#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 TwinVQFrameType {
- 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)
-};
-
-#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
-
-/**
- * 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[PPC_SHAPE_LEN_MAX];
-
- uint8_t gain_bits[CHANNELS_MAX];
- uint8_t sub_gain_bits[CHANNELS_MAX * SUBBLOCKS_MAX];
-
- 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 p_coef[CHANNELS_MAX];
- int g_coef[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;
-}
+#include "twinvq.h"
/**
* Evaluate a single LPC amplitude spectrum envelope coefficient from the line
{
int i;
const TwinVQModeTab *mtab = tctx->mtab;
- int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
+ int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub;
for (i = 0; i < size_s / 2; i++) {
float cos_i = tctx->cos_tabs[0][i];
interpolate(lpc + size / 2 - step + 1, lpc[size / 2],
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);
}
/**
}
}
-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,
enum TwinVQFrameType ftype, float *out)
{
const TwinVQModeTab *mtab = tctx->mtab;
- const TwinVQFrameData *bits = &tctx->bits;
+ const TwinVQFrameData *bits = &tctx->bits[tctx->cur_frame];
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);
+ float step = TWINVQ_AMP_MAX / ((1 << TWINVQ_GAIN_BITS) - 1);
+ float sub_step = TWINVQ_SUB_AMP_MAX / ((1 << TWINVQ_SUB_GAIN_BITS) - 1);
- if (ftype == FT_LONG) {
+ if (ftype == TWINVQ_FT_LONG) {
for (i = 0; i < tctx->avctx->channels; i++)
out[i] = (1.0 / (1 << 13)) *
- mulawinv(step * 0.5 + step * bits->gain_bits[i],
- AMP_MAX, MULAW_MU);
+ twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
+ TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);
} else {
for (i = 0; i < tctx->avctx->channels; i++) {
float val = (1.0 / (1 << 23)) *
- mulawinv(step * 0.5 + step * bits->gain_bits[i],
- AMP_MAX, MULAW_MU);
+ twinvq_mulawinv(step * 0.5 + step * bits->gain_bits[i],
+ TWINVQ_AMP_MAX, TWINVQ_MULAW_MU);
for (j = 0; j < sub; j++)
out[i * sub + j] =
- val * mulawinv(sub_step * 0.5 +
- sub_step * bits->sub_gain_bits[i * sub + j],
- SUB_AMP_MAX, MULAW_MU);
+ val * twinvq_mulawinv(sub_step * 0.5 +
+ sub_step * bits->sub_gain_bits[i * sub + j],
+ TWINVQ_SUB_AMP_MAX, TWINVQ_MULAW_MU);
}
}
}
lsp[i] = 2 * cos(lsp[i]);
switch (ftype) {
- case FT_LONG:
+ case TWINVQ_FT_LONG:
eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
break;
- case FT_MEDIUM:
+ case TWINVQ_FT_MEDIUM:
eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
break;
- case FT_SHORT:
+ case TWINVQ_FT_SHORT:
eval_lpcenv(tctx, lsp, lpc);
break;
}
float *out2 = out;
float *prev_buf;
int types_sizes[] = {
- mtab->size / mtab->fmode[FT_LONG].sub,
- mtab->size / mtab->fmode[FT_MEDIUM].sub,
- mtab->size / (mtab->fmode[FT_SHORT].sub * 2),
+ mtab->size / mtab->fmode[TWINVQ_FT_LONG].sub,
+ mtab->size / mtab->fmode[TWINVQ_FT_MEDIUM].sub,
+ mtab->size / (mtab->fmode[TWINVQ_FT_SHORT].sub * 2),
};
wsize = types_sizes[wtype_to_wsize[wtype]];
prev_buf = prev + (size - bsize) / 2;
for (j = 0; j < mtab->fmode[ftype].sub; j++) {
- int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;
+ int sub_wtype = ftype == TWINVQ_FT_MEDIUM ? 8 : wtype;
if (!j && wtype == 4)
sub_wtype = 4;
memcpy(out2, buf1 + bsize * j + wsize / 2,
(bsize - wsize / 2) * sizeof(float));
- out2 += ftype == FT_MEDIUM ? (bsize - wsize) / 2 : bsize - wsize;
+ out2 += ftype == TWINVQ_FT_MEDIUM ? (bsize - wsize) / 2 : bsize - wsize;
prev_buf = buf1 + bsize * j + bsize / 2;
}
}
static void imdct_output(TwinVQContext *tctx, enum TwinVQFrameType ftype,
- int wtype, float **out)
+ int wtype, float **out, int offset)
{
const TwinVQModeTab *mtab = tctx->mtab;
float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
int size1, size2, i;
+ float *out1, *out2;
for (i = 0; i < tctx->avctx->channels; i++)
imdct_and_window(tctx, ftype, wtype,
size2 = tctx->last_block_pos[0];
size1 = mtab->size - size2;
- memcpy(&out[0][0], prev_buf, size1 * sizeof(out[0][0]));
- memcpy(&out[0][size1], tctx->curr_frame, size2 * sizeof(out[0][0]));
+ out1 = &out[0][0] + offset;
+ memcpy(out1, prev_buf, size1 * sizeof(*out1));
+ memcpy(out1 + size1, tctx->curr_frame, size2 * sizeof(*out1));
if (tctx->avctx->channels == 2) {
- memcpy(&out[1][0], &prev_buf[2 * mtab->size],
- size1 * sizeof(out[1][0]));
- memcpy(&out[1][size1], &tctx->curr_frame[2 * mtab->size],
- size2 * sizeof(out[1][0]));
- tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
+ out2 = &out[1][0] + offset;
+ memcpy(out2, &prev_buf[2 * mtab->size],
+ size1 * sizeof(*out2));
+ memcpy(out2 + size1, &tctx->curr_frame[2 * mtab->size],
+ size2 * sizeof(*out2));
+ tctx->fdsp.butterflies_float(out1, out2, mtab->size);
}
}
-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,
enum TwinVQFrameType ftype)
{
const TwinVQModeTab *mtab = tctx->mtab;
- TwinVQFrameData *bits = &tctx->bits;
+ TwinVQFrameData *bits = &tctx->bits[tctx->cur_frame];
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];
+ float gain[TWINVQ_CHANNELS_MAX * TWINVQ_SUBBLOCKS_MAX];
+ float ppc_shape[TWINVQ_PPC_SHAPE_LEN_MAX * TWINVQ_CHANNELS_MAX * 4];
int i, j;
dec_gain(tctx, ftype, gain);
- if (ftype == FT_LONG) {
+ if (ftype == TWINVQ_FT_LONG) {
int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len * channels - 1) /
tctx->n_div[3];
- dequant(tctx, bits->ppc_coeffs, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
- mtab->ppc_shape_cb + cb_len_p * PPC_SHAPE_CB_SIZE, cb_len_p);
+ dequant(tctx, bits->ppc_coeffs, ppc_shape,
+ TWINVQ_FT_PPC, mtab->ppc_shape_cb,
+ mtab->ppc_shape_cb + cb_len_p * TWINVQ_PPC_SHAPE_CB_SIZE,
+ cb_len_p);
}
for (i = 0; i < channels; i++) {
float *chunk = out + mtab->size * i;
- float lsp[LSP_COEFS_MAX];
+ float lsp[TWINVQ_LSP_COEFS_MAX];
for (j = 0; j < sub; j++) {
- dec_bark_env(tctx, bits->bark1[i][j], bits->bark_use_hist[i][j], i,
- tctx->tmp_buf, gain[sub * i + j], ftype);
+ 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->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);
- float v = 1.0 / 8192 *
- mulawinv(pgain_step * bits->g_coef[i] +
- pgain_step / 2,
- 25000.0, PGAIN_MU);
-
- decode_ppc(tctx, bits->p_coef[i],
- ppc_shape + i * mtab->ppc_shape_len, v, chunk);
- }
+ if (ftype == TWINVQ_FT_LONG)
+ tctx->decode_ppc(tctx, bits->p_coef[i], bits->g_coef[i],
+ ppc_shape + i * mtab->ppc_shape_len, chunk);
decode_lsp(tctx, bits->lpc_idx1[i], bits->lpc_idx2[i],
bits->lpc_hist_idx[i], lsp, tctx->lsp_hist[i]);
}
}
-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 const enum TwinVQFrameType wtype_to_ftype_table[] = {
- FT_LONG, FT_LONG, FT_SHORT, FT_LONG,
- FT_MEDIUM, FT_LONG, FT_LONG, FT_MEDIUM, FT_MEDIUM
+const enum TwinVQFrameType ff_twinvq_wtype_to_ftype_table[] = {
+ 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
};
-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;
-
- init_get_bits(&gb, buf, buf_size * 8);
- skip_bits(&gb, get_bits(&gb, 8));
-
- bits->window_type = get_bits(&gb, 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 == FT_LONG) {
- for (i = 0; i < channels; i++)
- bits->gain_bits[i] = get_bits(&gb, GAIN_BITS);
- } else {
- for (i = 0; i < channels; i++) {
- bits->gain_bits[i] = get_bits(&gb, GAIN_BITS);
- for (j = 0; j < sub; j++)
- bits->sub_gain_bits[i * sub + j] = get_bits(&gb, 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 == 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 ff_twinvq_decode_frame(AVCodecContext *avctx, void *data,
+ int *got_frame_ptr, AVPacket *avpkt)
{
AVFrame *frame = data;
const uint8_t *buf = avpkt->data;
float **out = NULL;
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 */
if (tctx->discarded_packets >= 2) {
- frame->nb_samples = mtab->size;
+ frame->nb_samples = mtab->size * tctx->frames_per_packet;
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;
}
- if ((ret = twinvq_read_bitstream(avctx, tctx, buf, buf_size)) < 0)
+ if (buf_size < avctx->block_align) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Frame too small (%d bytes). Truncated file?\n", buf_size);
+ return AVERROR(EINVAL);
+ }
+
+ if ((ret = tctx->read_bitstream(avctx, tctx, buf, buf_size)) < 0)
return ret;
- read_and_decode_spectrum(tctx, tctx->spectrum, tctx->bits.ftype);
+ for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet;
+ tctx->cur_frame++) {
+ read_and_decode_spectrum(tctx, tctx->spectrum,
+ tctx->bits[tctx->cur_frame].ftype);
- imdct_output(tctx, tctx->bits.ftype, tctx->bits.window_type, out);
+ imdct_output(tctx, tctx->bits[tctx->cur_frame].ftype,
+ tctx->bits[tctx->cur_frame].window_type, out,
+ tctx->cur_frame * mtab->size);
- FFSWAP(float *, tctx->curr_frame, tctx->prev_frame);
+ FFSWAP(float *, tctx->curr_frame, tctx->prev_frame);
+ }
if (tctx->discarded_packets < 2) {
tctx->discarded_packets++;
*got_frame_ptr = 1;
- return buf_size;
+ // VQF can deliver packets 1 byte greater than block align
+ if (buf_size == avctx->block_align + 1)
+ return buf_size;
+ return avctx->block_align;
}
/**
{
int i, j, ret;
const TwinVQModeTab *mtab = tctx->mtab;
- int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
- int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
+ int size_s = mtab->size / mtab->fmode[TWINVQ_FT_SHORT].sub;
+ int size_m = mtab->size / mtab->fmode[TWINVQ_FT_MEDIUM].sub;
int channels = tctx->avctx->channels;
float norm = channels == 1 ? 2.0 : 1.0;
for (i = 0; i < line_len[0]; i++) {
int shift;
- if (num_blocks == 1 ||
- (ftype == FT_LONG && num_vect % num_blocks) ||
- (ftype != FT_LONG && num_vect & 1) ||
+ if (num_blocks == 1 ||
+ (ftype == TWINVQ_FT_LONG && num_vect % num_blocks) ||
+ (ftype != TWINVQ_FT_LONG && num_vect & 1) ||
i == line_len[1]) {
shift = 0;
- } else if (ftype == FT_LONG) {
+ } else if (ftype == TWINVQ_FT_LONG) {
shift = i;
} else
shift = i * i;
const TwinVQModeTab *mtab = tctx->mtab;
int16_t *tmp_perm = (int16_t *)tctx->tmp_buf;
- if (ftype == FT_PPC) {
+ if (ftype == TWINVQ_FT_PPC) {
size = tctx->avctx->channels;
block_size = mtab->ppc_shape_len;
} else {
mtab->fmode[i].bark_n_bit + 1);
bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
- WINDOW_TYPE_BITS + n_ch * GAIN_BITS;
+ TWINVQ_WINDOW_TYPE_BITS + n_ch * TWINVQ_GAIN_BITS;
for (i = 0; i < 2; i++)
bsize_no_main_cb[i] =
- lsp_bits_per_block + n_ch * GAIN_BITS + WINDOW_TYPE_BITS +
- mtab->fmode[i].sub * (bse_bits[i] + n_ch * SUB_GAIN_BITS);
+ lsp_bits_per_block + n_ch * TWINVQ_GAIN_BITS +
+ TWINVQ_WINDOW_TYPE_BITS +
+ mtab->fmode[i].sub * (bse_bits[i] + n_ch * TWINVQ_SUB_GAIN_BITS);
+
+ if (tctx->codec == TWINVQ_CODEC_METASOUND && !tctx->is_6kbps) {
+ bsize_no_main_cb[1] += 2;
+ bsize_no_main_cb[2] += 2;
+ }
// The remaining bits are all used for the main spectrum coefficients
for (i = 0; i < 4; i++) {
tctx->length_change[i] = num_rounded_up;
}
- for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
+ for (frametype = TWINVQ_FT_SHORT; frametype <= TWINVQ_FT_PPC; 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;
int i;
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;
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");
+ if (!avctx->block_align) {
+ avctx->block_align = tctx->frame_size + 7 >> 3;
+ } else if (avctx->block_align * 8 < tctx->frame_size) {
+ av_log(avctx, AV_LOG_ERROR, "Block align is %d bits, expected %d\n",
+ avctx->block_align * 8, tctx->frame_size);
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");
+ tctx->frames_per_packet = avctx->block_align * 8 / tctx->frame_size;
+ if (tctx->frames_per_packet > TWINVQ_MAX_FRAMES_PER_PACKET) {
+ av_log(avctx, AV_LOG_ERROR, "Too many frames per packet (%d)\n",
+ tctx->frames_per_packet);
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);
+ avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
if ((ret = init_mdct_win(tctx))) {
av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
- twinvq_decode_close(avctx);
+ ff_twinvq_decode_close(avctx);
return ret;
}
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;
}
-
-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 },
-};
projects / ffmpeg / blobdiff