| 1 | /* |
| 2 | * Voxware MetaSound decoder |
| 3 | * Copyright (c) 2013 Konstantin Shishkov |
| 4 | * based on TwinVQ decoder |
| 5 | * Copyright (c) 2009 Vitor Sessak |
| 6 | * |
| 7 | * This file is part of FFmpeg. |
| 8 | * |
| 9 | * FFmpeg is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU Lesser General Public |
| 11 | * License as published by the Free Software Foundation; either |
| 12 | * version 2.1 of the License, or (at your option) any later version. |
| 13 | * |
| 14 | * FFmpeg is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | * Lesser General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU Lesser General Public |
| 20 | * License along with FFmpeg; if not, write to the Free Software |
| 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 22 | */ |
| 23 | |
| 24 | #include <inttypes.h> |
| 25 | #include <math.h> |
| 26 | #include <stdint.h> |
| 27 | |
| 28 | #define BITSTREAM_READER_LE |
| 29 | #include "libavutil/channel_layout.h" |
| 30 | #include "libavutil/float_dsp.h" |
| 31 | #include "avcodec.h" |
| 32 | #include "get_bits.h" |
| 33 | #include "fft.h" |
| 34 | #include "internal.h" |
| 35 | #include "lsp.h" |
| 36 | #include "sinewin.h" |
| 37 | |
| 38 | #include "twinvq.h" |
| 39 | #include "metasound_data.h" |
| 40 | |
| 41 | static void add_peak(float period, int width, const float *shape, |
| 42 | float ppc_gain, float *speech, int len) |
| 43 | { |
| 44 | int i, j, center; |
| 45 | const float *shape_end = shape + len; |
| 46 | |
| 47 | // First peak centered around zero |
| 48 | for (i = 0; i < width / 2; i++) |
| 49 | speech[i] += ppc_gain * *shape++; |
| 50 | |
| 51 | for (i = 1; i < ROUNDED_DIV(len, width); i++) { |
| 52 | center = (int)(i * period + 0.5); |
| 53 | for (j = -width / 2; j < (width + 1) / 2; j++) |
| 54 | speech[j + center] += ppc_gain * *shape++; |
| 55 | } |
| 56 | |
| 57 | // For the last block, be careful not to go beyond the end of the buffer |
| 58 | center = (int)(i * period + 0.5); |
| 59 | for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++) |
| 60 | speech[j + center] += ppc_gain * *shape++; |
| 61 | } |
| 62 | |
| 63 | static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef, |
| 64 | const float *shape, float *speech) |
| 65 | { |
| 66 | const TwinVQModeTab *mtab = tctx->mtab; |
| 67 | int isampf = tctx->avctx->sample_rate / 1000; |
| 68 | int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels); |
| 69 | int width; |
| 70 | |
| 71 | float ratio = (float)mtab->size / isampf; |
| 72 | float min_period, max_period, period_range, period; |
| 73 | float some_mult; |
| 74 | |
| 75 | float pgain_base, pgain_step, ppc_gain; |
| 76 | |
| 77 | if (tctx->avctx->channels == 1) { |
| 78 | min_period = log2(ratio * 0.2); |
| 79 | max_period = min_period + log2(6); |
| 80 | } else { |
| 81 | min_period = (int)(ratio * 0.2 * 400 + 0.5) / 400.0; |
| 82 | max_period = (int)(ratio * 0.2 * 400 * 6 + 0.5) / 400.0; |
| 83 | } |
| 84 | period_range = max_period - min_period; |
| 85 | period = min_period + period_coef * period_range / |
| 86 | ((1 << mtab->ppc_period_bit) - 1); |
| 87 | if (tctx->avctx->channels == 1) |
| 88 | period = powf(2.0, period); |
| 89 | else |
| 90 | period = (int)(period * 400 + 0.5) / 400.0; |
| 91 | |
| 92 | switch (isampf) { |
| 93 | case 8: some_mult = 2.0; break; |
| 94 | case 11: some_mult = 3.0; break; |
| 95 | case 16: some_mult = 3.0; break; |
| 96 | case 22: some_mult = ibps == 32 ? 2.0 : 4.0; break; |
| 97 | case 44: some_mult = 8.0; break; |
| 98 | default: some_mult = 4.0; |
| 99 | } |
| 100 | |
| 101 | width = (int)(some_mult / (mtab->size / period) * mtab->ppc_shape_len); |
| 102 | if (isampf == 22 && ibps == 32) |
| 103 | width = (int)((2.0 / period + 1) * width + 0.5); |
| 104 | |
| 105 | pgain_base = tctx->avctx->channels == 2 ? 25000.0 : 20000.0; |
| 106 | pgain_step = pgain_base / ((1 << mtab->pgain_bit) - 1); |
| 107 | ppc_gain = 1.0 / 8192 * |
| 108 | twinvq_mulawinv(pgain_step * g_coef + pgain_step / 2, |
| 109 | pgain_base, TWINVQ_PGAIN_MU); |
| 110 | |
| 111 | add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len); |
| 112 | } |
| 113 | |
| 114 | static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist, |
| 115 | int ch, float *out, float gain, |
| 116 | enum TwinVQFrameType ftype) |
| 117 | { |
| 118 | const TwinVQModeTab *mtab = tctx->mtab; |
| 119 | int i, j; |
| 120 | float *hist = tctx->bark_hist[ftype][ch]; |
| 121 | float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype]; |
| 122 | int bark_n_coef = mtab->fmode[ftype].bark_n_coef; |
| 123 | int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef; |
| 124 | int idx = 0; |
| 125 | |
| 126 | if (tctx->avctx->channels == 1) |
| 127 | val = 0.5; |
| 128 | for (i = 0; i < fw_cb_len; i++) |
| 129 | for (j = 0; j < bark_n_coef; j++, idx++) { |
| 130 | float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] * |
| 131 | (1.0 / 2048); |
| 132 | float st; |
| 133 | |
| 134 | if (tctx->avctx->channels == 1) |
| 135 | st = use_hist ? |
| 136 | tmp2 + val * hist[idx] + 1.0 : tmp2 + 1.0; |
| 137 | else |
| 138 | st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0 |
| 139 | : tmp2 + 1.0; |
| 140 | |
| 141 | hist[idx] = tmp2; |
| 142 | if (st < 0.1) |
| 143 | st = 0.1; |
| 144 | |
| 145 | twinvq_memset_float(out, st * gain, |
| 146 | mtab->fmode[ftype].bark_tab[idx]); |
| 147 | out += mtab->fmode[ftype].bark_tab[idx]; |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb, |
| 152 | uint8_t *dst, enum TwinVQFrameType ftype) |
| 153 | { |
| 154 | int i; |
| 155 | |
| 156 | for (i = 0; i < tctx->n_div[ftype]; i++) { |
| 157 | int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]); |
| 158 | |
| 159 | *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]); |
| 160 | *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]); |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | static int metasound_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx, |
| 165 | const uint8_t *buf, int buf_size) |
| 166 | { |
| 167 | TwinVQFrameData *bits; |
| 168 | const TwinVQModeTab *mtab = tctx->mtab; |
| 169 | int channels = tctx->avctx->channels; |
| 170 | int sub; |
| 171 | GetBitContext gb; |
| 172 | int i, j, k; |
| 173 | |
| 174 | init_get_bits(&gb, buf, buf_size * 8); |
| 175 | |
| 176 | for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet; |
| 177 | tctx->cur_frame++) { |
| 178 | bits = tctx->bits + tctx->cur_frame; |
| 179 | |
| 180 | bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS); |
| 181 | |
| 182 | if (bits->window_type > 8) { |
| 183 | av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n"); |
| 184 | return AVERROR_INVALIDDATA; |
| 185 | } |
| 186 | |
| 187 | bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[tctx->cur_frame].window_type]; |
| 188 | |
| 189 | sub = mtab->fmode[bits->ftype].sub; |
| 190 | |
| 191 | if (bits->ftype != TWINVQ_FT_SHORT && !tctx->is_6kbps) |
| 192 | get_bits(&gb, 2); |
| 193 | |
| 194 | read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype); |
| 195 | |
| 196 | for (i = 0; i < channels; i++) |
| 197 | for (j = 0; j < sub; j++) |
| 198 | for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++) |
| 199 | bits->bark1[i][j][k] = |
| 200 | get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit); |
| 201 | |
| 202 | for (i = 0; i < channels; i++) |
| 203 | for (j = 0; j < sub; j++) |
| 204 | bits->bark_use_hist[i][j] = get_bits1(&gb); |
| 205 | |
| 206 | if (bits->ftype == TWINVQ_FT_LONG) { |
| 207 | for (i = 0; i < channels; i++) |
| 208 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
| 209 | } else { |
| 210 | for (i = 0; i < channels; i++) { |
| 211 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
| 212 | for (j = 0; j < sub; j++) |
| 213 | bits->sub_gain_bits[i * sub + j] = |
| 214 | get_bits(&gb, TWINVQ_SUB_GAIN_BITS); |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | for (i = 0; i < channels; i++) { |
| 219 | bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0); |
| 220 | bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1); |
| 221 | |
| 222 | for (j = 0; j < mtab->lsp_split; j++) |
| 223 | bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2); |
| 224 | } |
| 225 | |
| 226 | if (bits->ftype == TWINVQ_FT_LONG) { |
| 227 | read_cb_data(tctx, &gb, bits->ppc_coeffs, 3); |
| 228 | for (i = 0; i < channels; i++) { |
| 229 | bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit); |
| 230 | bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit); |
| 231 | } |
| 232 | } |
| 233 | |
| 234 | // subframes are aligned to nibbles |
| 235 | if (get_bits_count(&gb) & 3) |
| 236 | skip_bits(&gb, 4 - (get_bits_count(&gb) & 3)); |
| 237 | } |
| 238 | |
| 239 | return (get_bits_count(&gb) + 7) / 8; |
| 240 | } |
| 241 | |
| 242 | typedef struct MetasoundProps { |
| 243 | uint32_t tag; |
| 244 | int bit_rate; |
| 245 | int channels; |
| 246 | int sample_rate; |
| 247 | } MetasoundProps; |
| 248 | |
| 249 | static const MetasoundProps codec_props[] = { |
| 250 | { MKTAG('V','X','0','3'), 6, 1, 8000 }, |
| 251 | { MKTAG('V','X','0','4'), 12, 2, 8000 }, |
| 252 | |
| 253 | { MKTAG('V','O','X','i'), 8, 1, 8000 }, |
| 254 | { MKTAG('V','O','X','j'), 10, 1, 11025 }, |
| 255 | { MKTAG('V','O','X','k'), 16, 1, 16000 }, |
| 256 | { MKTAG('V','O','X','L'), 24, 1, 22050 }, |
| 257 | { MKTAG('V','O','X','q'), 32, 1, 44100 }, |
| 258 | { MKTAG('V','O','X','r'), 40, 1, 44100 }, |
| 259 | { MKTAG('V','O','X','s'), 48, 1, 44100 }, |
| 260 | { MKTAG('V','O','X','t'), 16, 2, 8000 }, |
| 261 | { MKTAG('V','O','X','u'), 20, 2, 11025 }, |
| 262 | { MKTAG('V','O','X','v'), 32, 2, 16000 }, |
| 263 | { MKTAG('V','O','X','w'), 48, 2, 22050 }, |
| 264 | { MKTAG('V','O','X','x'), 64, 2, 44100 }, |
| 265 | { MKTAG('V','O','X','y'), 80, 2, 44100 }, |
| 266 | { MKTAG('V','O','X','z'), 96, 2, 44100 }, |
| 267 | |
| 268 | { 0, 0, 0, 0 } |
| 269 | }; |
| 270 | |
| 271 | static av_cold int metasound_decode_init(AVCodecContext *avctx) |
| 272 | { |
| 273 | int isampf, ibps; |
| 274 | TwinVQContext *tctx = avctx->priv_data; |
| 275 | uint32_t tag; |
| 276 | const MetasoundProps *props = codec_props; |
| 277 | |
| 278 | if (!avctx->extradata || avctx->extradata_size < 16) { |
| 279 | av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n"); |
| 280 | return AVERROR_INVALIDDATA; |
| 281 | } |
| 282 | |
| 283 | tag = AV_RL32(avctx->extradata + 12); |
| 284 | |
| 285 | for (;;) { |
| 286 | if (!props->tag) { |
| 287 | av_log(avctx, AV_LOG_ERROR, "Could not find tag %08"PRIX32"\n", tag); |
| 288 | return AVERROR_INVALIDDATA; |
| 289 | } |
| 290 | if (props->tag == tag) { |
| 291 | avctx->sample_rate = props->sample_rate; |
| 292 | avctx->channels = props->channels; |
| 293 | avctx->bit_rate = props->bit_rate * 1000; |
| 294 | isampf = avctx->sample_rate / 1000; |
| 295 | break; |
| 296 | } |
| 297 | props++; |
| 298 | } |
| 299 | |
| 300 | if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) { |
| 301 | av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n", |
| 302 | avctx->channels); |
| 303 | return AVERROR_INVALIDDATA; |
| 304 | } |
| 305 | avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO |
| 306 | : AV_CH_LAYOUT_STEREO; |
| 307 | |
| 308 | ibps = avctx->bit_rate / (1000 * avctx->channels); |
| 309 | |
| 310 | switch ((avctx->channels << 16) + (isampf << 8) + ibps) { |
| 311 | case (1 << 16) + ( 8 << 8) + 6: |
| 312 | tctx->mtab = &ff_metasound_mode0806; |
| 313 | break; |
| 314 | case (2 << 16) + ( 8 << 8) + 6: |
| 315 | tctx->mtab = &ff_metasound_mode0806s; |
| 316 | break; |
| 317 | case (1 << 16) + ( 8 << 8) + 8: |
| 318 | tctx->mtab = &ff_metasound_mode0808; |
| 319 | break; |
| 320 | case (2 << 16) + ( 8 << 8) + 8: |
| 321 | tctx->mtab = &ff_metasound_mode0808s; |
| 322 | break; |
| 323 | case (1 << 16) + (11 << 8) + 10: |
| 324 | tctx->mtab = &ff_metasound_mode1110; |
| 325 | break; |
| 326 | case (2 << 16) + (11 << 8) + 10: |
| 327 | tctx->mtab = &ff_metasound_mode1110s; |
| 328 | break; |
| 329 | case (1 << 16) + (16 << 8) + 16: |
| 330 | tctx->mtab = &ff_metasound_mode1616; |
| 331 | break; |
| 332 | case (2 << 16) + (16 << 8) + 16: |
| 333 | tctx->mtab = &ff_metasound_mode1616s; |
| 334 | break; |
| 335 | case (1 << 16) + (22 << 8) + 24: |
| 336 | tctx->mtab = &ff_metasound_mode2224; |
| 337 | break; |
| 338 | case (2 << 16) + (22 << 8) + 24: |
| 339 | tctx->mtab = &ff_metasound_mode2224s; |
| 340 | break; |
| 341 | case (1 << 16) + (44 << 8) + 32: |
| 342 | tctx->mtab = &ff_metasound_mode4432; |
| 343 | break; |
| 344 | case (2 << 16) + (44 << 8) + 32: |
| 345 | tctx->mtab = &ff_metasound_mode4432s; |
| 346 | break; |
| 347 | case (1 << 16) + (44 << 8) + 40: |
| 348 | tctx->mtab = &ff_metasound_mode4440; |
| 349 | break; |
| 350 | case (2 << 16) + (44 << 8) + 40: |
| 351 | tctx->mtab = &ff_metasound_mode4440s; |
| 352 | break; |
| 353 | case (1 << 16) + (44 << 8) + 48: |
| 354 | tctx->mtab = &ff_metasound_mode4448; |
| 355 | break; |
| 356 | case (2 << 16) + (44 << 8) + 48: |
| 357 | tctx->mtab = &ff_metasound_mode4448s; |
| 358 | break; |
| 359 | default: |
| 360 | av_log(avctx, AV_LOG_ERROR, |
| 361 | "This version does not support %d kHz - %d kbit/s/ch mode.\n", |
| 362 | isampf, ibps); |
| 363 | return AVERROR(ENOSYS); |
| 364 | } |
| 365 | |
| 366 | tctx->codec = TWINVQ_CODEC_METASOUND; |
| 367 | tctx->read_bitstream = metasound_read_bitstream; |
| 368 | tctx->dec_bark_env = dec_bark_env; |
| 369 | tctx->decode_ppc = decode_ppc; |
| 370 | tctx->frame_size = avctx->bit_rate * tctx->mtab->size |
| 371 | / avctx->sample_rate; |
| 372 | tctx->is_6kbps = ibps == 6; |
| 373 | |
| 374 | return ff_twinvq_decode_init(avctx); |
| 375 | } |
| 376 | |
| 377 | AVCodec ff_metasound_decoder = { |
| 378 | .name = "metasound", |
| 379 | .long_name = NULL_IF_CONFIG_SMALL("Voxware MetaSound"), |
| 380 | .type = AVMEDIA_TYPE_AUDIO, |
| 381 | .id = AV_CODEC_ID_METASOUND, |
| 382 | .priv_data_size = sizeof(TwinVQContext), |
| 383 | .init = metasound_decode_init, |
| 384 | .close = ff_twinvq_decode_close, |
| 385 | .decode = ff_twinvq_decode_frame, |
| 386 | .capabilities = CODEC_CAP_DR1, |
| 387 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
| 388 | AV_SAMPLE_FMT_NONE }, |
| 389 | }; |