---
updated for HEAD and some comments were addressed
---
libavcodec/Makefile | 2 +-
libavcodec/twinvq.c | 667 +-----------------------------------------------
libavcodec/twinvq.h | 162 ++++++++++++
libavcodec/twinvqdec.c | 566 ++++++++++++++++++++++++++++++++++++++++
4 files changed, 736 insertions(+), 661 deletions(-)
create mode 100644 libavcodec/twinvq.h
create mode 100644 libavcodec/twinvqdec.c
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index d9c6e33..7203c64 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -355,7 +355,7 @@ OBJS-$(CONFIG_TRUESPEECH_DECODER) += truespeech.o
OBJS-$(CONFIG_TSCC_DECODER) += tscc.o msrledec.o
OBJS-$(CONFIG_TSCC2_DECODER) += tscc2.o
OBJS-$(CONFIG_TTA_DECODER) += tta.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_ULTI_DECODER) += ulti.o
OBJS-$(CONFIG_UTVIDEO_DECODER) += utvideodec.o utvideo.o
diff --git a/libavcodec/twinvq.c b/libavcodec/twinvq.c
index 8c304c6..a009801 100644
--- a/libavcodec/twinvq.c
+++ b/libavcodec/twinvq.c
@@ -22,209 +22,13 @@
#include <math.h>
#include <stdint.h>
-#include "libavutil/channel_layout.h"
#include "libavutil/float_dsp.h"
#include "avcodec.h"
-#include "get_bits.h"
#include "fft.h"
-#include "internal.h"
#include "lsp.h"
#include "sinewin.h"
-#include "twinvq_data.h"
-enum FrameType {
- FT_SHORT = 0, ///< Short frame (divided in n sub-blocks)
- FT_MEDIUM, ///< Medium frame (divided in m<n sub-blocks)
- FT_LONG, ///< Long frame (single sub-block + PPC)
- FT_PPC, ///< Periodic Peak Component (part of the long frame)
-};
-
-/**
- * Parameters and tables that are different for each frame type
- */
-struct FrameMode {
- 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
-};
-
-/**
- * Parameters and tables that are different for every combination of
- * bitrate/sample rate
- */
-typedef struct {
- struct FrameMode 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;
-} ModeTab;
-
-static const ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 TwinContext {
- AVCodecContext *avctx;
- AVFloatDSPContext fdsp;
- FFTContext mdct_ctx[3];
-
- const ModeTab *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;
-} TwinContext;
-
-#define PPC_SHAPE_CB_SIZE 64
-#define PPC_SHAPE_LEN_MAX 60
-#define SUB_AMP_MAX 4500.0
-#define MULAW_MU 100.0
-#define GAIN_BITS 8
-#define AMP_MAX 13000.0
-#define SUB_GAIN_BITS 5
-#define WINDOW_TYPE_BITS 4
-#define PGAIN_MU 200
-#define LSP_COEFS_MAX 20
-#define LSP_SPLIT_MAX 4
-#define CHANNELS_MAX 2
-#define SUBBLOCKS_MAX 16
-#define BARK_N_COEF_MAX 4
-
-/** @note not speed critical, hence not optimized */
-static void memset_float(float *buf, float val, int size)
-{
- while (size--)
- *buf++ = val;
-}
+#include "twinvq.h"
/**
* Evaluate a single LPC amplitude spectrum envelope coefficient from the line
@@ -359,182 +163,6 @@ static void eval_lpcenv_2parts(TwinContext *tctx, enum
FrameType ftype,
}
/**
- * Inverse quantization. Read CB coefficients for cb1 and cb2 from the
- * bitstream, sum the corresponding vectors and write the result to *out
- * after permutation.
- */
-static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
- enum FrameType ftype,
- const int16_t *cb0, const int16_t *cb1, int cb_len)
-{
- int pos = 0;
- int i, j;
-
- for (i = 0; i < tctx->n_div[ftype]; i++) {
- int tmp0, tmp1;
- int sign0 = 1;
- int sign1 = 1;
- const int16_t *tab0, *tab1;
- int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
- int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
-
- int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
- if (bits == 7) {
- if (get_bits1(gb))
- sign0 = -1;
- bits = 6;
- }
- tmp0 = get_bits(gb, bits);
-
- bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
-
- if (bits == 7) {
- if (get_bits1(gb))
- sign1 = -1;
-
- bits = 6;
- }
- tmp1 = get_bits(gb, bits);
-
- tab0 = cb0 + tmp0 * cb_len;
- tab1 = cb1 + tmp1 * cb_len;
-
- for (j = 0; j < length; j++)
- out[tctx->permut[ftype][pos + j]] = sign0 * tab0[j] +
- sign1 * tab1[j];
-
- pos += length;
- }
-}
-
-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(TwinContext *tctx, int period_coef, const float *shape,
- float ppc_gain, float *speech)
-{
- const ModeTab *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(TwinContext *tctx, GetBitContext *gb, enum FrameType
ftype,
- float *out)
-{
- const ModeTab *mtab = tctx->mtab;
- int i, j;
- int sub = mtab->fmode[ftype].sub;
- float step = AMP_MAX / ((1 << GAIN_BITS) - 1);
- float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
-
- if (ftype == FT_LONG) {
- for (i = 0; i < tctx->avctx->channels; i++)
- out[i] = (1.0 / (1 << 13)) *
- mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
- AMP_MAX, MULAW_MU);
- } else {
- for (i = 0; i < tctx->avctx->channels; i++) {
- float val = (1.0 / (1 << 23)) *
- mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
- AMP_MAX, MULAW_MU);
-
- for (j = 0; j < sub; j++)
- out[i * sub + j] =
- val * mulawinv(sub_step * 0.5 +
- sub_step * get_bits(gb, SUB_GAIN_BITS),
- SUB_AMP_MAX, MULAW_MU);
- }
- }
-}
-
-/**
* Rearrange the LSP coefficients so that they have a minimum distance of
* min_dist. This function does it exactly as described in section of 3.2.4
* of the G.729 specification (but interestingly is different from what the
@@ -553,7 +181,7 @@ static void rearrange_lsp(int order, float *lsp, float
min_dist)
}
}
-static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
+void ff_tvq_decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
int lpc_hist_idx, float *lsp, float *hist)
{
const ModeTab *mtab = tctx->mtab;
@@ -593,7 +221,7 @@ static void decode_lsp(TwinContext *tctx, int lpc_idx1,
uint8_t *lpc_idx2,
ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp);
}
-static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
+void ff_tvq_dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
enum FrameType ftype, float *lpc)
{
int i;
@@ -668,7 +296,7 @@ static void imdct_and_window(TwinContext *tctx, enum
FrameType ftype, int wtype,
tctx->last_block_pos[ch] = (size + first_wsize) / 2;
}
-static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
+void ff_tvq_imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
float **out)
{
const ModeTab *mtab = tctx->mtab;
@@ -699,184 +327,10 @@ static void imdct_output(TwinContext *tctx, enum
FrameType ftype, int wtype,
}
}
-static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
- int ch, float *out, float gain, enum FrameType ftype)
-{
- const ModeTab *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(TwinContext *tctx, GetBitContext *gb,
- float *out, enum FrameType ftype)
-{
- const ModeTab *mtab = tctx->mtab;
- int channels = tctx->avctx->channels;
- int sub = mtab->fmode[ftype].sub;
- int block_size = mtab->size / sub;
- float gain[CHANNELS_MAX * SUBBLOCKS_MAX];
- float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
- uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX];
- uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
-
- uint8_t lpc_idx1[CHANNELS_MAX];
- uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX];
- uint8_t lpc_hist_idx[CHANNELS_MAX];
-
- int i, j, k;
-
- dequant(tctx, gb, out, ftype,
- mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
- mtab->fmode[ftype].cb_len_read);
-
- for (i = 0; i < channels; i++)
- for (j = 0; j < sub; j++)
- for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
- bark1[i][j][k] =
- get_bits(gb, mtab->fmode[ftype].bark_n_bit);
-
- for (i = 0; i < channels; i++)
- for (j = 0; j < sub; j++)
- bark_use_hist[i][j] = get_bits1(gb);
-
- dec_gain(tctx, gb, ftype, gain);
-
- for (i = 0; i < channels; i++) {
- lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
- lpc_idx1[i] = get_bits(gb, tctx->mtab->lsp_bit1);
-
- for (j = 0; j < tctx->mtab->lsp_split; j++)
- lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
- }
-
- if (ftype == FT_LONG) {
- int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len * channels - 1) /
- tctx->n_div[3];
- dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
- mtab->ppc_shape_cb + cb_len_p * PPC_SHAPE_CB_SIZE, cb_len_p);
- }
-
- for (i = 0; i < channels; i++) {
- float *chunk = out + mtab->size * i;
- float lsp[LSP_COEFS_MAX];
-
- for (j = 0; j < sub; j++) {
- dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
- tctx->tmp_buf, gain[sub * i + j], ftype);
-
- tctx->fdsp.vector_fmul(chunk + block_size * j,
- chunk + block_size * j,
- tctx->tmp_buf, block_size);
- }
-
- if (ftype == FT_LONG) {
- float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1);
- int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
- int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
- float v = 1.0 / 8192 *
- mulawinv(pgain_step * g_coef + pgain_step / 2,
- 25000.0, PGAIN_MU);
-
- decode_ppc(tctx, p_coef, ppc_shape + i * mtab->ppc_shape_len, v,
- chunk);
- }
-
- decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
- tctx->lsp_hist[i]);
-
- dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
-
- for (j = 0; j < mtab->fmode[ftype].sub; j++) {
- tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
- chunk += block_size;
- }
- }
-}
-
-static int twin_decode_frame(AVCodecContext *avctx, void *data,
- int *got_frame_ptr, AVPacket *avpkt)
-{
- AVFrame *frame = data;
- const uint8_t *buf = avpkt->data;
- int buf_size = avpkt->size;
- TwinContext *tctx = avctx->priv_data;
- GetBitContext gb;
- const ModeTab *mtab = tctx->mtab;
- float **out = NULL;
- enum FrameType ftype;
- int window_type, ret;
- static const enum FrameType wtype_to_ftype_table[] = {
- FT_LONG, FT_LONG, FT_SHORT, FT_LONG,
- FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM
- };
-
- 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 */
- if (tctx->discarded_packets >= 2) {
- frame->nb_samples = mtab->size;
- if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
- av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
- return ret;
- }
- out = (float **)frame->extended_data;
- }
-
- init_get_bits(&gb, buf, buf_size * 8);
- skip_bits(&gb, get_bits(&gb, 8));
- window_type = get_bits(&gb, WINDOW_TYPE_BITS);
-
- if (window_type > 8) {
- av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
- return -1;
- }
-
- ftype = wtype_to_ftype_table[window_type];
-
- read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
-
- imdct_output(tctx, ftype, window_type, out);
-
- FFSWAP(float *, tctx->curr_frame, tctx->prev_frame);
-
- if (tctx->discarded_packets < 2) {
- tctx->discarded_packets++;
- *got_frame_ptr = 0;
- return buf_size;
- }
-
- *got_frame_ptr = 1;
-
- return buf_size;
-}
-
/**
* Init IMDCT and windowing tables
*/
-static av_cold int init_mdct_win(TwinContext *tctx)
+av_cold int ff_tvq_init_mdct_win(TwinContext *tctx)
{
int i, j, ret;
const ModeTab *mtab = tctx->mtab;
@@ -1019,7 +473,7 @@ static av_cold void construct_perm_table(TwinContext *tctx,
size * block_size);
}
-static av_cold void init_bitstream_params(TwinContext *tctx)
+av_cold void ff_tvq_init_bitstream_params(TwinContext *tctx)
{
const ModeTab *mtab = tctx->mtab;
int n_ch = tctx->avctx->channels;
@@ -1088,7 +542,7 @@ static av_cold void init_bitstream_params(TwinContext
*tctx)
construct_perm_table(tctx, frametype);
}
-static av_cold int twin_decode_close(AVCodecContext *avctx)
+av_cold int ff_tvq_decode_close(AVCodecContext *avctx)
{
TwinContext *tctx = avctx->priv_data;
int i;
@@ -1105,110 +559,3 @@ static av_cold int twin_decode_close(AVCodecContext
*avctx)
return 0;
}
-
-static av_cold int twin_decode_init(AVCodecContext *avctx)
-{
- int ret, isampf, ibps;
- TwinContext *tctx = avctx->priv_data;
-
- tctx->avctx = avctx;
- 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 > 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);
-
- 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);
- if ((ret = init_mdct_win(tctx))) {
- av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
- twin_decode_close(avctx);
- return ret;
- }
- init_bitstream_params(tctx);
-
- memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));
-
- return 0;
-}
-
-AVCodec ff_twinvq_decoder = {
- .name = "twinvq",
- .type = AVMEDIA_TYPE_AUDIO,
- .id = AV_CODEC_ID_TWINVQ,
- .priv_data_size = sizeof(TwinContext),
- .init = twin_decode_init,
- .close = twin_decode_close,
- .decode = twin_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 },
-};
diff --git a/libavcodec/twinvq.h b/libavcodec/twinvq.h
new file mode 100644
index 0000000..1332007
--- /dev/null
+++ b/libavcodec/twinvq.h
@@ -0,0 +1,162 @@
+/*
+ * TwinVQ decoder
+ * Copyright (c) 2009 Vitor Sessak
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; 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"
+
+enum FrameType {
+ FT_SHORT = 0, ///< Short frame (divided in n sub-blocks)
+ FT_MEDIUM, ///< Medium frame (divided in m<n sub-blocks)
+ FT_LONG, ///< Long frame (single sub-block + PPC)
+ FT_PPC, ///< Periodic Peak Component (part of the long frame)
+};
+
+/**
+ * Parameters and tables that are different for each frame type
+ */
+struct FrameMode {
+ 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
+};
+
+/**
+ * Parameters and tables that are different for every combination of
+ * bitrate/sample rate
+ */
+typedef struct {
+ struct FrameMode 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;
+} ModeTab;
+
+typedef struct TwinContext {
+ AVCodecContext *avctx;
+ AVFloatDSPContext fdsp;
+ FFTContext mdct_ctx[3];
+
+ const ModeTab *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;
+} TwinContext;
+
+#define PPC_SHAPE_CB_SIZE 64
+#define PPC_SHAPE_LEN_MAX 60
+#define SUB_AMP_MAX 4500.0
+#define MULAW_MU 100.0
+#define GAIN_BITS 8
+#define AMP_MAX 13000.0
+#define SUB_GAIN_BITS 5
+#define WINDOW_TYPE_BITS 4
+#define PGAIN_MU 200
+#define LSP_COEFS_MAX 20
+#define LSP_SPLIT_MAX 4
+#define CHANNELS_MAX 2
+#define SUBBLOCKS_MAX 16
+#define BARK_N_COEF_MAX 4
+
+/** @note not speed critical, hence not optimized */
+static inline void memset_float(float *buf, float val, int size)
+{
+ while (size--)
+ *buf++ = val;
+}
+
+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;
+}
+
+void ff_tvq_decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
+ int lpc_hist_idx, float *lsp, float *hist);
+void ff_tvq_dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
+ enum FrameType ftype, float *lpc);
+void ff_tvq_imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
+ float **out);
+av_cold int ff_tvq_init_mdct_win(TwinContext *tctx);
+av_cold void ff_tvq_init_bitstream_params(TwinContext *tctx);
+av_cold int ff_tvq_decode_close(AVCodecContext *avctx);
+
+#endif /* AVCODEC_TWINVQ_H */
diff --git a/libavcodec/twinvqdec.c b/libavcodec/twinvqdec.c
new file mode 100644
index 0000000..7de1228
--- /dev/null
+++ b/libavcodec/twinvqdec.c
@@ -0,0 +1,566 @@
+/*
+ * TwinVQ decoder
+ * Copyright (c) 2009 Vitor Sessak
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; 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 "libavutil/float_dsp.h"
+#include "avcodec.h"
+#include "get_bits.h"
+#include "fft.h"
+#include "internal.h"
+#include "lsp.h"
+#include "sinewin.h"
+#include "twinvq.h"
+#include "twinvq_data.h"
+
+static const ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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 ModeTab 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
+};
+
+/**
+ * Inverse quantization. Read CB coefficients for cb1 and cb2 from the
+ * bitstream, sum the corresponding vectors and write the result to *out
+ * after permutation.
+ */
+static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
+ enum FrameType ftype,
+ const int16_t *cb0, const int16_t *cb1, int cb_len)
+{
+ int pos = 0;
+ int i, j;
+
+ for (i = 0; i < tctx->n_div[ftype]; i++) {
+ int tmp0, tmp1;
+ int sign0 = 1;
+ int sign1 = 1;
+ const int16_t *tab0, *tab1;
+ int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
+ int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
+
+ int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
+ if (bits == 7) {
+ if (get_bits1(gb))
+ sign0 = -1;
+ bits = 6;
+ }
+ tmp0 = get_bits(gb, bits);
+
+ bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
+
+ if (bits == 7) {
+ if (get_bits1(gb))
+ sign1 = -1;
+
+ bits = 6;
+ }
+ tmp1 = get_bits(gb, bits);
+
+ tab0 = cb0 + tmp0 * cb_len;
+ tab1 = cb1 + tmp1 * cb_len;
+
+ for (j = 0; j < length; j++)
+ out[tctx->permut[ftype][pos + j]] = sign0 * tab0[j] +
+ sign1 * tab1[j];
+
+ pos += length;
+ }
+}
+
+/**
+ * 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(TwinContext *tctx, int period_coef, const float *shape,
+ float ppc_gain, float *speech)
+{
+ const ModeTab *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(TwinContext *tctx, GetBitContext *gb, enum FrameType
ftype,
+ float *out)
+{
+ const ModeTab *mtab = tctx->mtab;
+ int i, j;
+ int sub = mtab->fmode[ftype].sub;
+ float step = AMP_MAX / ((1 << GAIN_BITS) - 1);
+ float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
+
+ if (ftype == FT_LONG) {
+ for (i = 0; i < tctx->avctx->channels; i++)
+ out[i] = (1.0 / (1 << 13)) *
+ mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
+ AMP_MAX, MULAW_MU);
+ } else {
+ for (i = 0; i < tctx->avctx->channels; i++) {
+ float val = (1.0 / (1 << 23)) *
+ mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
+ AMP_MAX, MULAW_MU);
+
+ for (j = 0; j < sub; j++)
+ out[i * sub + j] =
+ val * mulawinv(sub_step * 0.5 +
+ sub_step * get_bits(gb, SUB_GAIN_BITS),
+ SUB_AMP_MAX, MULAW_MU);
+ }
+ }
+}
+
+static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
+ int ch, float *out, float gain, enum FrameType ftype)
+{
+ const ModeTab *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(TwinContext *tctx, GetBitContext *gb,
+ float *out, enum FrameType ftype)
+{
+ const ModeTab *mtab = tctx->mtab;
+ int channels = tctx->avctx->channels;
+ int sub = mtab->fmode[ftype].sub;
+ int block_size = mtab->size / sub;
+ float gain[CHANNELS_MAX * SUBBLOCKS_MAX];
+ float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
+ uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX];
+ uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
+
+ uint8_t lpc_idx1[CHANNELS_MAX];
+ uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX];
+ uint8_t lpc_hist_idx[CHANNELS_MAX];
+
+ int i, j, k;
+
+ dequant(tctx, gb, out, ftype,
+ mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
+ mtab->fmode[ftype].cb_len_read);
+
+ for (i = 0; i < channels; i++)
+ for (j = 0; j < sub; j++)
+ for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
+ bark1[i][j][k] =
+ get_bits(gb, mtab->fmode[ftype].bark_n_bit);
+
+ for (i = 0; i < channels; i++)
+ for (j = 0; j < sub; j++)
+ bark_use_hist[i][j] = get_bits1(gb);
+
+ dec_gain(tctx, gb, ftype, gain);
+
+ for (i = 0; i < channels; i++) {
+ lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
+ lpc_idx1[i] = get_bits(gb, tctx->mtab->lsp_bit1);
+
+ for (j = 0; j < tctx->mtab->lsp_split; j++)
+ lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
+ }
+
+ if (ftype == FT_LONG) {
+ int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len * channels - 1) /
+ tctx->n_div[3];
+ dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
+ mtab->ppc_shape_cb + cb_len_p * PPC_SHAPE_CB_SIZE, cb_len_p);
+ }
+
+ for (i = 0; i < channels; i++) {
+ float *chunk = out + mtab->size * i;
+ float lsp[LSP_COEFS_MAX];
+
+ for (j = 0; j < sub; j++) {
+ dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
+ tctx->tmp_buf, gain[sub * i + j], ftype);
+
+ tctx->fdsp.vector_fmul(chunk + block_size * j,
+ chunk + block_size * j,
+ tctx->tmp_buf, block_size);
+ }
+
+ if (ftype == FT_LONG) {
+ float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1);
+ int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
+ int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
+ float v = 1.0 / 8192 *
+ mulawinv(pgain_step * g_coef + pgain_step / 2,
+ 25000.0, PGAIN_MU);
+
+ decode_ppc(tctx, p_coef, ppc_shape + i * mtab->ppc_shape_len, v,
+ chunk);
+ }
+
+ ff_tvq_decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
+ tctx->lsp_hist[i]);
+
+ ff_tvq_dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
+
+ for (j = 0; j < mtab->fmode[ftype].sub; j++) {
+ tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
+ chunk += block_size;
+ }
+ }
+}
+
+static int twin_decode_frame(AVCodecContext *avctx, void *data,
+ int *got_frame_ptr, AVPacket *avpkt)
+{
+ AVFrame *frame = data;
+ const uint8_t *buf = avpkt->data;
+ int buf_size = avpkt->size;
+ TwinContext *tctx = avctx->priv_data;
+ GetBitContext gb;
+ const ModeTab *mtab = tctx->mtab;
+ float **out = NULL;
+ enum FrameType ftype;
+ int window_type, ret;
+ static const enum FrameType wtype_to_ftype_table[] = {
+ FT_LONG, FT_LONG, FT_SHORT, FT_LONG,
+ FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM
+ };
+
+ 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 */
+ if (tctx->discarded_packets >= 2) {
+ frame->nb_samples = mtab->size;
+ if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
+ av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+ return ret;
+ }
+ out = (float **)frame->extended_data;
+ }
+
+ init_get_bits(&gb, buf, buf_size * 8);
+ skip_bits(&gb, get_bits(&gb, 8));
+ window_type = get_bits(&gb, WINDOW_TYPE_BITS);
+
+ if (window_type > 8) {
+ av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
+ return -1;
+ }
+
+ ftype = wtype_to_ftype_table[window_type];
+
+ read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
+
+ ff_tvq_imdct_output(tctx, ftype, window_type, out);
+
+ FFSWAP(float *, tctx->curr_frame, tctx->prev_frame);
+
+ if (tctx->discarded_packets < 2) {
+ tctx->discarded_packets++;
+ *got_frame_ptr = 0;
+ return buf_size;
+ }
+
+ *got_frame_ptr = 1;
+
+ return buf_size;
+}
+
+static av_cold int twin_decode_init(AVCodecContext *avctx)
+{
+ int ret, isampf, ibps;
+ TwinContext *tctx = avctx->priv_data;
+
+ tctx->avctx = avctx;
+ 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 > 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);
+
+ 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);
+ if ((ret = ff_tvq_init_mdct_win(tctx))) {
+ av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
+ ff_tvq_decode_close(avctx);
+ return ret;
+ }
+ ff_tvq_init_bitstream_params(tctx);
+
+ memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));
+
+ return 0;
+}
+
+AVCodec ff_twinvq_decoder = {
+ .name = "twinvq",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .id = AV_CODEC_ID_TWINVQ,
+ .priv_data_size = sizeof(TwinContext),
+ .init = twin_decode_init,
+ .close = ff_tvq_decode_close,
+ .decode = twin_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 },
+};
--
1.7.9.5
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
