t size; ///< frame size in samples
uint8_t n_lsp; ///< number of lsp coefficients
const float *lspcodebook;
/* number of bits of the different LSP CB coefficients */
uint8_t lsp_bit0;
uint8_t lsp_bit1;
uint8_t lsp_bit2;
uint8_t lsp_split; ///< number of CB entries for the LSP decoding
const int16_t *ppc_shape_cb; ///< PPC shape CB
/** number of the bits for the PPC period value */
uint8_t ppc_period_bit;
uint8_t ppc_shape_bit; ///< number of bits of the PPC shape CB coeffs
uint8_t ppc_shape_len; ///< size of PPC shape CB
uint8_t pgain_bit; ///< bits for PPC gain
/** constant for peak period to peak width conversion */
uint16_t peak_per2wid;
} TwinVQModeTab;
static const TwinVQModeTab mode_08_08 = {
{
{ 8, bark_tab_s08_64, 10, tab.fcb08s, 1, 5, tab.cb0808s0, tab.cb0808s1, 18 },
{ 2, bark_tab_m08_256, 20, tab.fcb08m, 2, 5, tab.cb0808m0, tab.cb0808m1, 16 },
{ 1, bark_tab_l08_512, 30, tab.fcb08l, 3, 6, tab.cb0808l0, tab.cb0808l1, 17 }
},
512, 12, tab.lsp08, 1, 5, 3, 3, tab.shape08, 8, 28, 20, 6, 40
};
static const TwinVQModeTab mode_11_08 = {
{
{ 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1108s0, tab.cb1108s1, 29 },
{ 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1108m0, tab.cb1108m1, 24 },
{ 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1108l0, tab.cb1108l1, 27 }
},
512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
};
static const TwinVQModeTab mode_11_10 = {
{
{ 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1110s0, tab.cb1110s1, 21 },
{ 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1110m0, tab.cb1110m1, 18 },
{ 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1110l0, tab.cb1110l1, 20 }
},
512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
};
static const TwinVQModeTab mode_16_16 = {
{
{ 8, bark_tab_s16_128, 10, tab.fcb16s, 1, 5, tab.cb1616s0, tab.cb1616s1, 16 },
{ 2, bark_tab_m16_512, 20, tab.fcb16m, 2, 5, tab.cb1616m0, tab.cb1616m1, 15 },
{ 1, bark_tab_l16_1024, 30, tab.fcb16l, 3, 6, tab.cb1616l0, tab.cb1616l1, 16 }
},
1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16, 9, 56, 60, 7, 180
};
static const TwinVQModeTab mode_22_20 = {
{
{ 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18 },
{ 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17 },
{ 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18 }
},
1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};
static const TwinVQModeTab mode_22_24 = {
{
{ 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15 },
{ 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14 },
{ 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15 }
},
1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};
static const TwinVQModeTab mode_22_32 = {
{
{ 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11 },
{ 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11 },
{ 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12 }
},
512, 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
};
static const TwinVQModeTab mode_44_40 = {
{
{ 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4440s0, tab.cb4440s1, 18 },
{ 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4440m0, tab.cb4440m1, 17 },
{ 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4440l0, tab.cb4440l1, 17 }
},
2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
};
static const TwinVQModeTab mode_44_48 = {
{
{ 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4448s0, tab.cb4448s1, 15 },
{ 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4448m0, tab.cb4448m1, 14 },
{ 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4448l0, tab.cb4448l1, 14 }
},
2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
};
typedef struct TwinVQContext {
AVCodecContext *avctx;
AVFloatDSPContext fdsp;
FFTContext mdct_ctx[3];
const TwinVQModeTab *mtab;
// history
float lsp_hist[2][20]; ///< LSP coefficients of the last frame
float bark_hist[3][2][40]; ///< BSE coefficients of last frame
// bitstream parameters
int16_t permut[4][4096];
uint8_t length[4][2]; ///< main codebook stride
uint8_t length_change[4];
uint8_t bits_main_spec[2][4][2]; ///< bits for the main codebook
int bits_main_spec_change[4];
int n_div[4];
float *spectrum;
float *curr_frame; ///< non-interleaved output
float *prev_frame; ///< non-interleaved previous frame
int last_block_pos[2];
int discarded_packets;
float *cos_tabs[3];
// scratch buffers
float *tmp_buf;
TwinVQFrameData bits;
} TwinVQContext;
/** @note not speed critical, hence not optimized */
static void memset_float(float *buf, float val, int size)
{
while (size--)
*buf++ = val;
}
/** /**
* Evaluate a single LPC amplitude spectrum envelope coefficient from the line * Evaluate a single LPC amplitude spectrum envelope coefficient from the line
@ -378,7 +160,8 @@ static void eval_lpcenv_2parts(TwinVQContext *tctx, enum TwinVQFrameType ftype,
interpolate(lpc + size / 2 - step + 1, lpc[size / 2], interpolate(lpc + size / 2 - step + 1, lpc[size / 2],
lpc[size / 2 - step], step); lpc[size / 2 - step], step);
memset_float(lpc + size - 2 * step + 1, lpc[size - 2 * step], 2 * step - 1); twinvq_memset_float(lpc + size - 2 * step + 1, lpc[size - 2 * step],
2 * step - 1);
} }
/** /**
@ -428,104 +211,6 @@ static void dequant(TwinVQContext *tctx, const uint8_t *cb_bits, float *out,
} }
} }
static inline float mulawinv(float y, float clip, float mu)
{
y = av_clipf(y / clip, -1, 1);
return clip * FFSIGN(y) * (exp(log(1 + mu) * fabs(y)) - 1) / mu;
}
/**
* Evaluate a * b / 400 rounded to the nearest integer. When, for example,
* a * b == 200 and the nearest integer is ill-defined, use a table to emulate
* the following broken float-based implementation used by the binary decoder:
*
* @code
* static int very_broken_op(int a, int b)
* {
* static float test; // Ugh, force gcc to do the division first...
*
* test = a / 400.0;
* return b * test + 0.5;
* }
* @endcode
*
* @note if this function is replaced by just ROUNDED_DIV(a * b, 400.0), the
* stddev between the original file (before encoding with Yamaha encoder) and
* the decoded output increases, which leads one to believe that the encoder
* expects exactly this broken calculation.
*/
static int very_broken_op(int a, int b)
{
int x = a * b + 200;
int size;
const uint8_t *rtab;
if (x % 400 || b % 5)
return x / 400;
x /= 400;
size = tabs[b / 5].size;
rtab = tabs[b / 5].tab;
return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size];
}
/**
* Sum to data a periodic peak of a given period, width and shape.
*
* @param period the period of the peak divised by 400.0
*/
static void add_peak(int period, int width, const float *shape,
float ppc_gain, float *speech, int len)
{
int i, j;
const float *shape_end = shape + len;
int center;
// First peak centered around zero
for (i = 0; i < width / 2; i++)
speech[i] += ppc_gain * *shape++;
for (i = 1; i < ROUNDED_DIV(len, width); i++) {
center = very_broken_op(period, i);
for (j = -width / 2; j < (width + 1) / 2; j++)
speech[j + center] += ppc_gain * *shape++;
}
// For the last block, be careful not to go beyond the end of the buffer
center = very_broken_op(period, i);
for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++)
speech[j + center] += ppc_gain * *shape++;
}
static void decode_ppc(TwinVQContext *tctx, int period_coef,
const float *shape, float ppc_gain, float *speech)
{
const TwinVQModeTab *mtab = tctx->mtab;
int isampf = tctx->avctx->sample_rate / 1000;
int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels);
int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf);
int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf);
int period_range = max_period - min_period;
// This is actually the period multiplied by 400. It is just linearly coded
// between its maximum and minimum value.
int period = min_period +
ROUNDED_DIV(period_coef * period_range,
(1 << mtab->ppc_period_bit) - 1);
int width;
if (isampf == 22 && ibps == 32) {
// For some unknown reason, NTT decided to code this case differently...
width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid,
400 * mtab->size);
} else
width = period * mtab->peak_per2wid / (400 * mtab->size);
add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
}
static void dec_gain(TwinVQContext *tctx, static void dec_gain(TwinVQContext *tctx,
enum TwinVQFrameType ftype, float *out) enum TwinVQFrameType ftype, float *out)
{ {
@ -539,17 +224,17 @@ static void dec_gain(TwinVQContext *tctx,
if (ftype == TWINVQ_FT_LONG) { if (ftype == TWINVQ_FT_LONG) {
for (i = 0; i < tctx->avctx->channels; i++) for (i = 0; i < tctx->avctx->channels; i++)
out[i] = (1.0 / (1 << 13)) * out[i] = (1.0 / (1 << 13)) *
mulawinv(step * 0.5 + step * bits->gain_bits[i], twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
TWINVQ_AMP_MAX, TWINVQ_MULAW_MU); TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);
} else { } else {
for (i = 0; i < tctx->avctx->channels; i++) { for (i = 0; i < tctx->avctx->channels; i++) {
float val = (1.0 / (1 << 23)) * float val = (1.0 / (1 << 23)) *
mulawinv(step * 0.5 + step * bits->gain_bits[i], twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
TWINVQ_AMP_MAX, TWINVQ_MULAW_MU); TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);
for (j = 0; j < sub; j++) for (j = 0; j < sub; j++)
out[i * sub + j] = out[i * sub + j] =
val * mulawinv(sub_step * 0.5 + val * twinvq_mulawinv(sub_step * 0.5 +
sub_step * bits->sub_gain_bits[i * sub + j], sub_step * bits->sub_gain_bits[i * sub + j],
TWINVQ_SUB_AMP_MAX, TWINVQ_MULAW_MU); TWINVQ_SUB_AMP_MAX, TWINVQ_MULAW_MU);
} }
@ -721,34 +406,6 @@ static void imdct_output(TwinVQContext *tctx, enum TwinVQFrameType ftype,
} }
} }
static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist,
int ch, float *out, float gain,
enum TwinVQFrameType ftype)
{
const TwinVQModeTab *mtab = tctx->mtab;
int i, j;
float *hist = tctx->bark_hist[ftype][ch];
float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype];
int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
int idx = 0;
for (i = 0; i < fw_cb_len; i++)
for (j = 0; j < bark_n_coef; j++, idx++) {
float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] *
(1.0 / 4096);
float st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
: tmp2 + 1.0;
hist[idx] = tmp2;
if (st < -1.0)
st = 1.0;
memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
out += mtab->fmode[ftype].bark_tab[idx];
}
}
static void read_and_decode_spectrum(TwinVQContext *tctx, float *out, static void read_and_decode_spectrum(TwinVQContext *tctx, float *out,
enum TwinVQFrameType ftype) enum TwinVQFrameType ftype)
{ {
@ -782,7 +439,8 @@ static void read_and_decode_spectrum(TwinVQContext *tctx, float *out,
float lsp[TWINVQ_LSP_COEFS_MAX]; float lsp[TWINVQ_LSP_COEFS_MAX];
for (j = 0; j < sub; j++) { for (j = 0; j < sub; j++) {
dec_bark_env(tctx, bits->bark1[i][j], bits->bark_use_hist[i][j], i, tctx->dec_bark_env(tctx, bits->bark1[i][j],
bits->bark_use_hist[i][j], i,
tctx->tmp_buf, gain[sub * i + j], ftype); tctx->tmp_buf, gain[sub * i + j], ftype);
tctx->fdsp.vector_fmul(chunk + block_size * j, tctx->fdsp.vector_fmul(chunk + block_size * j,
@ -790,16 +448,9 @@ static void read_and_decode_spectrum(TwinVQContext *tctx, float *out,
tctx->tmp_buf, block_size); tctx->tmp_buf, block_size);
} }
if (ftype == TWINVQ_FT_LONG) { if (ftype == TWINVQ_FT_LONG)
float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1); tctx->decode_ppc(tctx, bits->p_coef[i], bits->g_coef[i],
float v = 1.0 / 8192 * ppc_shape + i * mtab->ppc_shape_len, chunk);
mulawinv(pgain_step * bits->g_coef[i] +
pgain_step / 2,
25000.0, TWINVQ_PGAIN_MU);
decode_ppc(tctx, bits->p_coef[i],
ppc_shape + i * mtab->ppc_shape_len, v, chunk);
}
decode_lsp(tctx, bits->lpc_idx1[i], bits->lpc_idx2[i], decode_lsp(tctx, bits->lpc_idx1[i], bits->lpc_idx2[i],
bits->lpc_hist_idx[i], lsp, tctx->lsp_hist[i]); bits->lpc_hist_idx[i], lsp, tctx->lsp_hist[i]);
@ -813,93 +464,13 @@ static void read_and_decode_spectrum(TwinVQContext *tctx, float *out,
} }
} }
static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb, const enum TwinVQFrameType ff_twinvq_wtype_to_ftype_table[] = {
uint8_t *dst, enum TwinVQFrameType ftype)
{
int i;
for (i = 0; i < tctx->n_div[ftype]; i++) {
int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);
*dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]);
*dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]);
}
}
static const enum TwinVQFrameType wtype_to_ftype_table[] = {
TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_SHORT, TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_SHORT, TWINVQ_FT_LONG,
TWINVQ_FT_MEDIUM, TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_MEDIUM, TWINVQ_FT_MEDIUM, TWINVQ_FT_LONG, TWINVQ_FT_LONG, TWINVQ_FT_MEDIUM,
TWINVQ_FT_MEDIUM TWINVQ_FT_MEDIUM
}; };
static int twinvq_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx, int ff_twinvq_decode_frame(AVCodecContext *avctx, void *data,
const uint8_t *buf, int buf_size)
{
TwinVQFrameData *bits = &tctx->bits;
const TwinVQModeTab *mtab = tctx->mtab;
int channels = tctx->avctx->channels;
int sub;
GetBitContext gb;
int i, j, k;
init_get_bits(&gb, buf, buf_size * 8);
skip_bits(&gb, get_bits(&gb, 8));
bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
if (bits->window_type > 8) {
av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
return AVERROR_INVALIDDATA;
}
bits->ftype = wtype_to_ftype_table[tctx->bits.window_type];
sub = mtab->fmode[bits->ftype].sub;
read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
for (i = 0; i < channels; i++)
for (j = 0; j < sub; j++)
for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++)
bits->bark1[i][j][k] =
get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit);
for (i = 0; i < channels; i++)
for (j = 0; j < sub; j++)
bits->bark_use_hist[i][j] = get_bits1(&gb);
if (bits->ftype == TWINVQ_FT_LONG) {
for (i = 0; i < channels; i++)
bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
} else {
for (i = 0; i < channels; i++) {
bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
for (j = 0; j < sub; j++)
bits->sub_gain_bits[i * sub + j] =
get_bits(&gb, TWINVQ_SUB_GAIN_BITS);
}
}
for (i = 0; i < channels; i++) {
bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0);
bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1);
for (j = 0; j < mtab->lsp_split; j++)
bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
}
if (bits->ftype == TWINVQ_FT_LONG) {
read_cb_data(tctx, &gb, bits->ppc_coeffs, 3);
for (i = 0; i < channels; i++) {
bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit);
bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit);
}
}
return 0;
}
static int twinvq_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame_ptr, AVPacket *avpkt) int *got_frame_ptr, AVPacket *avpkt)
{ {
AVFrame *frame = data; AVFrame *frame = data;
@ -910,12 +481,6 @@ static int twinvq_decode_frame(AVCodecContext *avctx, void *data,
float **out = NULL; float **out = NULL;
int ret; int ret;
if (buf_size * 8 < avctx->bit_rate * mtab->size / avctx->sample_rate + 8) {
av_log(avctx, AV_LOG_ERROR,
"Frame too small (%d bytes). Truncated file?\n", buf_size);
return AVERROR(EINVAL);
}
/* get output buffer */ /* get output buffer */
if (tctx->discarded_packets >= 2) { if (tctx->discarded_packets >= 2) {
frame->nb_samples = mtab->size; frame->nb_samples = mtab->size;
@ -924,7 +489,7 @@ static int twinvq_decode_frame(AVCodecContext *avctx, void *data,
out = (float **)frame->extended_data; out = (float **)frame->extended_data;
} }
if ((ret = twinvq_read_bitstream(avctx, tctx, buf, buf_size)) < 0) if ((ret = tctx->read_bitstream(avctx, tctx, buf, buf_size)) < 0)
return ret; return ret;
read_and_decode_spectrum(tctx, tctx->spectrum, tctx->bits.ftype); read_and_decode_spectrum(tctx, tctx->spectrum, tctx->bits.ftype);
@ -1160,7 +725,7 @@ static av_cold void init_bitstream_params(TwinVQContext *tctx)
construct_perm_table(tctx, frametype); construct_perm_table(tctx, frametype);
} }
static av_cold int twinvq_decode_close(AVCodecContext *avctx) av_cold int ff_twinvq_decode_close(AVCodecContext *avctx)
{ {
TwinVQContext *tctx = avctx->priv_data; TwinVQContext *tctx = avctx->priv_data;
int i; int i;
@ -1178,114 +743,24 @@ static av_cold int twinvq_decode_close(AVCodecContext *avctx)
return 0; return 0;
} }
static av_cold int twinvq_decode_init(AVCodecContext *avctx) av_cold int ff_twinvq_decode_init(AVCodecContext *avctx)
{ {
int ret, isampf, ibps; int ret;
TwinVQContext *tctx = avctx->priv_data; TwinVQContext *tctx = avctx->priv_data;
tctx->avctx = avctx; tctx->avctx = avctx;
avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
if (!avctx->extradata || avctx->extradata_size < 12) {
av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
return AVERROR_INVALIDDATA;
}
avctx->channels = AV_RB32(avctx->extradata) + 1;
avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
isampf = AV_RB32(avctx->extradata + 8);
if (isampf < 8 || isampf > 44) {
av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
return AVERROR_INVALIDDATA;
}
switch (isampf) {
case 44:
avctx->sample_rate = 44100;
break;
case 22:
avctx->sample_rate = 22050;
break;
case 11:
avctx->sample_rate = 11025;
break;
default:
avctx->sample_rate = isampf * 1000;
break;
}
if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
avctx->channels);
return -1;
}
avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
: AV_CH_LAYOUT_STEREO;
ibps = avctx->bit_rate / (1000 * avctx->channels);
if (ibps > 255U) {
av_log(avctx, AV_LOG_ERROR, "unsupported per channel bitrate %dkbps\n", ibps);
return AVERROR_INVALIDDATA;
}
switch ((isampf << 8) + ibps) {
case (8 << 8) + 8:
tctx->mtab = &mode_08_08;
break;
case (11 << 8) + 8:
tctx->mtab = &mode_11_08;
break;
case (11 << 8) + 10:
tctx->mtab = &mode_11_10;
break;
case (16 << 8) + 16:
tctx->mtab = &mode_16_16;
break;
case (22 << 8) + 20:
tctx->mtab = &mode_22_20;
break;
case (22 << 8) + 24:
tctx->mtab = &mode_22_24;
break;
case (22 << 8) + 32:
tctx->mtab = &mode_22_32;
break;
case (44 << 8) + 40:
tctx->mtab = &mode_44_40;
break;
case (44 << 8) + 48:
tctx->mtab = &mode_44_48;
break;
default:
av_log(avctx, AV_LOG_ERROR,
"This version does not support %d kHz - %d kbit/s/ch mode.\n",
isampf, isampf);
return -1;
}
avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
if ((ret = init_mdct_win(tctx))) { if ((ret = init_mdct_win(tctx))) {
av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n"); av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
twinvq_decode_close(avctx); ff_twinvq_decode_close(avctx);
return ret; return ret;
} }
init_bitstream_params(tctx); init_bitstream_params(tctx);
memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist)); twinvq_memset_float(tctx->bark_hist[0][0], 0.1,
FF_ARRAY_ELEMS(tctx->bark_hist));
return 0; return 0;
} }
AVCodec ff_twinvq_decoder = {
.name = "twinvq",
.type = AVMEDIA_TYPE_AUDIO,
.id = AV_CODEC_ID_TWINVQ,
.priv_data_size = sizeof(TwinVQContext),
.init = twinvq_decode_init,
.close = twinvq_decode_close,
.decode = twinvq_decode_frame,
.capabilities = CODEC_CAP_DR1,
.long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
AV_SAMPLE_FMT_NONE },
};

198
libavcodec/twinvq.h Normal file
View File

@ -0,0 +1,198 @@
/*
* TwinVQ decoder
* Copyright (c) 2009 Vitor Sessak
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_TWINVQ_H
#define AVCODEC_TWINVQ_H
#include <math.h>
#include <stdint.h>
#include "libavutil/common.h"
#include "libavutil/float_dsp.h"
#include "avcodec.h"
#include "fft.h"
#include "internal.h"
enum TwinVQCodec {
TWINVQ_CODEC_VQF,
TWINVQ_CODEC_METASOUND,
};
enum TwinVQFrameType {
TWINVQ_FT_SHORT = 0, ///< Short frame (divided in n sub-blocks)
TWINVQ_FT_MEDIUM, ///< Medium frame (divided in m<n sub-blocks)
TWINVQ_FT_LONG, ///< Long frame (single sub-block + PPC)
TWINVQ_FT_PPC, ///< Periodic Peak Component (part of the long frame)
};
#define TWINVQ_PPC_SHAPE_CB_SIZE 64
#define TWINVQ_PPC_SHAPE_LEN_MAX 60
#define TWINVQ_SUB_AMP_MAX 4500.0
#define TWINVQ_MULAW_MU 100.0
#define TWINVQ_GAIN_BITS 8
#define TWINVQ_AMP_MAX 13000.0
#define TWINVQ_SUB_GAIN_BITS 5
#define TWINVQ_WINDOW_TYPE_BITS 4
#define TWINVQ_PGAIN_MU 200
#define TWINVQ_LSP_COEFS_MAX 20
#define TWINVQ_LSP_SPLIT_MAX 4
#define TWINVQ_CHANNELS_MAX 2
#define TWINVQ_SUBBLOCKS_MAX 16
#define TWINVQ_BARK_N_COEF_MAX 4
/**
* Parameters and tables that are different for each frame type
*/
struct TwinVQFrameMode {
uint8_t sub; ///< Number subblocks in each frame
const uint16_t *bark_tab;
/** number of distinct bark scale envelope values */
uint8_t bark_env_size;
const int16_t *bark_cb; ///< codebook for the bark scale envelope (BSE)
uint8_t bark_n_coef;///< number of BSE CB coefficients to read
uint8_t bark_n_bit; ///< number of bits of the BSE coefs
//@{
/** main codebooks for spectrum data */
const int16_t *cb0;
const int16_t *cb1;
//@}
uint8_t cb_len_read; ///< number of spectrum coefficients to read
};
typedef struct TwinVQFrameData {
int window_type;
enum TwinVQFrameType ftype;
uint8_t main_coeffs[1024];
uint8_t ppc_coeffs[TWINVQ_PPC_SHAPE_LEN_MAX];
uint8_t gain_bits[TWINVQ_CHANNELS_MAX];
uint8_t sub_gain_bits[TWINVQ_CHANNELS_MAX * TWINVQ_SUBBLOCKS_MAX];
uint8_t bark1[TWINVQ_CHANNELS_MAX][TWINVQ_SUBBLOCKS_MAX][TWINVQ_BARK_N_COEF_MAX];
uint8_t bark_use_hist[TWINVQ_CHANNELS_MAX][TWINVQ_SUBBLOCKS_MAX];
uint8_t lpc_idx1[TWINVQ_CHANNELS_MAX];
uint8_t lpc_idx2[TWINVQ_CHANNELS_MAX][TWINVQ_LSP_SPLIT_MAX];
uint8_t lpc_hist_idx[TWINVQ_CHANNELS_MAX];
int p_coef[TWINVQ_CHANNELS_MAX];
int g_coef[TWINVQ_CHANNELS_MAX];
} TwinVQFrameData;
/**
* Parameters and tables that are different for every combination of
* bitrate/sample rate
*/
typedef struct TwinVQModeTab {
struct TwinVQFrameMode fmode[3]; ///< frame type-dependant parameters
uint16_t size; ///< frame size in samples
uint8_t n_lsp; ///< number of lsp coefficients
const float *lspcodebook;
/* number of bits of the different LSP CB coefficients */
uint8_t lsp_bit0;
uint8_t lsp_bit1;
uint8_t lsp_bit2;
uint8_t lsp_split; ///< number of CB entries for the LSP decoding
const int16_t *ppc_shape_cb; ///< PPC shape CB
/** number of the bits for the PPC period value */
uint8_t ppc_period_bit;
uint8_t ppc_shape_bit; ///< number of bits of the PPC shape CB coeffs
uint8_t ppc_shape_len; ///< size of PPC shape CB
uint8_t pgain_bit; ///< bits for PPC gain
/** constant for peak period to peak width conversion */
uint16_t peak_per2wid;
} TwinVQModeTab;
typedef struct TwinVQContext {
AVCodecContext *avctx;
AVFloatDSPContext fdsp;
FFTContext mdct_ctx[3];
const TwinVQModeTab *mtab;
// history
float lsp_hist[2][20]; ///< LSP coefficients of the last frame
float bark_hist[3][2][40]; ///< BSE coefficients of last frame
// bitstream parameters
int16_t permut[4][4096];
uint8_t length[4][2]; ///< main codebook stride
uint8_t length_change[4];
uint8_t bits_main_spec[2][4][2]; ///< bits for the main codebook
int bits_main_spec_change[4];
int n_div[4];
float *spectrum;
float *curr_frame; ///< non-interleaved output
float *prev_frame; ///< non-interleaved previous frame
int last_block_pos[2];
int discarded_packets;
float *cos_tabs[3];
// scratch buffers
float *tmp_buf;
TwinVQFrameData bits;
enum TwinVQCodec codec;
int (*read_bitstream)(AVCodecContext *avctx, struct TwinVQContext *tctx,
const uint8_t *buf, int buf_size);
void (*dec_bark_env)(struct TwinVQContext *tctx, const uint8_t *in,
int use_hist, int ch, float *out, float gain,
enum TwinVQFrameType ftype);
void (*decode_ppc)(struct TwinVQContext *tctx, int period_coef, int g_coef,
const float *shape, float *speech);
} TwinVQContext;
extern const enum TwinVQFrameType ff_twinvq_wtype_to_ftype_table[];
/** @note not speed critical, hence not optimized */
static inline void twinvq_memset_float(float *buf, float val, int size)
{
while (size--)
*buf++ = val;
}
static inline float twinvq_mulawinv(float y, float clip, float mu)
{
y = av_clipf(y / clip, -1, 1);
return clip * FFSIGN(y) * (exp(log(1 + mu) * fabs(y)) - 1) / mu;
}
int ff_twinvq_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame_ptr, AVPacket *avpkt);
av_cold int ff_twinvq_decode_close(AVCodecContext *avctx);
av_cold int ff_twinvq_decode_init(AVCodecContext *avctx);
#endif /* AVCODEC_TWINVQ_DATA_H */

426
libavcodec/twinvqdec.c Normal file
View File

@ -0,0 +1,426 @@
/*
* TwinVQ decoder
* Copyright (c) 2009 Vitor Sessak
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <math.h>
#include <stdint.h>
#include "libavutil/channel_layout.h"
#include "avcodec.h"
#include "get_bits.h"
#include "internal.h"
#include "twinvq.h"
#include "twinvq_data.h"
static const TwinVQModeTab mode_08_08 = {
{
{ 8, bark_tab_s08_64, 10, tab.fcb08s, 1, 5, tab.cb0808s0, tab.cb0808s1, 18 },
{ 2, bark_tab_m08_256, 20, tab.fcb08m, 2, 5, tab.cb0808m0, tab.cb0808m1, 16 },
{ 1, bark_tab_l08_512, 30, tab.fcb08l, 3, 6, tab.cb0808l0, tab.cb0808l1, 17 }
},
512, 12, tab.lsp08, 1, 5, 3, 3, tab.shape08, 8, 28, 20, 6, 40
};
static const TwinVQModeTab mode_11_08 = {
{
{ 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1108s0, tab.cb1108s1, 29 },
{ 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1108m0, tab.cb1108m1, 24 },
{ 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1108l0, tab.cb1108l1, 27 }
},
512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
};
static const TwinVQModeTab mode_11_10 = {
{
{ 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1110s0, tab.cb1110s1, 21 },
{ 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1110m0, tab.cb1110m1, 18 },
{ 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1110l0, tab.cb1110l1, 20 }
},
512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90
};
static const TwinVQModeTab mode_16_16 = {
{
{ 8, bark_tab_s16_128, 10, tab.fcb16s, 1, 5, tab.cb1616s0, tab.cb1616s1, 16 },
{ 2, bark_tab_m16_512, 20, tab.fcb16m, 2, 5, tab.cb1616m0, tab.cb1616m1, 15 },
{ 1, bark_tab_l16_1024, 30, tab.fcb16l, 3, 6, tab.cb1616l0, tab.cb1616l1, 16 }
},
1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16, 9, 56, 60, 7, 180
};
static const TwinVQModeTab mode_22_20 = {
{
{ 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18 },
{ 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17 },
{ 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18 }
},
1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};
static const TwinVQModeTab mode_22_24 = {
{
{ 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15 },
{ 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14 },
{ 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15 }
},
1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};
static const TwinVQModeTab mode_22_32 = {
{
{ 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11 },
{ 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11 },
{ 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12 }
},
512, 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
};
static const TwinVQModeTab mode_44_40 = {
{
{ 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4440s0, tab.cb4440s1, 18 },
{ 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4440m0, tab.cb4440m1, 17 },
{ 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4440l0, tab.cb4440l1, 17 }
},
2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
};
static const TwinVQModeTab mode_44_48 = {
{
{ 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4448s0, tab.cb4448s1, 15 },
{ 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4448m0, tab.cb4448m1, 14 },
{ 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4448l0, tab.cb4448l1, 14 }
},
2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432
};
/**
* Evaluate a * b / 400 rounded to the nearest integer. When, for example,
* a * b == 200 and the nearest integer is ill-defined, use a table to emulate
* the following broken float-based implementation used by the binary decoder:
*
* @code
* static int very_broken_op(int a, int b)
* {
* static float test; // Ugh, force gcc to do the division first...
*
* test = a / 400.0;
* return b * test + 0.5;
* }
* @endcode
*
* @note if this function is replaced by just ROUNDED_DIV(a * b, 400.0), the
* stddev between the original file (before encoding with Yamaha encoder) and
* the decoded output increases, which leads one to believe that the encoder
* expects exactly this broken calculation.
*/
static int very_broken_op(int a, int b)
{
int x = a * b + 200;
int size;
const uint8_t *rtab;
if (x % 400 || b % 5)
return x / 400;
x /= 400;
size = tabs[b / 5].size;
rtab = tabs[b / 5].tab;
return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size];
}
/**
* Sum to data a periodic peak of a given period, width and shape.
*
* @param period the period of the peak divised by 400.0
*/
static void add_peak(int period, int width, const float *shape,
float ppc_gain, float *speech, int len)
{
int i, j;
const float *shape_end = shape + len;
int center;
// First peak centered around zero
for (i = 0; i < width / 2; i++)
speech[i] += ppc_gain * *shape++;
for (i = 1; i < ROUNDED_DIV(len, width); i++) {
center = very_broken_op(period, i);
for (j = -width / 2; j < (width + 1) / 2; j++)
speech[j + center] += ppc_gain * *shape++;
}
// For the last block, be careful not to go beyond the end of the buffer
center = very_broken_op(period, i);
for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++)
speech[j + center] += ppc_gain * *shape++;
}
static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef,
const float *shape, float *speech)
{
const TwinVQModeTab *mtab = tctx->mtab;
int isampf = tctx->avctx->sample_rate / 1000;
int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels);
int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf);
int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf);
int period_range = max_period - min_period;
float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1);
float ppc_gain = 1.0 / 8192 *
twinvq_mulawinv(pgain_step * g_coef +
pgain_step / 2,
25000.0, TWINVQ_PGAIN_MU);
// This is actually the period multiplied by 400. It is just linearly coded
// between its maximum and minimum value.
int period = min_period +
ROUNDED_DIV(period_coef * period_range,
(1 << mtab->ppc_period_bit) - 1);
int width;
if (isampf == 22 && ibps == 32) {
// For some unknown reason, NTT decided to code this case differently...
width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid,
400 * mtab->size);
} else
width = period * mtab->peak_per2wid / (400 * mtab->size);
add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
}
static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist,
int ch, float *out, float gain,
enum TwinVQFrameType ftype)
{
const TwinVQModeTab *mtab = tctx->mtab;
int i, j;
float *hist = tctx->bark_hist[ftype][ch];
float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype];
int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
int idx = 0;
for (i = 0; i < fw_cb_len; i++)
for (j = 0; j < bark_n_coef; j++, idx++) {
float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] *
(1.0 / 4096);
float st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
: tmp2 + 1.0;
hist[idx] = tmp2;
if (st < -1.0)
st = 1.0;
twinvq_memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
out += mtab->fmode[ftype].bark_tab[idx];
}
}
static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb,
uint8_t *dst, enum TwinVQFrameType ftype)
{
int i;
for (i = 0; i < tctx->n_div[ftype]; i++) {
int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);
*dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]);
*dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]);
}
}
static int twinvq_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
const uint8_t *buf, int buf_size)
{
TwinVQFrameData *bits = &tctx->bits;
const TwinVQModeTab *mtab = tctx->mtab;
int channels = tctx->avctx->channels;
int sub;
GetBitContext gb;
int i, j, k;
if (buf_size * 8 < avctx->bit_rate * mtab->size / avctx->sample_rate + 8) {
av_log(avctx, AV_LOG_ERROR,
"Frame too small (%d bytes). Truncated file?\n", buf_size);
return AVERROR(EINVAL);
}
init_get_bits(&gb, buf, buf_size * 8);
skip_bits(&gb, get_bits(&gb, 8));
bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
if (bits->window_type > 8) {
av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
return AVERROR_INVALIDDATA;
}
bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits.window_type];
sub = mtab->fmode[bits->ftype].sub;
read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
for (i = 0; i < channels; i++)
for (j = 0; j < sub; j++)
for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++)
bits->bark1[i][j][k] =
get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit);
for (i = 0; i < channels; i++)
for (j = 0; j < sub; j++)
bits->bark_use_hist[i][j] = get_bits1(&gb);
if (bits->ftype == TWINVQ_FT_LONG) {
for (i = 0; i < channels; i++)
bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
} else {
for (i = 0; i < channels; i++) {
bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
for (j = 0; j < sub; j++)
bits->sub_gain_bits[i * sub + j] = get_bits(&gb,
TWINVQ_SUB_GAIN_BITS);
}
}
for (i = 0; i < channels; i++) {
bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0);
bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1);
for (j = 0; j < mtab->lsp_split; j++)
bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
}
if (bits->ftype == TWINVQ_FT_LONG) {
read_cb_data(tctx, &gb, bits->ppc_coeffs, 3);
for (i = 0; i < channels; i++) {
bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit);
bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit);
}
}
return 0;
}
static av_cold int twinvq_decode_init(AVCodecContext *avctx)
{
int isampf, ibps;
TwinVQContext *tctx = avctx->priv_data;
if (!avctx->extradata || avctx->extradata_size < 12) {
av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
return AVERROR_INVALIDDATA;
}
avctx->channels = AV_RB32(avctx->extradata) + 1;
avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
isampf = AV_RB32(avctx->extradata + 8);
if (isampf < 8 || isampf > 44) {
av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
return AVERROR_INVALIDDATA;
}
switch (isampf) {
case 44:
avctx->sample_rate = 44100;
break;
case 22:
avctx->sample_rate = 22050;
break;
case 11:
avctx->sample_rate = 11025;
break;
default:
avctx->sample_rate = isampf * 1000;
break;
}
if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
avctx->channels);
return -1;
}
avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
: AV_CH_LAYOUT_STEREO;
ibps = avctx->bit_rate / (1000 * avctx->channels);
if (ibps > 255U) {
av_log(avctx, AV_LOG_ERROR, "unsupported per channel bitrate %dkbps\n", ibps);
return AVERROR_INVALIDDATA;
}
switch ((isampf << 8) + ibps) {
case (8 << 8) + 8:
tctx->mtab = &mode_08_08;
break;
case (11 << 8) + 8:
tctx->mtab = &mode_11_08;
break;
case (11 << 8) + 10:
tctx->mtab = &mode_11_10;
break;
case (16 << 8) + 16:
tctx->mtab = &mode_16_16;
break;
case (22 << 8) + 20:
tctx->mtab = &mode_22_20;
break;
case (22 << 8) + 24:
tctx->mtab = &mode_22_24;
break;
case (22 << 8) + 32:
tctx->mtab = &mode_22_32;
break;
case (44 << 8) + 40:
tctx->mtab = &mode_44_40;
break;
case (44 << 8) + 48:
tctx->mtab = &mode_44_48;
break;
default:
av_log(avctx, AV_LOG_ERROR,
"This version does not support %d kHz - %d kbit/s/ch mode.\n",
isampf, isampf);
return -1;
}
tctx->codec = TWINVQ_CODEC_VQF;
tctx->read_bitstream = twinvq_read_bitstream;
tctx->dec_bark_env = dec_bark_env;
tctx->decode_ppc = decode_ppc;
return ff_twinvq_decode_init(avctx);
}
AVCodec ff_twinvq_decoder = {
.name = "twinvq",
.type = AVMEDIA_TYPE_AUDIO,
.id = AV_CODEC_ID_TWINVQ,
.priv_data_size = sizeof(TwinVQContext),
.init = twinvq_decode_init,
.close = ff_twinvq_decode_close,
.decode = ff_twinvq_decode_frame,
.capabilities = CODEC_CAP_DR1,
.long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
AV_SAMPLE_FMT_NONE },
};