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