| ... | ... |
@@ -159,6 +159,8 @@ OBJS-$(CONFIG_FLIC_DECODER) += flicvideo.o |
| 159 | 159 |
OBJS-$(CONFIG_FOURXM_DECODER) += 4xm.o |
| 160 | 160 |
OBJS-$(CONFIG_FRAPS_DECODER) += fraps.o |
| 161 | 161 |
OBJS-$(CONFIG_FRWU_DECODER) += frwu.o |
| 162 |
+OBJS-$(CONFIG_G723_1_DECODER) += g723_1.o acelp_vectors.o \ |
|
| 163 |
+ celp_filters.o celp_math.o |
|
| 162 | 164 |
OBJS-$(CONFIG_G729_DECODER) += g729dec.o lsp.o celp_math.o acelp_filters.o acelp_pitch_delay.o acelp_vectors.o g729postfilter.o |
| 163 | 165 |
OBJS-$(CONFIG_GIF_DECODER) += gifdec.o lzw.o |
| 164 | 166 |
OBJS-$(CONFIG_GIF_ENCODER) += gif.o lzwenc.o |
| ... | ... |
@@ -258,6 +258,7 @@ void avcodec_register_all(void) |
| 258 | 258 |
REGISTER_DECODER (DSICINAUDIO, dsicinaudio); |
| 259 | 259 |
REGISTER_ENCDEC (EAC3, eac3); |
| 260 | 260 |
REGISTER_ENCDEC (FLAC, flac); |
| 261 |
+ REGISTER_DECODER (G723_1, g723_1); |
|
| 261 | 262 |
REGISTER_DECODER (G729, g729); |
| 262 | 263 |
REGISTER_DECODER (GSM, gsm); |
| 263 | 264 |
REGISTER_DECODER (GSM_MS, gsm_ms); |
| 351 | 352 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,1081 @@ |
| 0 |
+/* |
|
| 1 |
+ * G.723.1 compatible decoder |
|
| 2 |
+ * Copyright (c) 2006 Benjamin Larsson |
|
| 3 |
+ * Copyright (c) 2010 Mohamed Naufal Basheer |
|
| 4 |
+ * |
|
| 5 |
+ * This file is part of FFmpeg. |
|
| 6 |
+ * |
|
| 7 |
+ * FFmpeg is free software; you can redistribute it and/or |
|
| 8 |
+ * modify it under the terms of the GNU Lesser General Public |
|
| 9 |
+ * License as published by the Free Software Foundation; either |
|
| 10 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
| 11 |
+ * |
|
| 12 |
+ * FFmpeg is distributed in the hope that it will be useful, |
|
| 13 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
| 14 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
| 15 |
+ * Lesser General Public License for more details. |
|
| 16 |
+ * |
|
| 17 |
+ * You should have received a copy of the GNU Lesser General Public |
|
| 18 |
+ * License along with FFmpeg; if not, write to the Free Software |
|
| 19 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
| 20 |
+ */ |
|
| 21 |
+ |
|
| 22 |
+/** |
|
| 23 |
+ * @file |
|
| 24 |
+ * G.723.1 compatible decoder |
|
| 25 |
+ */ |
|
| 26 |
+ |
|
| 27 |
+#include "avcodec.h" |
|
| 28 |
+#define ALT_BITSTREAM_READER_LE |
|
| 29 |
+#include "get_bits.h" |
|
| 30 |
+#include "acelp_vectors.h" |
|
| 31 |
+#include "celp_filters.h" |
|
| 32 |
+#include "celp_math.h" |
|
| 33 |
+#include "lsp.h" |
|
| 34 |
+#include "libavutil/lzo.h" |
|
| 35 |
+#include "g723_1_data.h" |
|
| 36 |
+ |
|
| 37 |
+typedef struct g723_1_context {
|
|
| 38 |
+ G723_1_Subframe subframe[4]; |
|
| 39 |
+ FrameType cur_frame_type; |
|
| 40 |
+ FrameType past_frame_type; |
|
| 41 |
+ Rate cur_rate; |
|
| 42 |
+ uint8_t lsp_index[LSP_BANDS]; |
|
| 43 |
+ int pitch_lag[2]; |
|
| 44 |
+ int erased_frames; |
|
| 45 |
+ |
|
| 46 |
+ int16_t prev_lsp[LPC_ORDER]; |
|
| 47 |
+ int16_t prev_excitation[PITCH_MAX]; |
|
| 48 |
+ int16_t excitation[PITCH_MAX + FRAME_LEN]; |
|
| 49 |
+ int16_t synth_mem[LPC_ORDER]; |
|
| 50 |
+ int16_t fir_mem[LPC_ORDER]; |
|
| 51 |
+ int iir_mem[LPC_ORDER]; |
|
| 52 |
+ |
|
| 53 |
+ int random_seed; |
|
| 54 |
+ int interp_index; |
|
| 55 |
+ int interp_gain; |
|
| 56 |
+ int sid_gain; |
|
| 57 |
+ int cur_gain; |
|
| 58 |
+ int reflection_coef; |
|
| 59 |
+ int pf_gain; ///< formant postfilter |
|
| 60 |
+ ///< gain scaling unit memory |
|
| 61 |
+} G723_1_Context; |
|
| 62 |
+ |
|
| 63 |
+static av_cold int g723_1_decode_init(AVCodecContext *avctx) |
|
| 64 |
+{
|
|
| 65 |
+ G723_1_Context *p = avctx->priv_data; |
|
| 66 |
+ |
|
| 67 |
+ avctx->sample_fmt = SAMPLE_FMT_S16; |
|
| 68 |
+ p->pf_gain = 1 << 12; |
|
| 69 |
+ memcpy(p->prev_lsp, dc_lsp, LPC_ORDER * sizeof(int16_t)); |
|
| 70 |
+ |
|
| 71 |
+ return 0; |
|
| 72 |
+} |
|
| 73 |
+ |
|
| 74 |
+/** |
|
| 75 |
+ * Unpack the frame into parameters. |
|
| 76 |
+ * |
|
| 77 |
+ * @param p the context |
|
| 78 |
+ * @param buf pointer to the input buffer |
|
| 79 |
+ * @param buf_size size of the input buffer |
|
| 80 |
+ */ |
|
| 81 |
+static int unpack_bitstream(G723_1_Context *p, const uint8_t *buf, |
|
| 82 |
+ int buf_size) |
|
| 83 |
+{
|
|
| 84 |
+ GetBitContext gb; |
|
| 85 |
+ int ad_cb_len; |
|
| 86 |
+ int temp, info_bits, i; |
|
| 87 |
+ |
|
| 88 |
+ init_get_bits(&gb, buf, buf_size * 8); |
|
| 89 |
+ |
|
| 90 |
+ /* Extract frame type and rate info */ |
|
| 91 |
+ info_bits = get_bits(&gb, 2); |
|
| 92 |
+ |
|
| 93 |
+ if (info_bits == 3) {
|
|
| 94 |
+ p->cur_frame_type = UntransmittedFrame; |
|
| 95 |
+ return 0; |
|
| 96 |
+ } |
|
| 97 |
+ |
|
| 98 |
+ /* Extract 24 bit lsp indices, 8 bit for each band */ |
|
| 99 |
+ p->lsp_index[2] = get_bits(&gb, 8); |
|
| 100 |
+ p->lsp_index[1] = get_bits(&gb, 8); |
|
| 101 |
+ p->lsp_index[0] = get_bits(&gb, 8); |
|
| 102 |
+ |
|
| 103 |
+ if (info_bits == 2) {
|
|
| 104 |
+ p->cur_frame_type = SIDFrame; |
|
| 105 |
+ p->subframe[0].amp_index = get_bits(&gb, 6); |
|
| 106 |
+ return 0; |
|
| 107 |
+ } |
|
| 108 |
+ |
|
| 109 |
+ /* Extract the info common to both rates */ |
|
| 110 |
+ p->cur_rate = info_bits ? Rate5k3 : Rate6k3; |
|
| 111 |
+ p->cur_frame_type = ActiveFrame; |
|
| 112 |
+ |
|
| 113 |
+ p->pitch_lag[0] = get_bits(&gb, 7); |
|
| 114 |
+ if (p->pitch_lag[0] > 123) /* test if forbidden code */ |
|
| 115 |
+ return -1; |
|
| 116 |
+ p->pitch_lag[0] += PITCH_MIN; |
|
| 117 |
+ p->subframe[1].ad_cb_lag = get_bits(&gb, 2); |
|
| 118 |
+ |
|
| 119 |
+ p->pitch_lag[1] = get_bits(&gb, 7); |
|
| 120 |
+ if (p->pitch_lag[1] > 123) |
|
| 121 |
+ return -1; |
|
| 122 |
+ p->pitch_lag[1] += PITCH_MIN; |
|
| 123 |
+ p->subframe[3].ad_cb_lag = get_bits(&gb, 2); |
|
| 124 |
+ p->subframe[0].ad_cb_lag = 1; |
|
| 125 |
+ p->subframe[2].ad_cb_lag = 1; |
|
| 126 |
+ |
|
| 127 |
+ for (i = 0; i < SUBFRAMES; i++) {
|
|
| 128 |
+ /* Extract combined gain */ |
|
| 129 |
+ temp = get_bits(&gb, 12); |
|
| 130 |
+ ad_cb_len = 170; |
|
| 131 |
+ p->subframe[i].dirac_train = 0; |
|
| 132 |
+ if (p->cur_rate == Rate6k3 && p->pitch_lag[i >> 1] < SUBFRAME_LEN - 2) {
|
|
| 133 |
+ p->subframe[i].dirac_train = temp >> 11; |
|
| 134 |
+ temp &= 0x7ff; |
|
| 135 |
+ ad_cb_len = 85; |
|
| 136 |
+ } |
|
| 137 |
+ p->subframe[i].ad_cb_gain = FASTDIV(temp, GAIN_LEVELS); |
|
| 138 |
+ if (p->subframe[i].ad_cb_gain < ad_cb_len) {
|
|
| 139 |
+ p->subframe[i].amp_index = temp - p->subframe[i].ad_cb_gain * |
|
| 140 |
+ GAIN_LEVELS; |
|
| 141 |
+ } else {
|
|
| 142 |
+ return -1; |
|
| 143 |
+ } |
|
| 144 |
+ } |
|
| 145 |
+ |
|
| 146 |
+ p->subframe[0].grid_index = get_bits(&gb, 1); |
|
| 147 |
+ p->subframe[1].grid_index = get_bits(&gb, 1); |
|
| 148 |
+ p->subframe[2].grid_index = get_bits(&gb, 1); |
|
| 149 |
+ p->subframe[3].grid_index = get_bits(&gb, 1); |
|
| 150 |
+ |
|
| 151 |
+ if (p->cur_rate == Rate6k3) {
|
|
| 152 |
+ skip_bits(&gb, 1); /* skip reserved bit */ |
|
| 153 |
+ |
|
| 154 |
+ /* Compute pulse_pos index using the 13-bit combined position index */ |
|
| 155 |
+ temp = get_bits(&gb, 13); |
|
| 156 |
+ p->subframe[0].pulse_pos = temp / 810; |
|
| 157 |
+ |
|
| 158 |
+ temp -= p->subframe[0].pulse_pos * 810; |
|
| 159 |
+ p->subframe[1].pulse_pos = FASTDIV(temp, 90); |
|
| 160 |
+ |
|
| 161 |
+ temp -= p->subframe[1].pulse_pos * 90; |
|
| 162 |
+ p->subframe[2].pulse_pos = FASTDIV(temp, 9); |
|
| 163 |
+ p->subframe[3].pulse_pos = temp - p->subframe[2].pulse_pos * 9; |
|
| 164 |
+ |
|
| 165 |
+ p->subframe[0].pulse_pos = (p->subframe[0].pulse_pos << 16) + |
|
| 166 |
+ get_bits(&gb, 16); |
|
| 167 |
+ p->subframe[1].pulse_pos = (p->subframe[1].pulse_pos << 14) + |
|
| 168 |
+ get_bits(&gb, 14); |
|
| 169 |
+ p->subframe[2].pulse_pos = (p->subframe[2].pulse_pos << 16) + |
|
| 170 |
+ get_bits(&gb, 16); |
|
| 171 |
+ p->subframe[3].pulse_pos = (p->subframe[3].pulse_pos << 14) + |
|
| 172 |
+ get_bits(&gb, 14); |
|
| 173 |
+ |
|
| 174 |
+ p->subframe[0].pulse_sign = get_bits(&gb, 6); |
|
| 175 |
+ p->subframe[1].pulse_sign = get_bits(&gb, 5); |
|
| 176 |
+ p->subframe[2].pulse_sign = get_bits(&gb, 6); |
|
| 177 |
+ p->subframe[3].pulse_sign = get_bits(&gb, 5); |
|
| 178 |
+ } else { /* Rate5k3 */
|
|
| 179 |
+ p->subframe[0].pulse_pos = get_bits(&gb, 12); |
|
| 180 |
+ p->subframe[1].pulse_pos = get_bits(&gb, 12); |
|
| 181 |
+ p->subframe[2].pulse_pos = get_bits(&gb, 12); |
|
| 182 |
+ p->subframe[3].pulse_pos = get_bits(&gb, 12); |
|
| 183 |
+ |
|
| 184 |
+ p->subframe[0].pulse_sign = get_bits(&gb, 4); |
|
| 185 |
+ p->subframe[1].pulse_sign = get_bits(&gb, 4); |
|
| 186 |
+ p->subframe[2].pulse_sign = get_bits(&gb, 4); |
|
| 187 |
+ p->subframe[3].pulse_sign = get_bits(&gb, 4); |
|
| 188 |
+ } |
|
| 189 |
+ |
|
| 190 |
+ return 0; |
|
| 191 |
+} |
|
| 192 |
+ |
|
| 193 |
+/** |
|
| 194 |
+ * Bitexact implementation of sqrt(val/2). |
|
| 195 |
+ */ |
|
| 196 |
+static int16_t square_root(int val) |
|
| 197 |
+{
|
|
| 198 |
+ int16_t res = 0; |
|
| 199 |
+ int16_t exp = 0x4000; |
|
| 200 |
+ int i; |
|
| 201 |
+ |
|
| 202 |
+ for (i = 0; i < 14; i ++) {
|
|
| 203 |
+ int res_exp = res + exp; |
|
| 204 |
+ if (val >= res_exp * res_exp << 1) |
|
| 205 |
+ res += exp; |
|
| 206 |
+ exp >>= 1; |
|
| 207 |
+ } |
|
| 208 |
+ return res; |
|
| 209 |
+} |
|
| 210 |
+ |
|
| 211 |
+/** |
|
| 212 |
+ * Calculate the number of left-shifts required for normalizing the input. |
|
| 213 |
+ * |
|
| 214 |
+ * @param num input number |
|
| 215 |
+ * @param width width of the input, 16 bits(0) / 32 bits(1) |
|
| 216 |
+ */ |
|
| 217 |
+static int normalize_bits(int num, int width) |
|
| 218 |
+{
|
|
| 219 |
+ int i = 0; |
|
| 220 |
+ int bits = (width) ? 31 : 15; |
|
| 221 |
+ int limit = 1 << (bits - 1); |
|
| 222 |
+ |
|
| 223 |
+ if (num) {
|
|
| 224 |
+ if (num == -1) |
|
| 225 |
+ return bits; |
|
| 226 |
+ if (num < 0) |
|
| 227 |
+ num = ~num; |
|
| 228 |
+ for (i = 0; num < limit; i++) |
|
| 229 |
+ num <<= 1; |
|
| 230 |
+ } |
|
| 231 |
+ return i; |
|
| 232 |
+} |
|
| 233 |
+ |
|
| 234 |
+/** |
|
| 235 |
+ * Scale vector contents based on the largest of their absolutes. |
|
| 236 |
+ */ |
|
| 237 |
+static int scale_vector(int16_t *vector, int length) |
|
| 238 |
+{
|
|
| 239 |
+ int bits, scale, max = 0; |
|
| 240 |
+ int i; |
|
| 241 |
+ |
|
| 242 |
+ const int16_t shift_table[16] = {
|
|
| 243 |
+ 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, |
|
| 244 |
+ 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000, 0x7fff |
|
| 245 |
+ }; |
|
| 246 |
+ |
|
| 247 |
+ for (i = 0; i < length; i++) |
|
| 248 |
+ max = FFMAX(max, FFABS(vector[i])); |
|
| 249 |
+ |
|
| 250 |
+ bits = normalize_bits(max, 0); |
|
| 251 |
+ scale = shift_table[bits]; |
|
| 252 |
+ |
|
| 253 |
+ for (i = 0; i < length; i++) |
|
| 254 |
+ vector[i] = (int16_t)(av_clipl_int32(vector[i] * scale << 1) >> 4); |
|
| 255 |
+ |
|
| 256 |
+ return bits - 3; |
|
| 257 |
+} |
|
| 258 |
+ |
|
| 259 |
+/** |
|
| 260 |
+ * Perform inverse quantization of LSP frequencies. |
|
| 261 |
+ * |
|
| 262 |
+ * @param cur_lsp the current LSP vector |
|
| 263 |
+ * @param prev_lsp the previous LSP vector |
|
| 264 |
+ * @param lsp_index VQ indices |
|
| 265 |
+ * @param bad_frame bad frame flag |
|
| 266 |
+ */ |
|
| 267 |
+static void inverse_quant(int16_t *cur_lsp, int16_t *prev_lsp, |
|
| 268 |
+ uint8_t *lsp_index, int bad_frame) |
|
| 269 |
+{
|
|
| 270 |
+ int min_dist, pred; |
|
| 271 |
+ int i, j, temp, stable; |
|
| 272 |
+ |
|
| 273 |
+ /* Check for frame erasure */ |
|
| 274 |
+ if (!bad_frame) {
|
|
| 275 |
+ min_dist = 0x100; |
|
| 276 |
+ pred = 12288; |
|
| 277 |
+ } else {
|
|
| 278 |
+ min_dist = 0x200; |
|
| 279 |
+ pred = 23552; |
|
| 280 |
+ lsp_index[0] = lsp_index[1] = lsp_index[2] = 0; |
|
| 281 |
+ } |
|
| 282 |
+ |
|
| 283 |
+ /* Get the VQ table entry corresponding to the transmitted index */ |
|
| 284 |
+ cur_lsp[0] = lsp_band0[lsp_index[0]][0]; |
|
| 285 |
+ cur_lsp[1] = lsp_band0[lsp_index[0]][1]; |
|
| 286 |
+ cur_lsp[2] = lsp_band0[lsp_index[0]][2]; |
|
| 287 |
+ cur_lsp[3] = lsp_band1[lsp_index[1]][0]; |
|
| 288 |
+ cur_lsp[4] = lsp_band1[lsp_index[1]][1]; |
|
| 289 |
+ cur_lsp[5] = lsp_band1[lsp_index[1]][2]; |
|
| 290 |
+ cur_lsp[6] = lsp_band2[lsp_index[2]][0]; |
|
| 291 |
+ cur_lsp[7] = lsp_band2[lsp_index[2]][1]; |
|
| 292 |
+ cur_lsp[8] = lsp_band2[lsp_index[2]][2]; |
|
| 293 |
+ cur_lsp[9] = lsp_band2[lsp_index[2]][3]; |
|
| 294 |
+ |
|
| 295 |
+ /* Add predicted vector & DC component to the previously quantized vector */ |
|
| 296 |
+ for (i = 0; i < LPC_ORDER; i++) {
|
|
| 297 |
+ temp = ((prev_lsp[i] - dc_lsp[i]) * pred + (1 << 14)) >> 15; |
|
| 298 |
+ cur_lsp[i] += dc_lsp[i] + temp; |
|
| 299 |
+ } |
|
| 300 |
+ |
|
| 301 |
+ for (i = 0; i < LPC_ORDER; i++) {
|
|
| 302 |
+ cur_lsp[0] = FFMAX(cur_lsp[0], 0x180); |
|
| 303 |
+ cur_lsp[LPC_ORDER - 1] = FFMIN(cur_lsp[LPC_ORDER - 1], 0x7e00); |
|
| 304 |
+ |
|
| 305 |
+ /* Stability check */ |
|
| 306 |
+ for (j = 1; j < LPC_ORDER; j++) {
|
|
| 307 |
+ temp = min_dist + cur_lsp[j - 1] - cur_lsp[j]; |
|
| 308 |
+ if (temp > 0) {
|
|
| 309 |
+ temp >>= 1; |
|
| 310 |
+ cur_lsp[j - 1] -= temp; |
|
| 311 |
+ cur_lsp[j] += temp; |
|
| 312 |
+ } |
|
| 313 |
+ } |
|
| 314 |
+ stable = 1; |
|
| 315 |
+ for (j = 1; j < LPC_ORDER; j++) {
|
|
| 316 |
+ temp = cur_lsp[j - 1] + min_dist - cur_lsp[j] - 4; |
|
| 317 |
+ if (temp > 0) {
|
|
| 318 |
+ stable = 0; |
|
| 319 |
+ break; |
|
| 320 |
+ } |
|
| 321 |
+ } |
|
| 322 |
+ if (stable) |
|
| 323 |
+ break; |
|
| 324 |
+ } |
|
| 325 |
+ if (!stable) |
|
| 326 |
+ memcpy(cur_lsp, prev_lsp, LPC_ORDER * sizeof(int16_t)); |
|
| 327 |
+} |
|
| 328 |
+ |
|
| 329 |
+/** |
|
| 330 |
+ * Bitexact implementation of 2ab scaled by 1/2^16. |
|
| 331 |
+ * |
|
| 332 |
+ * @param a 32 bit multiplicand |
|
| 333 |
+ * @param b 16 bit multiplier |
|
| 334 |
+ */ |
|
| 335 |
+#define MULL2(a, b) \ |
|
| 336 |
+ ((((a) >> 16) * (b) << 1) + (((a) & 0xffff) * (b) >> 15)) |
|
| 337 |
+ |
|
| 338 |
+/** |
|
| 339 |
+ * Convert LSP frequencies to LPC coefficients. |
|
| 340 |
+ * |
|
| 341 |
+ * @param lpc buffer for LPC coefficients |
|
| 342 |
+ */ |
|
| 343 |
+static void lsp2lpc(int16_t *lpc) |
|
| 344 |
+{
|
|
| 345 |
+ int f1[LPC_ORDER / 2 + 1]; |
|
| 346 |
+ int f2[LPC_ORDER / 2 + 1]; |
|
| 347 |
+ int i, j; |
|
| 348 |
+ |
|
| 349 |
+ /* Calculate negative cosine */ |
|
| 350 |
+ for (j = 0; j < LPC_ORDER; j++) {
|
|
| 351 |
+ int index = lpc[j] >> 7; |
|
| 352 |
+ int offset = lpc[j] & 0x7f; |
|
| 353 |
+ int64_t temp1 = cos_tab[index] << 16; |
|
| 354 |
+ int temp2 = (cos_tab[index + 1] - cos_tab[index]) * |
|
| 355 |
+ ((offset << 8) + 0x80) << 1; |
|
| 356 |
+ |
|
| 357 |
+ lpc[j] = -(av_clipl_int32(((temp1 + temp2) << 1) + (1 << 15)) >> 16); |
|
| 358 |
+ } |
|
| 359 |
+ |
|
| 360 |
+ /* |
|
| 361 |
+ * Compute sum and difference polynomial coefficients |
|
| 362 |
+ * (bitexact alternative to lsp2poly() in lsp.c) |
|
| 363 |
+ */ |
|
| 364 |
+ /* Initialize with values in Q28 */ |
|
| 365 |
+ f1[0] = 1 << 28; |
|
| 366 |
+ f1[1] = (lpc[0] << 14) + (lpc[2] << 14); |
|
| 367 |
+ f1[2] = lpc[0] * lpc[2] + (2 << 28); |
|
| 368 |
+ |
|
| 369 |
+ f2[0] = 1 << 28; |
|
| 370 |
+ f2[1] = (lpc[1] << 14) + (lpc[3] << 14); |
|
| 371 |
+ f2[2] = lpc[1] * lpc[3] + (2 << 28); |
|
| 372 |
+ |
|
| 373 |
+ /* |
|
| 374 |
+ * Calculate and scale the coefficients by 1/2 in |
|
| 375 |
+ * each iteration for a final scaling factor of Q25 |
|
| 376 |
+ */ |
|
| 377 |
+ for (i = 2; i < LPC_ORDER / 2; i++) {
|
|
| 378 |
+ f1[i + 1] = f1[i - 1] + MULL2(f1[i], lpc[2 * i]); |
|
| 379 |
+ f2[i + 1] = f2[i - 1] + MULL2(f2[i], lpc[2 * i + 1]); |
|
| 380 |
+ |
|
| 381 |
+ for (j = i; j >= 2; j--) {
|
|
| 382 |
+ f1[j] = MULL2(f1[j - 1], lpc[2 * i]) + |
|
| 383 |
+ (f1[j] >> 1) + (f1[j - 2] >> 1); |
|
| 384 |
+ f2[j] = MULL2(f2[j - 1], lpc[2 * i + 1]) + |
|
| 385 |
+ (f2[j] >> 1) + (f2[j - 2] >> 1); |
|
| 386 |
+ } |
|
| 387 |
+ |
|
| 388 |
+ f1[0] >>= 1; |
|
| 389 |
+ f2[0] >>= 1; |
|
| 390 |
+ f1[1] = ((lpc[2 * i] << 16 >> i) + f1[1]) >> 1; |
|
| 391 |
+ f2[1] = ((lpc[2 * i + 1] << 16 >> i) + f2[1]) >> 1; |
|
| 392 |
+ } |
|
| 393 |
+ |
|
| 394 |
+ /* Convert polynomial coefficients to LPC coefficients */ |
|
| 395 |
+ for (i = 0; i < LPC_ORDER / 2; i++) {
|
|
| 396 |
+ int64_t ff1 = f1[i + 1] + f1[i]; |
|
| 397 |
+ int64_t ff2 = f2[i + 1] - f2[i]; |
|
| 398 |
+ |
|
| 399 |
+ lpc[i] = av_clipl_int32(((ff1 + ff2) << 3) + (1 << 15)) >> 16; |
|
| 400 |
+ lpc[LPC_ORDER - i - 1] = av_clipl_int32(((ff1 - ff2) << 3) + |
|
| 401 |
+ (1 << 15)) >> 16; |
|
| 402 |
+ } |
|
| 403 |
+} |
|
| 404 |
+ |
|
| 405 |
+/** |
|
| 406 |
+ * Quantize LSP frequencies by interpolation and convert them to |
|
| 407 |
+ * the corresponding LPC coefficients. |
|
| 408 |
+ * |
|
| 409 |
+ * @param lpc buffer for LPC coefficients |
|
| 410 |
+ * @param cur_lsp the current LSP vector |
|
| 411 |
+ * @param prev_lsp the previous LSP vector |
|
| 412 |
+ */ |
|
| 413 |
+static void lsp_interpolate(int16_t *lpc, int16_t *cur_lsp, int16_t *prev_lsp) |
|
| 414 |
+{
|
|
| 415 |
+ int i; |
|
| 416 |
+ int16_t *lpc_ptr = lpc; |
|
| 417 |
+ |
|
| 418 |
+ /* cur_lsp * 0.25 + prev_lsp * 0.75 */ |
|
| 419 |
+ ff_acelp_weighted_vector_sum(lpc, cur_lsp, prev_lsp, |
|
| 420 |
+ 4096, 12288, 1 << 13, 14, LPC_ORDER); |
|
| 421 |
+ ff_acelp_weighted_vector_sum(lpc + LPC_ORDER, cur_lsp, prev_lsp, |
|
| 422 |
+ 8192, 8192, 1 << 13, 14, LPC_ORDER); |
|
| 423 |
+ ff_acelp_weighted_vector_sum(lpc + 2 * LPC_ORDER, cur_lsp, prev_lsp, |
|
| 424 |
+ 12288, 4096, 1 << 13, 14, LPC_ORDER); |
|
| 425 |
+ memcpy(lpc + 3 * LPC_ORDER, cur_lsp, LPC_ORDER * sizeof(int16_t)); |
|
| 426 |
+ |
|
| 427 |
+ for (i = 0; i < SUBFRAMES; i++) {
|
|
| 428 |
+ lsp2lpc(lpc_ptr); |
|
| 429 |
+ lpc_ptr += LPC_ORDER; |
|
| 430 |
+ } |
|
| 431 |
+} |
|
| 432 |
+ |
|
| 433 |
+/** |
|
| 434 |
+ * Generate a train of dirac functions with period as pitch lag. |
|
| 435 |
+ */ |
|
| 436 |
+static void gen_dirac_train(int16_t *buf, int pitch_lag) |
|
| 437 |
+{
|
|
| 438 |
+ int16_t vector[SUBFRAME_LEN]; |
|
| 439 |
+ int i, j; |
|
| 440 |
+ |
|
| 441 |
+ memcpy(vector, buf, SUBFRAME_LEN * sizeof(int16_t)); |
|
| 442 |
+ for (i = pitch_lag; i < SUBFRAME_LEN; i += pitch_lag) {
|
|
| 443 |
+ for (j = 0; j < SUBFRAME_LEN - i; j++) |
|
| 444 |
+ buf[i + j] += vector[j]; |
|
| 445 |
+ } |
|
| 446 |
+} |
|
| 447 |
+ |
|
| 448 |
+/** |
|
| 449 |
+ * Generate fixed codebook excitation vector. |
|
| 450 |
+ * |
|
| 451 |
+ * @param vector decoded excitation vector |
|
| 452 |
+ * @param subfrm current subframe |
|
| 453 |
+ * @param cur_rate current bitrate |
|
| 454 |
+ * @param pitch_lag closed loop pitch lag |
|
| 455 |
+ * @param index current subframe index |
|
| 456 |
+ */ |
|
| 457 |
+static void gen_fcb_excitation(int16_t *vector, G723_1_Subframe subfrm, |
|
| 458 |
+ Rate cur_rate, int pitch_lag, int index) |
|
| 459 |
+{
|
|
| 460 |
+ int temp, i, j; |
|
| 461 |
+ |
|
| 462 |
+ memset(vector, 0, SUBFRAME_LEN * sizeof(int16_t)); |
|
| 463 |
+ |
|
| 464 |
+ if (cur_rate == Rate6k3) {
|
|
| 465 |
+ if (subfrm.pulse_pos >= max_pos[index]) |
|
| 466 |
+ return; |
|
| 467 |
+ |
|
| 468 |
+ /* Decode amplitudes and positions */ |
|
| 469 |
+ j = PULSE_MAX - pulses[index]; |
|
| 470 |
+ temp = subfrm.pulse_pos; |
|
| 471 |
+ for (i = 0; i < SUBFRAME_LEN / GRID_SIZE; i++) {
|
|
| 472 |
+ temp -= combinatorial_table[j][i]; |
|
| 473 |
+ if (temp >= 0) |
|
| 474 |
+ continue; |
|
| 475 |
+ temp += combinatorial_table[j++][i]; |
|
| 476 |
+ if (subfrm.pulse_sign & (1 << (PULSE_MAX - j))) {
|
|
| 477 |
+ vector[subfrm.grid_index + GRID_SIZE * i] = |
|
| 478 |
+ -fixed_cb_gain[subfrm.amp_index]; |
|
| 479 |
+ } else {
|
|
| 480 |
+ vector[subfrm.grid_index + GRID_SIZE * i] = |
|
| 481 |
+ fixed_cb_gain[subfrm.amp_index]; |
|
| 482 |
+ } |
|
| 483 |
+ if (j == PULSE_MAX) |
|
| 484 |
+ break; |
|
| 485 |
+ } |
|
| 486 |
+ if (subfrm.dirac_train == 1) |
|
| 487 |
+ gen_dirac_train(vector, pitch_lag); |
|
| 488 |
+ } else { /* Rate5k3 */
|
|
| 489 |
+ int cb_gain = fixed_cb_gain[subfrm.amp_index]; |
|
| 490 |
+ int cb_shift = subfrm.grid_index; |
|
| 491 |
+ int cb_sign = subfrm.pulse_sign; |
|
| 492 |
+ int cb_pos = subfrm.pulse_pos; |
|
| 493 |
+ int offset, beta, lag; |
|
| 494 |
+ |
|
| 495 |
+ for (i = 0; i < 8; i += 2) {
|
|
| 496 |
+ offset = ((cb_pos & 7) << 3) + cb_shift + i; |
|
| 497 |
+ vector[offset] = (cb_sign & 1) ? cb_gain : -cb_gain; |
|
| 498 |
+ cb_pos >>= 3; |
|
| 499 |
+ cb_sign >>= 1; |
|
| 500 |
+ } |
|
| 501 |
+ |
|
| 502 |
+ /* Enhance harmonic components */ |
|
| 503 |
+ lag = pitch_contrib[subfrm.ad_cb_gain << 1] + pitch_lag + |
|
| 504 |
+ subfrm.ad_cb_lag - 1; |
|
| 505 |
+ beta = pitch_contrib[(subfrm.ad_cb_gain << 1) + 1]; |
|
| 506 |
+ |
|
| 507 |
+ if (lag < SUBFRAME_LEN - 2) {
|
|
| 508 |
+ for (i = lag; i < SUBFRAME_LEN; i++) |
|
| 509 |
+ vector[i] += beta * vector[i - lag] >> 15; |
|
| 510 |
+ } |
|
| 511 |
+ } |
|
| 512 |
+} |
|
| 513 |
+ |
|
| 514 |
+/** |
|
| 515 |
+ * Get delayed contribution from the previous excitation vector. |
|
| 516 |
+ */ |
|
| 517 |
+static void get_residual(int16_t *residual, int16_t *prev_excitation, int lag) |
|
| 518 |
+{
|
|
| 519 |
+ int offset = PITCH_MAX - PITCH_ORDER / 2 - lag; |
|
| 520 |
+ int i; |
|
| 521 |
+ |
|
| 522 |
+ residual[0] = prev_excitation[offset]; |
|
| 523 |
+ residual[1] = prev_excitation[offset + 1]; |
|
| 524 |
+ |
|
| 525 |
+ offset += 2; |
|
| 526 |
+ for (i = 2; i < SUBFRAME_LEN + PITCH_ORDER - 1; i++) |
|
| 527 |
+ residual[i] = prev_excitation[offset + (i - 2) % lag]; |
|
| 528 |
+} |
|
| 529 |
+ |
|
| 530 |
+/** |
|
| 531 |
+ * Generate adaptive codebook excitation. |
|
| 532 |
+ */ |
|
| 533 |
+static void gen_acb_excitation(int16_t *vector, int16_t *prev_excitation, |
|
| 534 |
+ int pitch_lag, G723_1_Subframe subfrm, |
|
| 535 |
+ Rate cur_rate) |
|
| 536 |
+{
|
|
| 537 |
+ int16_t residual[SUBFRAME_LEN + PITCH_ORDER - 1]; |
|
| 538 |
+ const int16_t *cb_ptr; |
|
| 539 |
+ int lag = pitch_lag + subfrm.ad_cb_lag - 1; |
|
| 540 |
+ |
|
| 541 |
+ int i; |
|
| 542 |
+ int64_t sum; |
|
| 543 |
+ |
|
| 544 |
+ get_residual(residual, prev_excitation, lag); |
|
| 545 |
+ |
|
| 546 |
+ /* Select quantization table */ |
|
| 547 |
+ if (cur_rate == Rate6k3 && pitch_lag < SUBFRAME_LEN - 2) {
|
|
| 548 |
+ cb_ptr = adaptive_cb_gain85; |
|
| 549 |
+ } else |
|
| 550 |
+ cb_ptr = adaptive_cb_gain170; |
|
| 551 |
+ |
|
| 552 |
+ /* Calculate adaptive vector */ |
|
| 553 |
+ cb_ptr += subfrm.ad_cb_gain * 20; |
|
| 554 |
+ for (i = 0; i < SUBFRAME_LEN; i++) {
|
|
| 555 |
+ sum = ff_dot_product(residual + i, cb_ptr, PITCH_ORDER, 1); |
|
| 556 |
+ vector[i] = av_clipl_int32((sum << 1) + (1 << 15)) >> 16; |
|
| 557 |
+ } |
|
| 558 |
+} |
|
| 559 |
+ |
|
| 560 |
+/** |
|
| 561 |
+ * Estimate maximum auto-correlation around pitch lag. |
|
| 562 |
+ * |
|
| 563 |
+ * @param p the context |
|
| 564 |
+ * @param offset offset of the excitation vector |
|
| 565 |
+ * @param ccr_max pointer to the maximum auto-correlation |
|
| 566 |
+ * @param pitch_lag decoded pitch lag |
|
| 567 |
+ * @param length length of autocorrelation |
|
| 568 |
+ * @param dir forward lag(1) / backward lag(-1) |
|
| 569 |
+ */ |
|
| 570 |
+static int autocorr_max(G723_1_Context *p, int offset, int *ccr_max, |
|
| 571 |
+ int pitch_lag, int length, int dir) |
|
| 572 |
+{
|
|
| 573 |
+ int limit, ccr, lag = 0; |
|
| 574 |
+ int16_t *buf = p->excitation + offset; |
|
| 575 |
+ int i; |
|
| 576 |
+ |
|
| 577 |
+ pitch_lag = FFMIN(PITCH_MAX - 3, pitch_lag); |
|
| 578 |
+ limit = FFMIN(FRAME_LEN + PITCH_MAX - offset - length, pitch_lag + 3); |
|
| 579 |
+ |
|
| 580 |
+ for (i = pitch_lag - 3; i <= limit; i++) {
|
|
| 581 |
+ ccr = ff_dot_product(buf, buf + dir * i, length, 1); |
|
| 582 |
+ |
|
| 583 |
+ if (ccr > *ccr_max) {
|
|
| 584 |
+ *ccr_max = ccr; |
|
| 585 |
+ lag = i; |
|
| 586 |
+ } |
|
| 587 |
+ } |
|
| 588 |
+ return lag; |
|
| 589 |
+} |
|
| 590 |
+ |
|
| 591 |
+/** |
|
| 592 |
+ * Calculate pitch postfilter optimal and scaling gains. |
|
| 593 |
+ * |
|
| 594 |
+ * @param lag pitch postfilter forward/backward lag |
|
| 595 |
+ * @param ppf pitch postfilter parameters |
|
| 596 |
+ * @param cur_rate current bitrate |
|
| 597 |
+ * @param tgt_eng target energy |
|
| 598 |
+ * @param ccr cross-correlation |
|
| 599 |
+ * @param res_eng residual energy |
|
| 600 |
+ */ |
|
| 601 |
+static void comp_ppf_gains(int lag, PPFParam *ppf, Rate cur_rate, |
|
| 602 |
+ int tgt_eng, int ccr, int res_eng) |
|
| 603 |
+{
|
|
| 604 |
+ int pf_residual; /* square of postfiltered residual */ |
|
| 605 |
+ int64_t temp1, temp2; |
|
| 606 |
+ |
|
| 607 |
+ ppf->index = lag; |
|
| 608 |
+ |
|
| 609 |
+ temp1 = tgt_eng * res_eng >> 1; |
|
| 610 |
+ temp2 = ccr * ccr << 1; |
|
| 611 |
+ |
|
| 612 |
+ if (temp2 > temp1) {
|
|
| 613 |
+ if (ccr >= res_eng) {
|
|
| 614 |
+ ppf->opt_gain = ppf_gain_weight[cur_rate]; |
|
| 615 |
+ } else {
|
|
| 616 |
+ ppf->opt_gain = (ccr << 15) / res_eng * |
|
| 617 |
+ ppf_gain_weight[cur_rate] >> 15; |
|
| 618 |
+ } |
|
| 619 |
+ /* pf_res^2 = tgt_eng + 2*ccr*gain + res_eng*gain^2 */ |
|
| 620 |
+ temp1 = (tgt_eng << 15) + (ccr * ppf->opt_gain << 1); |
|
| 621 |
+ temp2 = (ppf->opt_gain * ppf->opt_gain >> 15) * res_eng; |
|
| 622 |
+ pf_residual = av_clipl_int32(temp1 + temp2 + (1 << 15)) >> 16; |
|
| 623 |
+ |
|
| 624 |
+ if (tgt_eng >= pf_residual << 1) {
|
|
| 625 |
+ temp1 = 0x7fff; |
|
| 626 |
+ } else {
|
|
| 627 |
+ temp1 = (tgt_eng << 14) / pf_residual; |
|
| 628 |
+ } |
|
| 629 |
+ |
|
| 630 |
+ /* scaling_gain = sqrt(tgt_eng/pf_res^2) */ |
|
| 631 |
+ ppf->sc_gain = square_root(temp1 << 16); |
|
| 632 |
+ } else {
|
|
| 633 |
+ ppf->opt_gain = 0; |
|
| 634 |
+ ppf->sc_gain = 0x7fff; |
|
| 635 |
+ } |
|
| 636 |
+ |
|
| 637 |
+ ppf->opt_gain = av_clip_int16(ppf->opt_gain * ppf->sc_gain >> 15); |
|
| 638 |
+} |
|
| 639 |
+ |
|
| 640 |
+/** |
|
| 641 |
+ * Calculate pitch postfilter parameters. |
|
| 642 |
+ * |
|
| 643 |
+ * @param p the context |
|
| 644 |
+ * @param offset offset of the excitation vector |
|
| 645 |
+ * @param pitch_lag decoded pitch lag |
|
| 646 |
+ * @param ppf pitch postfilter parameters |
|
| 647 |
+ * @param cur_rate current bitrate |
|
| 648 |
+ */ |
|
| 649 |
+static void comp_ppf_coeff(G723_1_Context *p, int offset, int pitch_lag, |
|
| 650 |
+ PPFParam *ppf, Rate cur_rate) |
|
| 651 |
+{
|
|
| 652 |
+ |
|
| 653 |
+ int16_t scale; |
|
| 654 |
+ int i; |
|
| 655 |
+ int64_t temp1, temp2; |
|
| 656 |
+ |
|
| 657 |
+ /* |
|
| 658 |
+ * 0 - target energy |
|
| 659 |
+ * 1 - forward cross-correlation |
|
| 660 |
+ * 2 - forward residual energy |
|
| 661 |
+ * 3 - backward cross-correlation |
|
| 662 |
+ * 4 - backward residual energy |
|
| 663 |
+ */ |
|
| 664 |
+ int energy[5] = {0, 0, 0, 0, 0};
|
|
| 665 |
+ int16_t *buf = p->excitation + offset; |
|
| 666 |
+ int fwd_lag = autocorr_max(p, offset, &energy[1], pitch_lag, |
|
| 667 |
+ SUBFRAME_LEN, 1); |
|
| 668 |
+ int back_lag = autocorr_max(p, offset, &energy[3], pitch_lag, |
|
| 669 |
+ SUBFRAME_LEN, -1); |
|
| 670 |
+ |
|
| 671 |
+ ppf->index = 0; |
|
| 672 |
+ ppf->opt_gain = 0; |
|
| 673 |
+ ppf->sc_gain = 0x7fff; |
|
| 674 |
+ |
|
| 675 |
+ /* Case 0, Section 3.6 */ |
|
| 676 |
+ if (!back_lag && !fwd_lag) |
|
| 677 |
+ return; |
|
| 678 |
+ |
|
| 679 |
+ /* Compute target energy */ |
|
| 680 |
+ energy[0] = ff_dot_product(buf, buf, SUBFRAME_LEN, 1); |
|
| 681 |
+ |
|
| 682 |
+ /* Compute forward residual energy */ |
|
| 683 |
+ if (fwd_lag) |
|
| 684 |
+ energy[2] = ff_dot_product(buf + fwd_lag, buf + fwd_lag, |
|
| 685 |
+ SUBFRAME_LEN, 1); |
|
| 686 |
+ |
|
| 687 |
+ /* Compute backward residual energy */ |
|
| 688 |
+ if (back_lag) |
|
| 689 |
+ energy[4] = ff_dot_product(buf - back_lag, buf - back_lag, |
|
| 690 |
+ SUBFRAME_LEN, 1); |
|
| 691 |
+ |
|
| 692 |
+ /* Normalize and shorten */ |
|
| 693 |
+ temp1 = 0; |
|
| 694 |
+ for (i = 0; i < 5; i++) |
|
| 695 |
+ temp1 = FFMAX(energy[i], temp1); |
|
| 696 |
+ |
|
| 697 |
+ scale = normalize_bits(temp1, 1); |
|
| 698 |
+ for (i = 0; i < 5; i++) |
|
| 699 |
+ energy[i] = av_clipl_int32(energy[i] << scale) >> 16; |
|
| 700 |
+ |
|
| 701 |
+ if (fwd_lag && !back_lag) { /* Case 1 */
|
|
| 702 |
+ comp_ppf_gains(fwd_lag, ppf, cur_rate, energy[0], energy[1], |
|
| 703 |
+ energy[2]); |
|
| 704 |
+ } else if (!fwd_lag) { /* Case 2 */
|
|
| 705 |
+ comp_ppf_gains(-back_lag, ppf, cur_rate, energy[0], energy[3], |
|
| 706 |
+ energy[4]); |
|
| 707 |
+ } else { /* Case 3 */
|
|
| 708 |
+ |
|
| 709 |
+ /* |
|
| 710 |
+ * Select the largest of energy[1]^2/energy[2] |
|
| 711 |
+ * and energy[3]^2/energy[4] |
|
| 712 |
+ */ |
|
| 713 |
+ temp1 = energy[4] * ((energy[1] * energy[1] + (1 << 14)) >> 15); |
|
| 714 |
+ temp2 = energy[2] * ((energy[3] * energy[3] + (1 << 14)) >> 15); |
|
| 715 |
+ if (temp1 >= temp2) {
|
|
| 716 |
+ comp_ppf_gains(fwd_lag, ppf, cur_rate, energy[0], energy[1], |
|
| 717 |
+ energy[2]); |
|
| 718 |
+ } else {
|
|
| 719 |
+ comp_ppf_gains(-back_lag, ppf, cur_rate, energy[0], energy[3], |
|
| 720 |
+ energy[4]); |
|
| 721 |
+ } |
|
| 722 |
+ } |
|
| 723 |
+} |
|
| 724 |
+ |
|
| 725 |
+/** |
|
| 726 |
+ * Classify frames as voiced/unvoiced. |
|
| 727 |
+ * |
|
| 728 |
+ * @param p the context |
|
| 729 |
+ * @param pitch_lag decoded pitch_lag |
|
| 730 |
+ * @param exc_eng excitation energy estimation |
|
| 731 |
+ * @param scale scaling factor of exc_eng |
|
| 732 |
+ * |
|
| 733 |
+ * @return residual interpolation index if voiced, 0 otherwise |
|
| 734 |
+ */ |
|
| 735 |
+static int comp_interp_index(G723_1_Context *p, int pitch_lag, |
|
| 736 |
+ int *exc_eng, int *scale) |
|
| 737 |
+{
|
|
| 738 |
+ int offset = PITCH_MAX + 2 * SUBFRAME_LEN; |
|
| 739 |
+ int16_t *buf = p->excitation + offset; |
|
| 740 |
+ |
|
| 741 |
+ int index, ccr, tgt_eng, best_eng, temp; |
|
| 742 |
+ |
|
| 743 |
+ *scale = scale_vector(p->excitation, FRAME_LEN + PITCH_MAX); |
|
| 744 |
+ |
|
| 745 |
+ /* Compute maximum backward cross-correlation */ |
|
| 746 |
+ ccr = 0; |
|
| 747 |
+ index = autocorr_max(p, offset, &ccr, pitch_lag, SUBFRAME_LEN * 2, -1); |
|
| 748 |
+ ccr = av_clipl_int32((int64_t)ccr + (1 << 15)) >> 16; |
|
| 749 |
+ |
|
| 750 |
+ /* Compute target energy */ |
|
| 751 |
+ tgt_eng = ff_dot_product(buf, buf, SUBFRAME_LEN * 2, 1); |
|
| 752 |
+ *exc_eng = av_clipl_int32(tgt_eng + (1 << 15)) >> 16; |
|
| 753 |
+ |
|
| 754 |
+ if (ccr <= 0) |
|
| 755 |
+ return 0; |
|
| 756 |
+ |
|
| 757 |
+ /* Compute best energy */ |
|
| 758 |
+ best_eng = ff_dot_product(buf - index, buf - index, |
|
| 759 |
+ SUBFRAME_LEN * 2, 1); |
|
| 760 |
+ best_eng = av_clipl_int32((int64_t)best_eng + (1 << 15)) >> 16; |
|
| 761 |
+ |
|
| 762 |
+ temp = best_eng * *exc_eng >> 3; |
|
| 763 |
+ |
|
| 764 |
+ if (temp < ccr * ccr) {
|
|
| 765 |
+ return index; |
|
| 766 |
+ } else |
|
| 767 |
+ return 0; |
|
| 768 |
+} |
|
| 769 |
+ |
|
| 770 |
+/** |
|
| 771 |
+ * Peform residual interpolation based on frame classification. |
|
| 772 |
+ * |
|
| 773 |
+ * @param buf decoded excitation vector |
|
| 774 |
+ * @param out output vector |
|
| 775 |
+ * @param lag decoded pitch lag |
|
| 776 |
+ * @param gain interpolated gain |
|
| 777 |
+ * @param rseed seed for random number generator |
|
| 778 |
+ */ |
|
| 779 |
+static void residual_interp(int16_t *buf, int16_t *out, int lag, |
|
| 780 |
+ int gain, int *rseed) |
|
| 781 |
+{
|
|
| 782 |
+ int i; |
|
| 783 |
+ if (lag) { /* Voiced */
|
|
| 784 |
+ int16_t *vector_ptr = buf + PITCH_MAX; |
|
| 785 |
+ /* Attenuate */ |
|
| 786 |
+ for (i = 0; i < lag; i++) |
|
| 787 |
+ vector_ptr[i - lag] = vector_ptr[i - lag] * 3 >> 2; |
|
| 788 |
+ av_memcpy_backptr((uint8_t*)vector_ptr, lag * sizeof(int16_t), |
|
| 789 |
+ FRAME_LEN * sizeof(int16_t)); |
|
| 790 |
+ memcpy(out, vector_ptr, FRAME_LEN * sizeof(int16_t)); |
|
| 791 |
+ } else { /* Unvoiced */
|
|
| 792 |
+ for (i = 0; i < FRAME_LEN; i++) {
|
|
| 793 |
+ *rseed = *rseed * 521 + 259; |
|
| 794 |
+ out[i] = gain * *rseed >> 15; |
|
| 795 |
+ } |
|
| 796 |
+ memset(buf, 0, (FRAME_LEN + PITCH_MAX) * sizeof(int16_t)); |
|
| 797 |
+ } |
|
| 798 |
+} |
|
| 799 |
+ |
|
| 800 |
+/** |
|
| 801 |
+ * Perform IIR filtering. |
|
| 802 |
+ * |
|
| 803 |
+ * @param fir_coef FIR coefficients |
|
| 804 |
+ * @param iir_coef IIR coefficients |
|
| 805 |
+ * @param src source vector |
|
| 806 |
+ * @param dest destination vector |
|
| 807 |
+ * @param width width of the output, 16 bits(0) / 32 bits(1) |
|
| 808 |
+ */ |
|
| 809 |
+#define iir_filter(fir_coef, iir_coef, src, dest, width)\ |
|
| 810 |
+{\
|
|
| 811 |
+ int m, n;\ |
|
| 812 |
+ int res_shift = 16 & ~-(width);\ |
|
| 813 |
+ int in_shift = 16 - res_shift;\ |
|
| 814 |
+\ |
|
| 815 |
+ for (m = 0; m < SUBFRAME_LEN; m++) {\
|
|
| 816 |
+ int64_t filter = 0;\ |
|
| 817 |
+ for (n = 1; n <= LPC_ORDER; n++) {\
|
|
| 818 |
+ filter -= (fir_coef)[n - 1] * (src)[m - n] -\ |
|
| 819 |
+ (iir_coef)[n - 1] * ((dest)[m - n] >> in_shift);\ |
|
| 820 |
+ }\ |
|
| 821 |
+\ |
|
| 822 |
+ (dest)[m] = av_clipl_int32(((src)[m] << 16) + (filter << 3) +\ |
|
| 823 |
+ (1 << 15)) >> res_shift;\ |
|
| 824 |
+ }\ |
|
| 825 |
+} |
|
| 826 |
+ |
|
| 827 |
+/** |
|
| 828 |
+ * Adjust gain of postfiltered signal. |
|
| 829 |
+ * |
|
| 830 |
+ * @param p the context |
|
| 831 |
+ * @param buf postfiltered output vector |
|
| 832 |
+ * @param energy input energy coefficient |
|
| 833 |
+ */ |
|
| 834 |
+static void gain_scale(G723_1_Context *p, int16_t * buf, int energy) |
|
| 835 |
+{
|
|
| 836 |
+ int num, denom, gain, bits1, bits2; |
|
| 837 |
+ int i; |
|
| 838 |
+ |
|
| 839 |
+ num = energy; |
|
| 840 |
+ denom = 0; |
|
| 841 |
+ for (i = 0; i < SUBFRAME_LEN; i++) {
|
|
| 842 |
+ int64_t temp = buf[i] >> 2; |
|
| 843 |
+ temp = av_clipl_int32(MUL64(temp, temp) << 1); |
|
| 844 |
+ denom = av_clipl_int32(denom + temp); |
|
| 845 |
+ } |
|
| 846 |
+ |
|
| 847 |
+ if (num && denom) {
|
|
| 848 |
+ bits1 = normalize_bits(num, 1); |
|
| 849 |
+ bits2 = normalize_bits(denom, 1); |
|
| 850 |
+ num = num << bits1 >> 1; |
|
| 851 |
+ denom <<= bits2; |
|
| 852 |
+ |
|
| 853 |
+ bits2 = 5 + bits1 - bits2; |
|
| 854 |
+ bits2 = FFMAX(0, bits2); |
|
| 855 |
+ |
|
| 856 |
+ gain = (num >> 1) / (denom >> 16); |
|
| 857 |
+ gain = square_root(gain << 16 >> bits2); |
|
| 858 |
+ } else {
|
|
| 859 |
+ gain = 1 << 12; |
|
| 860 |
+ } |
|
| 861 |
+ |
|
| 862 |
+ for (i = 0; i < SUBFRAME_LEN; i++) {
|
|
| 863 |
+ p->pf_gain = ((p->pf_gain << 4) - p->pf_gain + gain + (1 << 3)) >> 4; |
|
| 864 |
+ buf[i] = av_clip_int16((buf[i] * (p->pf_gain + (p->pf_gain >> 4)) + |
|
| 865 |
+ (1 << 10)) >> 11); |
|
| 866 |
+ } |
|
| 867 |
+} |
|
| 868 |
+ |
|
| 869 |
+/** |
|
| 870 |
+ * Perform formant filtering. |
|
| 871 |
+ * |
|
| 872 |
+ * @param p the context |
|
| 873 |
+ * @param lpc quantized lpc coefficients |
|
| 874 |
+ * @param buf output buffer |
|
| 875 |
+ */ |
|
| 876 |
+static void formant_postfilter(G723_1_Context *p, int16_t *lpc, int16_t *buf) |
|
| 877 |
+{
|
|
| 878 |
+ int16_t filter_coef[2][LPC_ORDER], *buf_ptr; |
|
| 879 |
+ int filter_signal[LPC_ORDER + FRAME_LEN], *signal_ptr; |
|
| 880 |
+ int i, j, k; |
|
| 881 |
+ |
|
| 882 |
+ memcpy(buf, p->fir_mem, LPC_ORDER * sizeof(int16_t)); |
|
| 883 |
+ memcpy(filter_signal, p->iir_mem, LPC_ORDER * sizeof(int)); |
|
| 884 |
+ |
|
| 885 |
+ for (i = LPC_ORDER, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++) {
|
|
| 886 |
+ for (k = 0; k < LPC_ORDER; k++) {
|
|
| 887 |
+ filter_coef[0][k] = (-lpc[k] * postfilter_tbl[0][k] + |
|
| 888 |
+ (1 << 14)) >> 15; |
|
| 889 |
+ filter_coef[1][k] = (-lpc[k] * postfilter_tbl[1][k] + |
|
| 890 |
+ (1 << 14)) >> 15; |
|
| 891 |
+ } |
|
| 892 |
+ iir_filter(filter_coef[0], filter_coef[1], buf + i, |
|
| 893 |
+ filter_signal + i, 1); |
|
| 894 |
+ } |
|
| 895 |
+ |
|
| 896 |
+ memcpy(p->fir_mem, buf + FRAME_LEN, LPC_ORDER * sizeof(int16_t)); |
|
| 897 |
+ memcpy(p->iir_mem, filter_signal + FRAME_LEN, LPC_ORDER * sizeof(int)); |
|
| 898 |
+ |
|
| 899 |
+ buf_ptr = buf + LPC_ORDER; |
|
| 900 |
+ signal_ptr = filter_signal + LPC_ORDER; |
|
| 901 |
+ for (i = 0; i < SUBFRAMES; i++) {
|
|
| 902 |
+ int16_t temp_vector[SUBFRAME_LEN]; |
|
| 903 |
+ int16_t temp; |
|
| 904 |
+ int auto_corr[2]; |
|
| 905 |
+ int scale, energy; |
|
| 906 |
+ |
|
| 907 |
+ /* Normalize */ |
|
| 908 |
+ memcpy(temp_vector, buf_ptr, SUBFRAME_LEN * sizeof(int16_t)); |
|
| 909 |
+ scale = scale_vector(temp_vector, SUBFRAME_LEN); |
|
| 910 |
+ |
|
| 911 |
+ /* Compute auto correlation coefficients */ |
|
| 912 |
+ auto_corr[0] = ff_dot_product(temp_vector, temp_vector + 1, |
|
| 913 |
+ SUBFRAME_LEN - 1, 1); |
|
| 914 |
+ auto_corr[1] = ff_dot_product(temp_vector, temp_vector, |
|
| 915 |
+ SUBFRAME_LEN, 1); |
|
| 916 |
+ |
|
| 917 |
+ /* Compute reflection coefficient */ |
|
| 918 |
+ temp = auto_corr[1] >> 16; |
|
| 919 |
+ if (temp) {
|
|
| 920 |
+ temp = (auto_corr[0] >> 2) / temp; |
|
| 921 |
+ } |
|
| 922 |
+ p->reflection_coef = ((p->reflection_coef << 2) - p->reflection_coef + |
|
| 923 |
+ temp + 2) >> 2; |
|
| 924 |
+ temp = (p->reflection_coef * 0xffffc >> 3) & 0xfffc; |
|
| 925 |
+ |
|
| 926 |
+ /* Compensation filter */ |
|
| 927 |
+ for (j = 0; j < SUBFRAME_LEN; j++) {
|
|
| 928 |
+ buf_ptr[j] = av_clipl_int32(signal_ptr[j] + |
|
| 929 |
+ ((signal_ptr[j - 1] >> 16) * |
|
| 930 |
+ temp << 1)) >> 16; |
|
| 931 |
+ } |
|
| 932 |
+ |
|
| 933 |
+ /* Compute normalized signal energy */ |
|
| 934 |
+ temp = 2 * scale + 4; |
|
| 935 |
+ if (temp < 0) {
|
|
| 936 |
+ energy = av_clipl_int32((int64_t)auto_corr[1] << -temp); |
|
| 937 |
+ } else |
|
| 938 |
+ energy = auto_corr[1] >> temp; |
|
| 939 |
+ |
|
| 940 |
+ gain_scale(p, buf_ptr, energy); |
|
| 941 |
+ |
|
| 942 |
+ buf_ptr += SUBFRAME_LEN; |
|
| 943 |
+ signal_ptr += SUBFRAME_LEN; |
|
| 944 |
+ } |
|
| 945 |
+} |
|
| 946 |
+ |
|
| 947 |
+static int g723_1_decode_frame(AVCodecContext *avctx, void *data, |
|
| 948 |
+ int *data_size, AVPacket *avpkt) |
|
| 949 |
+{
|
|
| 950 |
+ G723_1_Context *p = avctx->priv_data; |
|
| 951 |
+ const uint8_t *buf = avpkt->data; |
|
| 952 |
+ int buf_size = avpkt->size; |
|
| 953 |
+ int16_t *out = data; |
|
| 954 |
+ int dec_mode = buf[0] & 3; |
|
| 955 |
+ |
|
| 956 |
+ PPFParam ppf[SUBFRAMES]; |
|
| 957 |
+ int16_t cur_lsp[LPC_ORDER]; |
|
| 958 |
+ int16_t lpc[SUBFRAMES * LPC_ORDER]; |
|
| 959 |
+ int16_t acb_vector[SUBFRAME_LEN]; |
|
| 960 |
+ int16_t *vector_ptr; |
|
| 961 |
+ int bad_frame = 0, i, j; |
|
| 962 |
+ |
|
| 963 |
+ if (!buf_size || buf_size < frame_size[dec_mode]) {
|
|
| 964 |
+ *data_size = 0; |
|
| 965 |
+ return buf_size; |
|
| 966 |
+ } |
|
| 967 |
+ |
|
| 968 |
+ if (unpack_bitstream(p, buf, buf_size) < 0) {
|
|
| 969 |
+ bad_frame = 1; |
|
| 970 |
+ p->cur_frame_type = p->past_frame_type == ActiveFrame ? |
|
| 971 |
+ ActiveFrame : UntransmittedFrame; |
|
| 972 |
+ } |
|
| 973 |
+ |
|
| 974 |
+ *data_size = FRAME_LEN * sizeof(int16_t); |
|
| 975 |
+ if(p->cur_frame_type == ActiveFrame) {
|
|
| 976 |
+ if (!bad_frame) {
|
|
| 977 |
+ p->erased_frames = 0; |
|
| 978 |
+ } else if(p->erased_frames != 3) |
|
| 979 |
+ p->erased_frames++; |
|
| 980 |
+ |
|
| 981 |
+ inverse_quant(cur_lsp, p->prev_lsp, p->lsp_index, bad_frame); |
|
| 982 |
+ lsp_interpolate(lpc, cur_lsp, p->prev_lsp); |
|
| 983 |
+ |
|
| 984 |
+ /* Save the lsp_vector for the next frame */ |
|
| 985 |
+ memcpy(p->prev_lsp, cur_lsp, LPC_ORDER * sizeof(int16_t)); |
|
| 986 |
+ |
|
| 987 |
+ /* Generate the excitation for the frame */ |
|
| 988 |
+ memcpy(p->excitation, p->prev_excitation, PITCH_MAX * sizeof(int16_t)); |
|
| 989 |
+ vector_ptr = p->excitation + PITCH_MAX; |
|
| 990 |
+ if (!p->erased_frames) {
|
|
| 991 |
+ /* Update interpolation gain memory */ |
|
| 992 |
+ p->interp_gain = fixed_cb_gain[(p->subframe[2].amp_index + |
|
| 993 |
+ p->subframe[3].amp_index) >> 1]; |
|
| 994 |
+ for (i = 0; i < SUBFRAMES; i++) {
|
|
| 995 |
+ gen_fcb_excitation(vector_ptr, p->subframe[i], p->cur_rate, |
|
| 996 |
+ p->pitch_lag[i >> 1], i); |
|
| 997 |
+ gen_acb_excitation(acb_vector, &p->excitation[SUBFRAME_LEN * i], |
|
| 998 |
+ p->pitch_lag[i >> 1], p->subframe[i], |
|
| 999 |
+ p->cur_rate); |
|
| 1000 |
+ /* Get the total excitation */ |
|
| 1001 |
+ for (j = 0; j < SUBFRAME_LEN; j++) {
|
|
| 1002 |
+ vector_ptr[j] = av_clip_int16(vector_ptr[j] << 1); |
|
| 1003 |
+ vector_ptr[j] = av_clip_int16(vector_ptr[j] + |
|
| 1004 |
+ acb_vector[j]); |
|
| 1005 |
+ } |
|
| 1006 |
+ vector_ptr += SUBFRAME_LEN; |
|
| 1007 |
+ } |
|
| 1008 |
+ |
|
| 1009 |
+ vector_ptr = p->excitation + PITCH_MAX; |
|
| 1010 |
+ |
|
| 1011 |
+ /* Save the excitation */ |
|
| 1012 |
+ memcpy(out, vector_ptr, FRAME_LEN * sizeof(int16_t)); |
|
| 1013 |
+ |
|
| 1014 |
+ p->interp_index = comp_interp_index(p, p->pitch_lag[1], |
|
| 1015 |
+ &p->sid_gain, &p->cur_gain); |
|
| 1016 |
+ |
|
| 1017 |
+ for (i = PITCH_MAX, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++) |
|
| 1018 |
+ comp_ppf_coeff(p, i, p->pitch_lag[j >> 1], |
|
| 1019 |
+ ppf + j, p->cur_rate); |
|
| 1020 |
+ |
|
| 1021 |
+ /* Restore the original excitation */ |
|
| 1022 |
+ memcpy(p->excitation, p->prev_excitation, |
|
| 1023 |
+ PITCH_MAX * sizeof(int16_t)); |
|
| 1024 |
+ memcpy(vector_ptr, out, FRAME_LEN * sizeof(int16_t)); |
|
| 1025 |
+ |
|
| 1026 |
+ /* Peform pitch postfiltering */ |
|
| 1027 |
+ for (i = 0, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++) |
|
| 1028 |
+ ff_acelp_weighted_vector_sum(out + LPC_ORDER + i, vector_ptr + i, |
|
| 1029 |
+ vector_ptr + i + ppf[j].index, |
|
| 1030 |
+ ppf[j].sc_gain, ppf[j].opt_gain, |
|
| 1031 |
+ 1 << 14, 15, SUBFRAME_LEN); |
|
| 1032 |
+ } else {
|
|
| 1033 |
+ p->interp_gain = (p->interp_gain * 3 + 2) >> 2; |
|
| 1034 |
+ if (p->erased_frames == 3) {
|
|
| 1035 |
+ /* Mute output */ |
|
| 1036 |
+ memset(p->excitation, 0, |
|
| 1037 |
+ (FRAME_LEN + PITCH_MAX) * sizeof(int16_t)); |
|
| 1038 |
+ memset(out, 0, (FRAME_LEN + LPC_ORDER) * sizeof(int16_t)); |
|
| 1039 |
+ } else {
|
|
| 1040 |
+ /* Regenerate frame */ |
|
| 1041 |
+ residual_interp(p->excitation, out + LPC_ORDER, p->interp_index, |
|
| 1042 |
+ p->interp_gain, &p->random_seed); |
|
| 1043 |
+ } |
|
| 1044 |
+ } |
|
| 1045 |
+ /* Save the excitation for the next frame */ |
|
| 1046 |
+ memcpy(p->prev_excitation, p->excitation + FRAME_LEN, |
|
| 1047 |
+ PITCH_MAX * sizeof(int16_t)); |
|
| 1048 |
+ } else {
|
|
| 1049 |
+ memset(out, 0, *data_size); |
|
| 1050 |
+ av_log(avctx, AV_LOG_WARNING, |
|
| 1051 |
+ "G.723.1: Comfort noise generation not supported yet\n"); |
|
| 1052 |
+ return frame_size[dec_mode]; |
|
| 1053 |
+ } |
|
| 1054 |
+ |
|
| 1055 |
+ p->past_frame_type = p->cur_frame_type; |
|
| 1056 |
+ |
|
| 1057 |
+ memcpy(out, p->synth_mem, LPC_ORDER * sizeof(int16_t)); |
|
| 1058 |
+ for (i = LPC_ORDER, j = 0; j < SUBFRAMES; i += SUBFRAME_LEN, j++) |
|
| 1059 |
+ ff_celp_lp_synthesis_filter(out + i, &lpc[j * LPC_ORDER], |
|
| 1060 |
+ out + i, SUBFRAME_LEN, LPC_ORDER, |
|
| 1061 |
+ 0, 1, 1 << 12); |
|
| 1062 |
+ memcpy(p->synth_mem, out + FRAME_LEN, LPC_ORDER * sizeof(int16_t)); |
|
| 1063 |
+ |
|
| 1064 |
+ formant_postfilter(p, lpc, out); |
|
| 1065 |
+ |
|
| 1066 |
+ memmove(out, out + LPC_ORDER, *data_size); |
|
| 1067 |
+ |
|
| 1068 |
+ return frame_size[dec_mode]; |
|
| 1069 |
+} |
|
| 1070 |
+ |
|
| 1071 |
+AVCodec ff_g723_1_decoder = {
|
|
| 1072 |
+ .name = "g723_1", |
|
| 1073 |
+ .type = AVMEDIA_TYPE_AUDIO, |
|
| 1074 |
+ .id = CODEC_ID_G723_1, |
|
| 1075 |
+ .priv_data_size = sizeof(G723_1_Context), |
|
| 1076 |
+ .init = g723_1_decode_init, |
|
| 1077 |
+ .decode = g723_1_decode_frame, |
|
| 1078 |
+ .long_name = NULL_IF_CONFIG_SMALL("G.723.1"),
|
|
| 1079 |
+ .capabilities = CODEC_CAP_SUBFRAMES, |
|
| 1080 |
+}; |
| ... | ... |
@@ -21,7 +21,7 @@ |
| 21 | 21 |
#define AVCODEC_VERSION_H |
| 22 | 22 |
|
| 23 | 23 |
#define LIBAVCODEC_VERSION_MAJOR 53 |
| 24 |
-#define LIBAVCODEC_VERSION_MINOR 18 |
|
| 24 |
+#define LIBAVCODEC_VERSION_MINOR 19 |
|
| 25 | 25 |
#define LIBAVCODEC_VERSION_MICRO 0 |
| 26 | 26 |
|
| 27 | 27 |
#define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \ |
| ... | ... |
@@ -98,6 +98,7 @@ OBJS-$(CONFIG_GXF_DEMUXER) += gxf.o |
| 98 | 98 |
OBJS-$(CONFIG_GXF_MUXER) += gxfenc.o audiointerleave.o |
| 99 | 99 |
OBJS-$(CONFIG_G722_DEMUXER) += rawdec.o |
| 100 | 100 |
OBJS-$(CONFIG_G722_MUXER) += rawenc.o |
| 101 |
+OBJS-$(CONFIG_G723_1_DEMUXER) += g723_1.o |
|
| 101 | 102 |
OBJS-$(CONFIG_H261_DEMUXER) += h261dec.o rawdec.o |
| 102 | 103 |
OBJS-$(CONFIG_H261_MUXER) += rawenc.o |
| 103 | 104 |
OBJS-$(CONFIG_H263_DEMUXER) += h263dec.o rawdec.o |
| ... | ... |
@@ -100,6 +100,7 @@ void av_register_all(void) |
| 100 | 100 |
REGISTER_MUXER (FRAMECRC, framecrc); |
| 101 | 101 |
REGISTER_MUXER (FRAMEMD5, framemd5); |
| 102 | 102 |
REGISTER_MUXDEMUX (G722, g722); |
| 103 |
+ REGISTER_DEMUXER (G723_1, g723_1); |
|
| 103 | 104 |
REGISTER_MUXER (GIF, gif); |
| 104 | 105 |
REGISTER_DEMUXER (GSM, gsm); |
| 105 | 106 |
REGISTER_MUXDEMUX (GXF, gxf); |
| 106 | 107 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,83 @@ |
| 0 |
+/* |
|
| 1 |
+ * G.723.1 demuxer |
|
| 2 |
+ * Copyright (c) 2010 Mohamed Naufal Basheer |
|
| 3 |
+ * |
|
| 4 |
+ * This file is part of FFmpeg. |
|
| 5 |
+ * |
|
| 6 |
+ * FFmpeg is free software; you can redistribute it and/or |
|
| 7 |
+ * modify it under the terms of the GNU Lesser General Public |
|
| 8 |
+ * License as published by the Free Software Foundation; either |
|
| 9 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
| 10 |
+ * |
|
| 11 |
+ * FFmpeg is distributed in the hope that it will be useful, |
|
| 12 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
| 13 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
| 14 |
+ * Lesser General Public License for more details. |
|
| 15 |
+ * |
|
| 16 |
+ * You should have received a copy of the GNU Lesser General Public |
|
| 17 |
+ * License along with FFmpeg; if not, write to the Free Software |
|
| 18 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
| 19 |
+ */ |
|
| 20 |
+ |
|
| 21 |
+/** |
|
| 22 |
+ * @file |
|
| 23 |
+ * G.723.1 demuxer |
|
| 24 |
+ */ |
|
| 25 |
+ |
|
| 26 |
+#include "avformat.h" |
|
| 27 |
+ |
|
| 28 |
+static const uint8_t frame_size[4] = {24, 20, 4, 1};
|
|
| 29 |
+ |
|
| 30 |
+static int g723_1_init(AVFormatContext *s, AVFormatParameters *ap) |
|
| 31 |
+{
|
|
| 32 |
+ AVStream *st; |
|
| 33 |
+ |
|
| 34 |
+ st = av_new_stream(s, 0); |
|
| 35 |
+ if (!st) |
|
| 36 |
+ return AVERROR(ENOMEM); |
|
| 37 |
+ |
|
| 38 |
+ st->codec->codec_type = AVMEDIA_TYPE_AUDIO; |
|
| 39 |
+ st->codec->codec_id = CODEC_ID_G723_1; |
|
| 40 |
+ st->codec->channels = 1; |
|
| 41 |
+ st->codec->sample_rate = 8000; |
|
| 42 |
+ |
|
| 43 |
+ av_set_pts_info(st, 64, 1, st->codec->sample_rate); |
|
| 44 |
+ |
|
| 45 |
+ return 0; |
|
| 46 |
+} |
|
| 47 |
+ |
|
| 48 |
+static int g723_1_read_packet(AVFormatContext *s, AVPacket *pkt) |
|
| 49 |
+{
|
|
| 50 |
+ int size, byte, ret; |
|
| 51 |
+ |
|
| 52 |
+ pkt->pos = url_ftell(s->pb); |
|
| 53 |
+ byte = get_byte(s->pb); |
|
| 54 |
+ size = frame_size[byte & 3]; |
|
| 55 |
+ |
|
| 56 |
+ ret = av_new_packet(pkt, size); |
|
| 57 |
+ if (ret < 0) |
|
| 58 |
+ return ret; |
|
| 59 |
+ |
|
| 60 |
+ pkt->data[0] = byte; |
|
| 61 |
+ pkt->duration = 240; |
|
| 62 |
+ pkt->stream_index = 0; |
|
| 63 |
+ |
|
| 64 |
+ ret = get_buffer(s->pb, pkt->data + 1, size - 1); |
|
| 65 |
+ if (ret < size - 1) {
|
|
| 66 |
+ av_free_packet(pkt); |
|
| 67 |
+ return ret < 0 ? ret : AVERROR_EOF; |
|
| 68 |
+ } |
|
| 69 |
+ |
|
| 70 |
+ return pkt->size; |
|
| 71 |
+} |
|
| 72 |
+ |
|
| 73 |
+AVInputFormat ff_g723_1_demuxer = {
|
|
| 74 |
+ "g723_1", |
|
| 75 |
+ NULL_IF_CONFIG_SMALL("G.723.1 format"),
|
|
| 76 |
+ 0, |
|
| 77 |
+ NULL, |
|
| 78 |
+ g723_1_init, |
|
| 79 |
+ g723_1_read_packet, |
|
| 80 |
+ .extensions = "tco", |
|
| 81 |
+ .flags = AVFMT_GENERIC_INDEX |
|
| 82 |
+}; |
| ... | ... |
@@ -44,7 +44,7 @@ static const struct |
| 44 | 44 |
{
|
| 45 | 45 |
{0, "PCMU", AVMEDIA_TYPE_AUDIO, CODEC_ID_PCM_MULAW, 8000, 1},
|
| 46 | 46 |
{3, "GSM", AVMEDIA_TYPE_AUDIO, CODEC_ID_NONE, 8000, 1},
|
| 47 |
- {4, "G723", AVMEDIA_TYPE_AUDIO, CODEC_ID_NONE, 8000, 1},
|
|
| 47 |
+ {4, "G723", AVMEDIA_TYPE_AUDIO, CODEC_ID_G723_1, 8000, 1},
|
|
| 48 | 48 |
{5, "DVI4", AVMEDIA_TYPE_AUDIO, CODEC_ID_NONE, 8000, 1},
|
| 49 | 49 |
{6, "DVI4", AVMEDIA_TYPE_AUDIO, CODEC_ID_NONE, 16000, 1},
|
| 50 | 50 |
{7, "LPC", AVMEDIA_TYPE_AUDIO, CODEC_ID_NONE, 8000, 1},
|