48 #define MAX_LSPS_ALIGN16 16
51 #define MAX_FRAMESIZE 160
52 #define MAX_SIGNAL_HISTORY 416
53 #define MAX_SFRAMESIZE (MAX_FRAMESIZE * MAX_FRAMES)
55 #define SFRAME_CACHE_MAXSIZE 256
306 int cntr[8] = { 0 },
n, res;
308 memset(vbm_tree, 0xff,
sizeof(vbm_tree[0]) * 25);
309 for (
n = 0;
n < 17;
n++) {
313 vbm_tree[res * 3 + cntr[res]++] =
n;
323 10, 10, 10, 12, 12, 12,
326 static const uint16_t codes[] = {
327 0x0000, 0x0001, 0x0002,
328 0x000c, 0x000d, 0x000e,
329 0x003c, 0x003d, 0x003e,
330 0x00fc, 0x00fd, 0x00fe,
331 0x03fc, 0x03fd, 0x03fe,
332 0x0ffc, 0x0ffd, 0x0ffe,
333 0x3ffc, 0x3ffd, 0x3ffe, 0x3fff
337 bits, 1, 1, codes, 2, 2, 132);
345 int n,
flags, pitch_range, lsp16_flag;
358 "Invalid extradata size %d (should be 46)\n",
372 memcpy(&s->
sin[255], s->
cos, 256 *
sizeof(s->
cos[0]));
373 for (n = 0; n < 255; n++) {
381 "Invalid denoise filter strength %d (max=11)\n",
389 lsp16_flag = flags & 0x1000;
399 for (n = 0; n < s->
lsps; n++)
414 if (pitch_range <= 0) {
424 int min_sr = ((((1 << 8) - 50) * 400) + 0xFF) >> 8,
428 "Unsupported samplerate %d (min=%d, max=%d)\n",
478 const float *speech_synth,
482 float speech_energy = 0.0, postfilter_energy = 0.0, gain_scale_factor;
483 float mem = *gain_mem;
485 for (i = 0; i <
size; i++) {
486 speech_energy += fabsf(speech_synth[i]);
487 postfilter_energy += fabsf(in[i]);
489 gain_scale_factor = (1.0 -
alpha) * speech_energy / postfilter_energy;
491 for (i = 0; i <
size; i++) {
492 mem = alpha * mem + gain_scale_factor;
493 out[i] = in[i] *
mem;
521 float optimal_gain = 0, dot;
524 *best_hist_ptr =
NULL;
529 if (dot > optimal_gain) {
533 }
while (--ptr >= end);
535 if (optimal_gain <= 0)
541 if (optimal_gain <= dot) {
542 dot = dot / (dot + 0.6 * optimal_gain);
547 for (n = 0; n <
size; n++)
548 out[n] = best_hist_ptr[n] + dot * (in[n] - best_hist_ptr[n]);
577 int fcb_type,
float *
coeffs,
int remainder)
580 float irange, angle_mul, gain_mul, range, sq;
585 #define log_range(var, assign) do { \
586 float tmp = log10f(assign); var = tmp; \
587 max = FFMAX(max, tmp); min = FFMIN(min, tmp); \
589 log_range(last_coeff, lpcs[1] * lpcs[1]);
590 for (n = 1; n < 64; n++)
591 log_range(lpcs[n], lpcs[n * 2] * lpcs[n * 2] +
592 lpcs[n * 2 + 1] * lpcs[n * 2 + 1]);
603 irange = 64.0 / range;
607 for (n = 0; n <= 64; n++) {
610 idx =
lrint((max - lpcs[n]) * irange - 1);
613 lpcs[
n] = angle_mul * pwr;
616 idx = av_clipf((pwr * gain_mul - 0.0295) * 70.570526123, 0, INT_MAX / 2);
620 powf(1.0331663, idx - 127);
633 idx = 255 + av_clip(lpcs[64], -255, 255);
634 coeffs[0] = coeffs[0] * s->
cos[idx];
635 idx = 255 + av_clip(lpcs[64] - 2 * lpcs[63], -255, 255);
636 last_coeff = coeffs[64] * s->
cos[idx];
638 idx = 255 + av_clip(-lpcs[64] - 2 * lpcs[n - 1], -255, 255);
639 coeffs[n * 2 + 1] = coeffs[
n] * s->
sin[idx];
640 coeffs[n * 2] = coeffs[
n] * s->
cos[idx];
644 idx = 255 + av_clip( lpcs[64] - 2 * lpcs[n - 1], -255, 255);
645 coeffs[n * 2 + 1] = coeffs[
n] * s->
sin[idx];
646 coeffs[n * 2] = coeffs[
n] * s->
cos[idx];
654 memset(&coeffs[remainder], 0,
sizeof(coeffs[0]) * (128 - remainder));
658 coeffs[remainder - 1] = 0;
665 for (n = 0; n < remainder; n++)
696 float *synth_pf,
int size,
699 int remainder, lim,
n;
705 tilted_lpcs[0] = 1.0;
706 memcpy(&tilted_lpcs[1], lpcs,
sizeof(lpcs[0]) * s->
lsps);
707 memset(&tilted_lpcs[s->
lsps + 1], 0,
708 sizeof(tilted_lpcs[0]) * (128 - s->
lsps - 1));
710 tilted_lpcs, s->
lsps + 2);
716 remainder =
FFMIN(127 - size, size - 1);
721 memset(&synth_pf[size], 0,
sizeof(synth_pf[0]) * (128 - size));
726 for (n = 1; n < 64; n++) {
727 float v1 = synth_pf[n * 2], v2 = synth_pf[n * 2 + 1];
728 synth_pf[n * 2] = v1 *
coeffs[n * 2] - v2 *
coeffs[n * 2 + 1];
729 synth_pf[n * 2 + 1] = v2 * coeffs[n * 2] + v1 * coeffs[n * 2 + 1];
737 for (n = 0; n < lim; n++)
747 for (n = 0; n < lim; n++)
749 if (lim < remainder) {
778 float *samples,
int size,
779 const float *lpcs,
float *zero_exc_pf,
780 int fcb_type,
int pitch)
784 *synth_filter_in = zero_exc_pf;
793 synth_filter_in = synth_filter_in_buf;
797 synth_filter_in, size, s->
lsps);
798 memcpy(&synth_pf[-s->
lsps], &synth_pf[size - s->
lsps],
799 sizeof(synth_pf[0]) * s->
lsps);
811 (
const float[2]) { -1.99997, 1.0 },
812 (
const float[2]) { -1.9330735188, 0.93589198496 },
832 const uint16_t *values,
833 const uint16_t *
sizes,
836 const double *base_q)
840 memset(lsps, 0, num *
sizeof(*lsps));
841 for (n = 0; n < n_stages; n++) {
842 const uint8_t *t_off = &table[values[
n] * num];
843 double base = base_q[
n], mul = mul_q[
n];
845 for (m = 0; m < num; m++)
846 lsps[m] += base + mul * t_off[m];
848 table += sizes[
n] * num;
865 static const uint16_t vec_sizes[4] = { 256, 64, 32, 32 };
866 static const double mul_lsf[4] = {
867 5.2187144800e-3, 1.4626986422e-3,
868 9.6179549166e-4, 1.1325736225e-3
870 static const double base_lsf[4] = {
871 M_PI * -2.15522e-1,
M_PI * -6.1646e-2,
872 M_PI * -3.3486e-2,
M_PI * -5.7408e-2
890 double *i_lsps,
const double *old,
891 double *
a1,
double *
a2,
int q_mode)
893 static const uint16_t vec_sizes[3] = { 128, 64, 64 };
894 static const double mul_lsf[3] = {
895 2.5807601174e-3, 1.2354460219e-3, 1.1763821673e-3
897 static const double base_lsf[3] = {
898 M_PI * -1.07448e-1,
M_PI * -5.2706e-2,
M_PI * -5.1634e-2
900 const float (*ipol_tab)[2][10] = q_mode ?
912 for (n = 0; n < 10; n++) {
913 double delta = old[
n] - i_lsps[
n];
914 a1[
n] = ipol_tab[
interpol][0][
n] * delta + i_lsps[
n];
915 a1[10 +
n] = ipol_tab[
interpol][1][
n] * delta + i_lsps[
n];
927 static const uint16_t vec_sizes[5] = { 256, 64, 128, 64, 128 };
928 static const double mul_lsf[5] = {
929 3.3439586280e-3, 6.9908173703e-4,
930 3.3216608306e-3, 1.0334960326e-3,
933 static const double base_lsf[5] = {
934 M_PI * -1.27576e-1,
M_PI * -2.4292e-2,
935 M_PI * -1.28094e-1,
M_PI * -3.2128e-2,
959 double *i_lsps,
const double *old,
960 double *
a1,
double *
a2,
int q_mode)
962 static const uint16_t vec_sizes[3] = { 128, 128, 128 };
963 static const double mul_lsf[3] = {
964 1.2232979501e-3, 1.4062241527e-3, 1.6114744851e-3
966 static const double base_lsf[3] = {
969 const float (*ipol_tab)[2][16] = q_mode ?
981 for (n = 0; n < 16; n++) {
982 double delta = old[
n] - i_lsps[
n];
983 a1[
n] = ipol_tab[
interpol][0][
n] * delta + i_lsps[
n];
984 a1[16 +
n] = ipol_tab[
interpol][1][
n] * delta + i_lsps[
n];
1011 static const int16_t start_offset[94] = {
1012 -11, -9, -7, -5, -3, -1, 1, 3, 5, 7, 9, 11,
1013 13, 15, 18, 17, 19, 20, 21, 22, 23, 24, 25, 26,
1014 27, 28, 29, 30, 31, 32, 33, 35, 37, 39, 41, 43,
1015 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67,
1016 69, 71, 73, 75, 77, 79, 81, 83, 85, 87, 89, 91,
1017 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, 115,
1018 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139,
1019 141, 143, 145, 147, 149, 151, 153, 155, 157, 159
1025 if ((bits =
get_bits(gb, 6)) >= 54) {
1027 bits += (bits - 54) * 3 +
get_bits(gb, 2);
1033 for (offset = start_offset[bits]; offset < 0; offset += pitch[0]) ;
1046 if (start_offset[bits] < 0)
1063 uint16_t use_mask_mem[9];
1064 uint16_t *use_mask = use_mask_mem + 2;
1073 pulse_start,
n, idx, range, aidx, start_off = 0;
1082 if (block_idx == 0) {
1091 pulse_start = s->
aw_n_pulses[block_idx] > 0 ? pulse_off - range / 2 : 0;
1096 memset(&use_mask[-2], 0, 2 *
sizeof(use_mask[0]));
1097 memset( use_mask, -1, 5 *
sizeof(use_mask[0]));
1098 memset(&use_mask[5], 0, 2 *
sizeof(use_mask[0]));
1102 uint16_t *use_mask_ptr = &use_mask[idx >> 4];
1103 int first_sh = 16 - (idx & 15);
1104 *use_mask_ptr++ &= 0xFFFF
u << first_sh;
1105 excl_range -= first_sh;
1106 if (excl_range >= 16) {
1107 *use_mask_ptr++ = 0;
1108 *use_mask_ptr &= 0xFFFF >> (excl_range - 16);
1110 *use_mask_ptr &= 0xFFFF >> excl_range;
1115 for (n = 0; n <= aidx; pulse_start++) {
1116 for (idx = pulse_start; idx < 0; idx += fcb->
pitch_lag) ;
1118 if (use_mask[0]) idx = 0x0F;
1119 else if (use_mask[1]) idx = 0x1F;
1120 else if (use_mask[2]) idx = 0x2F;
1121 else if (use_mask[3]) idx = 0x3F;
1122 else if (use_mask[4]) idx = 0x4F;
1126 if (use_mask[idx >> 4] & (0x8000 >> (idx & 15))) {
1127 use_mask[idx >> 4] &= ~(0x8000 >> (idx & 15));
1133 fcb->
x[fcb->
n] = start_off;
1157 int n, v_mask, i_mask, sh, n_pulses;
1171 for (n = n_pulses - 1; n >= 0; n--, val >>= sh) {
1172 fcb->
y[fcb->
n] = (val & v_mask) ? -1.0 : 1.0;
1173 fcb->
x[fcb->
n] = (val & i_mask) * n_pulses + n +
1175 while (fcb->
x[fcb->
n] < 0)
1181 int num2 = (val & 0x1FF) >> 1,
delta, idx;
1183 if (num2 < 1 * 79) {
delta = 1; idx = num2 + 1; }
1184 else if (num2 < 2 * 78) {
delta = 3; idx = num2 + 1 - 1 * 77; }
1185 else if (num2 < 3 * 77) {
delta = 5; idx = num2 + 1 - 2 * 76; }
1186 else {
delta = 7; idx = num2 + 1 - 3 * 75; }
1187 v = (val & 0x200) ? -1.0 : 1.0;
1192 fcb->
x[fcb->
n + 1] = idx;
1193 fcb->
y[fcb->
n + 1] = (val & 1) ? -v : v;
1211 static int pRNG(
int frame_cntr,
int block_num,
int block_size)
1223 static const unsigned int div_tbl[9][2] = {
1224 { 8332, 3 * 715827883
U },
1225 { 4545, 0 * 390451573
U },
1226 { 3124, 11 * 268435456
U },
1227 { 2380, 15 * 204522253
U },
1228 { 1922, 23 * 165191050
U },
1229 { 1612, 23 * 138547333
U },
1230 { 1388, 27 * 119304648
U },
1231 { 1219, 16 * 104755300
U },
1232 { 1086, 39 * 93368855
U }
1234 unsigned int z,
y, x =
MUL16(block_num, 1877) + frame_cntr;
1235 if (x >= 0xFFFF) x -= 0xFFFF;
1237 y = x - 9 *
MULH(477218589, x);
1238 z = (uint16_t) (x * div_tbl[y][0] +
UMULH(x, div_tbl[y][1]));
1240 return z % (1000 - block_size);
1248 int block_idx,
int size,
1270 for (n = 0; n <
size; n++)
1279 int block_idx,
int size,
1280 int block_pitch_sh2,
1284 static const float gain_coeff[6] = {
1285 0.8169, -0.06545, 0.1726, 0.0185, -0.0359, 0.0458
1288 int n, idx, gain_weight;
1292 memset(pulses, 0,
sizeof(*pulses) * size);
1309 for (n = 0; n <
size; n++)
1321 for (n = 0; n < 5; n++) {
1327 fcb.
x[fcb.
n] = n + 5 * pos1;
1328 fcb.
y[fcb.
n++] = sign;
1329 if (n < frame_desc->dbl_pulses) {
1331 fcb.
x[fcb.
n] = n + 5 * pos2;
1332 fcb.
y[fcb.
n++] = (pos1 < pos2) ? -sign : sign;
1352 for (n = 0; n < gain_weight; n++)
1358 for (n = 0; n <
size; n +=
len) {
1360 int abs_idx = block_idx * size +
n;
1363 int pitch = (pitch_sh16 + 0x6FFF) >> 16;
1364 int idx_sh16 = ((pitch << 16) - pitch_sh16) * 8 + 0x58000;
1365 idx = idx_sh16 >> 16;
1368 next_idx_sh16 = (idx_sh16) &~ 0xFFFF;
1370 next_idx_sh16 = (idx_sh16 + 0x10000) &~ 0xFFFF;
1381 int block_pitch = block_pitch_sh2 >> 2;
1382 idx = block_pitch_sh2 & 3;
1389 sizeof(
float) * size);
1394 acb_gain, fcb_gain, size);
1414 int block_idx,
int size,
1415 int block_pitch_sh2,
1416 const double *lsps,
const double *prev_lsps,
1418 float *excitation,
float *synth)
1429 frame_desc, excitation);
1432 fac = (block_idx + 0.5) / frame_desc->
n_blocks;
1433 for (n = 0; n < s->
lsps; n++)
1434 i_lsps[n] = cos(prev_lsps[n] + fac * (lsps[n] - prev_lsps[n]));
1458 const double *lsps,
const double *prev_lsps,
1459 float *excitation,
float *synth)
1462 int n, n_blocks_x2, log_n_blocks_x2,
av_uninit(cur_pitch_val);
1470 "Invalid frame type VLC code, skipping\n");
1493 int fac = n * 2 + 1;
1495 pitch[
n] = (
MUL16(fac, cur_pitch_val) +
1537 last_block_pitch = av_clip(block_pitch,
1543 if (block_pitch < t1) {
1547 if (block_pitch <
t2) {
1552 if (block_pitch <
t3) {
1559 pitch[
n] = bl_pitch_sh2 >> 2;
1564 bl_pitch_sh2 = pitch[
n] << 2;
1573 synth_block(s, gb, n, block_nsamples, bl_pitch_sh2,
1575 &excitation[n * block_nsamples],
1576 &synth[n * block_nsamples]);
1585 for (n = 0; n < s->
lsps; n++)
1586 i_lsps[n] = cos(0.5 * (prev_lsps[n] + lsps[n]));
1592 for (n = 0; n < s->
lsps; n++)
1593 i_lsps[n] = cos(lsps[n]);
1595 postfilter(s, &synth[80], &samples[80], 80, lpcs,
1599 memcpy(samples, synth, 160 *
sizeof(synth[0]));
1639 lsps[0] =
FFMAX(lsps[0], 0.0015 *
M_PI);
1640 for (n = 1; n < num; n++)
1641 lsps[n] =
FFMAX(lsps[n], lsps[n - 1] + 0.0125 *
M_PI);
1642 lsps[num - 1] =
FFMIN(lsps[num - 1], 0.9985 *
M_PI);
1646 for (n = 1; n < num; n++) {
1647 if (lsps[n] < lsps[n - 1]) {
1648 for (m = 1; m < num; m++) {
1649 double tmp = lsps[
m];
1650 for (l = m - 1; l >= 0; l--) {
1651 if (lsps[l] <= tmp)
break;
1652 lsps[l + 1] = lsps[l];
1674 int n, need_bits, bd_idx;
1696 int aw_idx_is_ext = 0;
1726 need_bits = 2 * !aw_idx_is_ext;
1760 int n, res, n_samples = 480;
1769 s->
lsps *
sizeof(*synth));
1796 if ((n_samples =
get_bits(gb, 12)) > 480) {
1798 "Superframe encodes >480 samples (%d), not allowed\n",
1807 for (n = 0; n < s->
lsps; n++)
1808 prev_lsps[n] = s->
prev_lsps[n] - mean_lsf[n];
1815 for (n = 0; n < s->
lsps; n++) {
1816 lsps[0][
n] = mean_lsf[
n] + (a1[
n] - a2[n * 2]);
1817 lsps[1][
n] = mean_lsf[
n] + (a1[s->
lsps +
n] - a2[n * 2 + 1]);
1818 lsps[2][
n] += mean_lsf[
n];
1820 for (n = 0; n < 3; n++)
1829 samples = (
float *)frame->
data[0];
1832 for (n = 0; n < 3; n++) {
1836 if (s->
lsps == 10) {
1841 for (m = 0; m < s->
lsps; m++)
1842 lsps[n][m] += mean_lsf[m];
1848 lsps[n], n == 0 ? s->
prev_lsps : lsps[n - 1],
1850 &synth[s->
lsps + n * MAX_FRAMESIZE]))) {
1870 s->
lsps *
sizeof(*synth));
1904 }
while (res == 0x3F);
1929 int rmn_bytes, rmn_bits;
1932 if (rmn_bits < nbits)
1936 rmn_bits &= 7; rmn_bytes >>= 3;
1937 if ((rmn_bits =
FFMIN(rmn_bits, nbits)) > 0)
1940 FFMIN(nbits - rmn_bits, rmn_bytes << 3));
1955 int *got_frame_ptr,
AVPacket *avpkt)
1994 if (res > avpkt->
size) {
1996 "Trying to skip %d bytes in packet of size %d\n",
2016 }
else if (*got_frame_ptr) {
2020 if (res > avpkt->
size) {
2022 "Trying to skip %d bytes in packet of size %d\n",
2065 for (n = 0; n < s->
lsps; n++)
Description of frame types.
static void aw_pulse_set1(WMAVoiceContext *s, GetBitContext *gb, int block_idx, AMRFixed *fcb)
Apply first set of pitch-adaptive window pulses.
av_cold void ff_rdft_end(RDFTContext *s)
static const uint8_t wmavoice_dq_lsp16r2[0x500]
const char const char void * val
int do_apf
whether to apply the averaged projection filter (APF)
#define AVERROR_INVALIDDATA
Invalid data found when processing input.
static int pRNG(int frame_cntr, int block_num, int block_size)
Generate a random number from frame_cntr and block_idx, which will lief in the range [0...
static av_cold int decode_vbmtree(GetBitContext *gb, int8_t vbm_tree[25])
Set up the variable bit mode (VBM) tree from container extradata.
void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs, const float *in, int buffer_length, int filter_length)
LP synthesis filter.
float gain_pred_err[6]
cache for gain prediction
This structure describes decoded (raw) audio or video data.
void(* dct_calc)(struct DCTContext *s, FFTSample *data)
int aw_next_pulse_off_cache
the position (relative to start of the second block) at which pulses should start to be positioned...
int frame_lsp_bitsize
size (in bits) of LSPs, when encoded per-frame (independent coding)
ptrdiff_t const GLvoid * data
static void flush(AVCodecContext *avctx)
float postfilter_agc
gain control memory, used in adaptive_gain_control()
void ff_acelp_apply_order_2_transfer_function(float *out, const float *in, const float zero_coeffs[2], const float pole_coeffs[2], float gain, float mem[2], int n)
Apply an order 2 rational transfer function in-place.
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
static unsigned int get_bits(GetBitContext *s, int n)
Read 1-25 bits.
static void postfilter(WMAVoiceContext *s, const float *synth, float *samples, int size, const float *lpcs, float *zero_exc_pf, int fcb_type, int pitch)
Averaging projection filter, the postfilter used in WMAVoice.
memory handling functions
void ff_weighted_vector_sumf(float *out, const float *in_a, const float *in_b, float weight_coeff_a, float weight_coeff_b, int length)
float implementation of weighted sum of two vectors.
static void skip_bits_long(GetBitContext *s, int n)
static av_cold int init(AVCodecContext *avctx)
#define avpriv_request_sample(...)
float synth_filter_out_buf[0x80+MAX_LSPS_ALIGN16]
aligned buffer for postfilter speech synthesis
static void aw_parse_coords(WMAVoiceContext *s, GetBitContext *gb, const int *pitch)
Parse the offset of the first pitch-adaptive window pulses, and the distribution of pulses between th...
int aw_n_pulses[2]
number of AW-pulses in each block; note that this number can be negative (in which case it basically ...
static int interpol(MBContext *mb, uint32_t *color, int x, int y, int linesize)
void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
Copy the content of src to the bitstream.
static void stabilize_lsps(double *lsps, int num)
Ensure minimum value for first item, maximum value for last value, proper spacing between each value ...
#define DECLARE_ALIGNED(n, t, v)
static const float wmavoice_gain_codebook_fcb[128]
static const uint8_t wmavoice_dq_lsp16i1[0x640]
static const uint8_t wmavoice_dq_lsp16r1[0x500]
int spillover_nbits
number of bits of the previous packet's last superframe preceding this packet's first full superframe...
void ff_set_fixed_vector(float *out, const AMRFixed *in, float scale, int size)
Add fixed vector to an array from a sparse representation.
int block_pitch_nbits
number of bits used to specify the first block's pitch value
static const uint8_t wmavoice_dq_lsp16i3[0x300]
static int synth_frame(AVCodecContext *ctx, GetBitContext *gb, int frame_idx, float *samples, const double *lsps, const double *prev_lsps, float *excitation, float *synth)
Synthesize output samples for a single frame.
static void calc_input_response(WMAVoiceContext *s, float *lpcs, int fcb_type, float *coeffs, int remainder)
Derive denoise filter coefficients (in real domain) from the LPCs.
static void dequant_lsp10i(GetBitContext *gb, double *lsps)
Parse 10 independently-coded LSPs.
#define MAX_LSPS_ALIGN16
same as MAX_LSPS; needs to be multiple
int block_align
number of bytes per packet if constant and known or 0 Used by some WAV based audio codecs...
no adaptive codebook (only hardcoded fixed)
static int aw_pulse_set2(WMAVoiceContext *s, GetBitContext *gb, int block_idx, AMRFixed *fcb)
Apply second set of pitch-adaptive window pulses.
static const float wmavoice_ipol1_coeffs[17 *9]
static const uint8_t wmavoice_dq_lsp16i2[0x3c0]
#define av_assert0(cond)
assert() equivalent, that is always enabled.
int spillover_bitsize
number of bits used to specify spillover_nbits in the packet header = ceil(log2(ctx->block_align << 3...
float avpriv_scalarproduct_float_c(const float *v1, const float *v2, int len)
Return the scalar product of two vectors.
int block_delta_pitch_nbits
number of bits used to specify the delta pitch between this and the last block's pitch value...
enum AVSampleFormat sample_fmt
audio sample format
Sparse representation for the algebraic codebook (fixed) vector.
static const uint8_t wmavoice_dq_lsp16r3[0x600]
#define INIT_VLC_STATIC(vlc, bits, a, b, c, d, e, f, g, static_size)
static const float wmavoice_gain_codebook_acb[128]
uint8_t log_n_blocks
log2(n_blocks)
int aw_first_pulse_off[2]
index of first sample to which to apply AW-pulses, or -0xff if unset
Per-block pitch with signal generation using a Hamming sinc window function.
static av_cold int end(AVCodecContext *avctx)
int has_residual_lsps
if set, superframes contain one set of LSPs that cover all frames, encoded as independent and residua...
float tilted_lpcs_pf[0x80]
aligned buffer for LPC tilting
uint8_t * extradata
some codecs need / can use extradata like Huffman tables.
static float tilt_factor(const float *lpcs, int n_lpcs)
Get the tilt factor of a formant filter from its transfer function.
static av_cold void init_static_data(void)
static const uint8_t wmavoice_dq_lsp10r[0x1400]
static void dequant_lsps(double *lsps, int num, const uint16_t *values, const uint16_t *sizes, int n_stages, const uint8_t *table, const double *mul_q, const double *base_q)
Dequantize LSPs.
static int check_bits_for_superframe(GetBitContext *orig_gb, WMAVoiceContext *s)
Test if there's enough bits to read 1 superframe.
static const float wmavoice_ipol2_coeffs[32]
Hamming-window sinc function (num = 32, x = [ 0, 31 ]): (0.54 + 0.46 * cos(2 * M_PI * x / (num - 1)))...
static int get_bits_count(const GetBitContext *s)
float dcf_mem[2]
DC filter history.
bitstream reader API header.
static av_cold void wmavoice_flush(AVCodecContext *ctx)
float synth_history[MAX_LSPS]
see excitation_history
double prev_lsps[MAX_LSPS]
LSPs of the last frame of the previous superframe.
static void copy_bits(PutBitContext *pb, const uint8_t *data, int size, GetBitContext *gb, int nbits)
Copy (unaligned) bits from gb/data/size to pb.
static int get_bits_left(GetBitContext *gb)
static double alpha(void *priv, double x, double y)
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
static const double wmavoice_mean_lsf16[2][16]
int sframe_cache_size
set to >0 if we have data from an (incomplete) superframe from a previous packet that spilled over in...
static const float wmavoice_lsp10_intercoeff_b[32][2][10]
int block_pitch_range
range of the block pitch
static const float wmavoice_std_codebook[1000]
static const int sizes[][2]
int last_acb_type
frame type [0-2] of the previous frame
static const struct endianess table[]
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
static const float wmavoice_gain_silence[256]
int denoise_filter_cache_size
samples in denoise_filter_cache
int history_nsamples
number of samples in history for signal prediction (through ACB)
static const uint8_t wmavoice_dq_lsp10i[0xf00]
static const float wmavoice_lsp10_intercoeff_a[32][2][10]
static const float wmavoice_energy_table[128]
LUT for 1.071575641632 * pow(1.0331663, n - 127)
Windows Media Voice (WMAVoice) tables.
const char * name
Name of the codec implementation.
int denoise_tilt_corr
Whether to apply tilt correction to the Wiener filter coefficients (postfilter)
int aw_idx_is_ext
whether the AW index was encoded in 8 bits (instead of 6)
static const uint8_t offset[127][2]
Libavcodec external API header.
uint16_t block_conv_table[4]
boundaries for block pitch unit/scale conversion
DCTContext dst
contexts for phase shift (in Hilbert transform, part of postfilter)
int lsp_def_mode
defines different sets of LSP defaults [0, 1]
uint64_t channel_layout
Audio channel layout.
void(* rdft_calc)(struct RDFTContext *s, FFTSample *z)
static int put_bits_count(PutBitContext *s)
int skip_bits_next
number of bits to skip at the next call to wmavoice_decode_packet() (since they're part of the previo...
hardcoded (fixed) codebook with per-block gain values
static void dequant_lsp16r(GetBitContext *gb, double *i_lsps, const double *old, double *a1, double *a2, int q_mode)
Parse 16 independently-coded LSPs, and then derive the tables to generate LSPs for the other frames f...
int min_pitch_val
base value for pitch parsing code
WMA Voice decoding context.
static void wiener_denoise(WMAVoiceContext *s, int fcb_type, float *synth_pf, int size, const float *lpcs)
This function applies a Wiener filter on the (noisy) speech signal as a means to denoise it...
int denoise_strength
strength of denoising in Wiener filter [0-11]
uint8_t sframe_cache[SFRAME_CACHE_MAXSIZE+AV_INPUT_BUFFER_PADDING_SIZE]
cache for superframe data split over multiple packets
audio channel layout utility functions
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
#define log_range(var, assign)
#define MAX_LSPS
maximum filter order
static VLC frame_type_vlc
Frame type VLC coding.
int pitch_nbits
number of bits used to specify the pitch value in the frame header
#define MAX_BLOCKS
maximum number of blocks per frame
float denoise_coeffs_pf[0x80]
aligned buffer for denoise coefficients
static void dequant_lsp10r(GetBitContext *gb, double *i_lsps, const double *old, double *a1, double *a2, int q_mode)
Parse 10 independently-coded LSPs, and then derive the tables to generate LSPs for the other frames f...
static av_always_inline unsigned UMULH(unsigned a, unsigned b)
static av_always_inline int get_vlc2(GetBitContext *s, VLC_TYPE(*table)[2], int bits, int max_depth)
Parse a vlc code.
static int kalman_smoothen(WMAVoiceContext *s, int pitch, const float *in, float *out, int size)
Kalman smoothing function.
void ff_tilt_compensation(float *mem, float tilt, float *samples, int size)
Apply tilt compensation filter, 1 - tilt * z-1.
static const float wmavoice_gain_universal[64]
void ff_acelp_lspd2lpc(const double *lsp, float *lpc, int lp_half_order)
Reconstruct LPC coefficients from the line spectral pair frequencies.
static av_cold int wmavoice_decode_init(AVCodecContext *ctx)
Set up decoder with parameters from demuxer (extradata etc.).
#define AVERROR_PATCHWELCOME
Not yet implemented in FFmpeg, patches welcome.
int sframe_lsp_bitsize
size (in bits) of LSPs, when encoded per superframe (residual coding)
static const uint8_t last_coeff[3]
static const struct frame_type_desc frame_descs[17]
float denoise_filter_cache[MAX_FRAMESIZE]
int sample_rate
samples per second
Pitch-adaptive window (AW) pulse signals, used in particular for low-bitrate streams.
void AAC_RENAME() ff_sine_window_init(INTFLOAT *window, int n)
Generate a sine window.
static int wmavoice_decode_packet(AVCodecContext *ctx, void *data, int *got_frame_ptr, AVPacket *avpkt)
Packet decoding: a packet is anything that the (ASF) demuxer contains, and we expect that the demuxer...
main external API structure.
static int parse_packet_header(WMAVoiceContext *s)
Parse the packet header at the start of each packet (input data to this decoder). ...
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
Get a buffer for a frame.
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(constint16_t *) pi >>8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(constint32_t *) pi >>24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(constfloat *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(constfloat *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(constfloat *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(constdouble *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(constdouble *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(constdouble *) pi *(1U<< 31))))#defineSET_CONV_FUNC_GROUP(ofmt, ifmt) staticvoidset_generic_function(AudioConvert *ac){}voidff_audio_convert_free(AudioConvert **ac){if(!*ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);}AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enumAVSampleFormatout_fmt, enumAVSampleFormatin_fmt, intchannels, intsample_rate, intapply_map){AudioConvert *ac;intin_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) returnNULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method!=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt)>2){ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc){av_free(ac);returnNULL;}returnac;}in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar){ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar?ac->channels:1;}elseif(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;elseac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);returnac;}intff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in){intuse_generic=1;intlen=in->nb_samples;intp;if(ac->dc){av_log(ac->avr, AV_LOG_TRACE,"%dsamples-audio_convert:%sto%s(dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));returnff_convert_dither(ac-> in
AVCodec ff_wmavoice_decoder
int8_t vbm_tree[25]
converts VLC codes to frame type
static unsigned int get_bits1(GetBitContext *s)
static void synth_block(WMAVoiceContext *s, GetBitContext *gb, int block_idx, int size, int block_pitch_sh2, const double *lsps, const double *prev_lsps, const struct frame_type_desc *frame_desc, float *excitation, float *synth)
Parse data in a single block.
static av_cold int wmavoice_decode_end(AVCodecContext *ctx)
static void skip_bits(GetBitContext *s, int n)
av_cold int ff_dct_init(DCTContext *s, int nbits, enum DCTTransformType inverse)
Set up DCT.
#define AV_CODEC_CAP_SUBFRAMES
Codec can output multiple frames per AVPacket Normally demuxers return one frame at a time...
int pitch_diff_sh16
((cur_pitch_val - last_pitch_val) << 16) / MAX_FRAMESIZE
static int init_get_bits(GetBitContext *s, const uint8_t *buffer, int bit_size)
Initialize GetBitContext.
#define MAX_SFRAMESIZE
maximum number of samples per superframe
int lsp_q_mode
defines quantizer defaults [0, 1]
int frame_cntr
current frame index [0 - 0xFFFE]; is only used for comfort noise in pRNG()
void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs, const float *in, int buffer_length, int filter_length)
LP zero synthesis filter.
Innovation (fixed) codebook pulse sets in combinations of either single pulses or pulse pairs...
static av_always_inline av_const long int lrint(double x)
comfort noise during silence generated from a hardcoded (fixed) codebook with per-frame (low) gain va...
static void adaptive_gain_control(float *out, const float *in, const float *speech_synth, int size, float alpha, float *gain_mem)
Adaptive gain control (as used in postfilter).
static const float mean_lsf[10]
#define SFRAME_CACHE_MAXSIZE
maximum cache size for frame data that
uint8_t fcb_type
Fixed codebook type (FCB_TYPE_*)
static void dequant_lsp16i(GetBitContext *gb, double *lsps)
Parse 16 independently-coded LSPs.
RDFTContext irdft
contexts for FFT-calculation in the postfilter (for denoise filter)
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
static int synth_superframe(AVCodecContext *ctx, AVFrame *frame, int *got_frame_ptr)
Synthesize output samples for a single superframe.
static int decode(AVCodecContext *avctx, void *data, int *got_sub, AVPacket *avpkt)
adaptive codebook with per-frame pitch, which we interpolate to get a per-sample pitch.
static void synth_block_hardcoded(WMAVoiceContext *s, GetBitContext *gb, int block_idx, int size, const struct frame_type_desc *frame_desc, float *excitation)
Parse hardcoded signal for a single block.
uint8_t n_blocks
amount of blocks per frame (each block (contains 160/n_blocks samples)
common internal api header.
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
static av_cold void wmavoice_init_static_data(AVCodec *codec)
float excitation_history[MAX_SIGNAL_HISTORY]
cache of the signal of previous superframes, used as a history for signal generation ...
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
int last_pitch_val
pitch value of the previous frame
#define AV_INPUT_BUFFER_PADDING_SIZE
Required number of additionally allocated bytes at the end of the input bitstream for decoding...
#define MAX_FRAMESIZE
maximum number of samples per frame
float silence_gain
set for use in blocks if ACB_TYPE_NONE
static const double wmavoice_mean_lsf10[2][10]
static const int16_t coeffs[]
int channels
number of audio channels
VLC_TYPE(* table)[2]
code, bits
av_cold void ff_dct_end(DCTContext *s)
void ff_acelp_interpolatef(float *out, const float *in, const float *filter_coeffs, int precision, int frac_pos, int filter_length, int length)
Floating point version of ff_acelp_interpolate()
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(constint16_t *) pi >>8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(constint32_t *) pi >>24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(constfloat *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(constfloat *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(constfloat *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(constdouble *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(constdouble *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(constdouble *) pi *(1U<< 31))))#defineSET_CONV_FUNC_GROUP(ofmt, ifmt) staticvoidset_generic_function(AudioConvert *ac){}voidff_audio_convert_free(AudioConvert **ac){if(!*ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);}AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enumAVSampleFormatout_fmt, enumAVSampleFormatin_fmt, intchannels, intsample_rate, intapply_map){AudioConvert *ac;intin_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) returnNULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method!=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt)>2){ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc){av_free(ac);returnNULL;}returnac;}in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar){ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar?ac->channels:1;}elseif(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;elseac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);returnac;}intff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in){intuse_generic=1;intlen=in->nb_samples;intp;if(ac->dc){av_log(ac->avr, AV_LOG_TRACE,"%dsamples-audio_convert:%sto%s(dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));returnff_convert_dither(ac-> out
int block_delta_pitch_hrange
1/2 range of the delta (full range is from -this to +this-1)
int max_pitch_val
max value + 1 for pitch parsing
int lsps
number of LSPs per frame [10 or 16]
#define MAX_FRAMES
maximum number of frames per superframe
static const int8_t pulses[4]
Number of non-zero pulses in the MP-MLQ excitation.
static const float wmavoice_lsp16_intercoeff_b[32][2][16]
void av_memcpy_backptr(uint8_t *dst, int back, int cnt)
deliberately overlapping memcpy implementation
PutBitContext pb
bitstream writer for sframe_cache
uint8_t acb_type
Adaptive codebook type (ACB_TYPE_*)
static const float wmavoice_denoise_power_table[12][64]
LUT for f(x,y) = pow((y + 6.9) / 64, 0.025 * (x + 1)).
int dc_level
Predicted amount of DC noise, based on which a DC removal filter is used.
#define VLC_NBITS
number of bits to read per VLC iteration
static const float wmavoice_lsp16_intercoeff_a[32][2][16]
float cos[511]
8-bit cosine/sine windows over [-pi,pi] range
#define AV_CH_LAYOUT_MONO
av_cold int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans)
Set up a real FFT.
int aw_pulse_range
the range over which aw_pulse_set1() can apply the pulse, relative to the value in aw_first_pulse_off...
This structure stores compressed data.
int nb_samples
number of audio samples (per channel) described by this frame
float zero_exc_pf[MAX_SIGNAL_HISTORY+MAX_SFRAMESIZE]
zero filter output (i.e.
#define AV_CODEC_CAP_DR1
Codec uses get_buffer() for allocating buffers and supports custom allocators.
static void synth_block_fcb_acb(WMAVoiceContext *s, GetBitContext *gb, int block_idx, int size, int block_pitch_sh2, const struct frame_type_desc *frame_desc, float *excitation)
Parse FCB/ACB signal for a single block.
uint8_t dbl_pulses
how many pulse vectors have pulse pairs (rather than just one single pulse) only if fcb_type == FCB_T...
#define MAX_SIGNAL_HISTORY
maximum excitation signal history
uint16_t frame_size
the amount of bits that make up the block data (per frame)
GetBitContext gb
packet bitreader.