blob: c2353f51b56e02561ddacfec6f08ebace3e27849
1 | /* |
2 | * TwinVQ decoder |
3 | * Copyright (c) 2009 Vitor Sessak |
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 | #include <math.h> |
23 | #include <stdint.h> |
24 | |
25 | #include "libavutil/channel_layout.h" |
26 | #include "avcodec.h" |
27 | #include "get_bits.h" |
28 | #include "internal.h" |
29 | #include "twinvq.h" |
30 | #include "twinvq_data.h" |
31 | |
32 | static const TwinVQModeTab mode_08_08 = { |
33 | { |
34 | { 8, bark_tab_s08_64, 10, tab.fcb08s, 1, 5, tab.cb0808s0, tab.cb0808s1, 18 }, |
35 | { 2, bark_tab_m08_256, 20, tab.fcb08m, 2, 5, tab.cb0808m0, tab.cb0808m1, 16 }, |
36 | { 1, bark_tab_l08_512, 30, tab.fcb08l, 3, 6, tab.cb0808l0, tab.cb0808l1, 17 } |
37 | }, |
38 | 512, 12, tab.lsp08, 1, 5, 3, 3, tab.shape08, 8, 28, 20, 6, 40 |
39 | }; |
40 | |
41 | static const TwinVQModeTab mode_11_08 = { |
42 | { |
43 | { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1108s0, tab.cb1108s1, 29 }, |
44 | { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1108m0, tab.cb1108m1, 24 }, |
45 | { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1108l0, tab.cb1108l1, 27 } |
46 | }, |
47 | 512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90 |
48 | }; |
49 | |
50 | static const TwinVQModeTab mode_11_10 = { |
51 | { |
52 | { 8, bark_tab_s11_64, 10, tab.fcb11s, 1, 5, tab.cb1110s0, tab.cb1110s1, 21 }, |
53 | { 2, bark_tab_m11_256, 20, tab.fcb11m, 2, 5, tab.cb1110m0, tab.cb1110m1, 18 }, |
54 | { 1, bark_tab_l11_512, 30, tab.fcb11l, 3, 6, tab.cb1110l0, tab.cb1110l1, 20 } |
55 | }, |
56 | 512, 16, tab.lsp11, 1, 6, 4, 3, tab.shape11, 9, 36, 30, 7, 90 |
57 | }; |
58 | |
59 | static const TwinVQModeTab mode_16_16 = { |
60 | { |
61 | { 8, bark_tab_s16_128, 10, tab.fcb16s, 1, 5, tab.cb1616s0, tab.cb1616s1, 16 }, |
62 | { 2, bark_tab_m16_512, 20, tab.fcb16m, 2, 5, tab.cb1616m0, tab.cb1616m1, 15 }, |
63 | { 1, bark_tab_l16_1024, 30, tab.fcb16l, 3, 6, tab.cb1616l0, tab.cb1616l1, 16 } |
64 | }, |
65 | 1024, 16, tab.lsp16, 1, 6, 4, 3, tab.shape16, 9, 56, 60, 7, 180 |
66 | }; |
67 | |
68 | static const TwinVQModeTab mode_22_20 = { |
69 | { |
70 | { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18 }, |
71 | { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17 }, |
72 | { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18 } |
73 | }, |
74 | 1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144 |
75 | }; |
76 | |
77 | static const TwinVQModeTab mode_22_24 = { |
78 | { |
79 | { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15 }, |
80 | { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14 }, |
81 | { 1, bark_tab_l22_1024, 32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15 } |
82 | }, |
83 | 1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144 |
84 | }; |
85 | |
86 | static const TwinVQModeTab mode_22_32 = { |
87 | { |
88 | { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11 }, |
89 | { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11 }, |
90 | { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12 } |
91 | }, |
92 | 512, 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72 |
93 | }; |
94 | |
95 | static const TwinVQModeTab mode_44_40 = { |
96 | { |
97 | { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4440s0, tab.cb4440s1, 18 }, |
98 | { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4440m0, tab.cb4440m1, 17 }, |
99 | { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4440l0, tab.cb4440l1, 17 } |
100 | }, |
101 | 2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432 |
102 | }; |
103 | |
104 | static const TwinVQModeTab mode_44_48 = { |
105 | { |
106 | { 16, bark_tab_s44_128, 10, tab.fcb44s, 1, 6, tab.cb4448s0, tab.cb4448s1, 15 }, |
107 | { 4, bark_tab_m44_512, 20, tab.fcb44m, 2, 6, tab.cb4448m0, tab.cb4448m1, 14 }, |
108 | { 1, bark_tab_l44_2048, 40, tab.fcb44l, 4, 6, tab.cb4448l0, tab.cb4448l1, 14 } |
109 | }, |
110 | 2048, 20, tab.lsp44, 1, 6, 4, 4, tab.shape44, 9, 84, 54, 7, 432 |
111 | }; |
112 | |
113 | /** |
114 | * Evaluate a * b / 400 rounded to the nearest integer. When, for example, |
115 | * a * b == 200 and the nearest integer is ill-defined, use a table to emulate |
116 | * the following broken float-based implementation used by the binary decoder: |
117 | * |
118 | * @code |
119 | * static int very_broken_op(int a, int b) |
120 | * { |
121 | * static float test; // Ugh, force gcc to do the division first... |
122 | * |
123 | * test = a / 400.0; |
124 | * return b * test + 0.5; |
125 | * } |
126 | * @endcode |
127 | * |
128 | * @note if this function is replaced by just ROUNDED_DIV(a * b, 400.0), the |
129 | * stddev between the original file (before encoding with Yamaha encoder) and |
130 | * the decoded output increases, which leads one to believe that the encoder |
131 | * expects exactly this broken calculation. |
132 | */ |
133 | static int very_broken_op(int a, int b) |
134 | { |
135 | int x = a * b + 200; |
136 | int size; |
137 | const uint8_t *rtab; |
138 | |
139 | if (x % 400 || b % 5) |
140 | return x / 400; |
141 | |
142 | x /= 400; |
143 | |
144 | size = tabs[b / 5].size; |
145 | rtab = tabs[b / 5].tab; |
146 | return x - rtab[size * av_log2(2 * (x - 1) / size) + (x - 1) % size]; |
147 | } |
148 | |
149 | /** |
150 | * Sum to data a periodic peak of a given period, width and shape. |
151 | * |
152 | * @param period the period of the peak divided by 400.0 |
153 | */ |
154 | static void add_peak(int period, int width, const float *shape, |
155 | float ppc_gain, float *speech, int len) |
156 | { |
157 | int i, j; |
158 | |
159 | const float *shape_end = shape + len; |
160 | int center; |
161 | |
162 | // First peak centered around zero |
163 | for (i = 0; i < width / 2; i++) |
164 | speech[i] += ppc_gain * *shape++; |
165 | |
166 | for (i = 1; i < ROUNDED_DIV(len, width); i++) { |
167 | center = very_broken_op(period, i); |
168 | for (j = -width / 2; j < (width + 1) / 2; j++) |
169 | speech[j + center] += ppc_gain * *shape++; |
170 | } |
171 | |
172 | // For the last block, be careful not to go beyond the end of the buffer |
173 | center = very_broken_op(period, i); |
174 | for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++) |
175 | speech[j + center] += ppc_gain * *shape++; |
176 | } |
177 | |
178 | static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef, |
179 | const float *shape, float *speech) |
180 | { |
181 | const TwinVQModeTab *mtab = tctx->mtab; |
182 | int isampf = tctx->avctx->sample_rate / 1000; |
183 | int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels); |
184 | int min_period = ROUNDED_DIV(40 * 2 * mtab->size, isampf); |
185 | int max_period = ROUNDED_DIV(40 * 2 * mtab->size * 6, isampf); |
186 | int period_range = max_period - min_period; |
187 | float pgain_step = 25000.0 / ((1 << mtab->pgain_bit) - 1); |
188 | float ppc_gain = 1.0 / 8192 * |
189 | twinvq_mulawinv(pgain_step * g_coef + |
190 | pgain_step / 2, |
191 | 25000.0, TWINVQ_PGAIN_MU); |
192 | |
193 | // This is actually the period multiplied by 400. It is just linearly coded |
194 | // between its maximum and minimum value. |
195 | int period = min_period + |
196 | ROUNDED_DIV(period_coef * period_range, |
197 | (1 << mtab->ppc_period_bit) - 1); |
198 | int width; |
199 | |
200 | if (isampf == 22 && ibps == 32) { |
201 | // For some unknown reason, NTT decided to code this case differently... |
202 | width = ROUNDED_DIV((period + 800) * mtab->peak_per2wid, |
203 | 400 * mtab->size); |
204 | } else |
205 | width = period * mtab->peak_per2wid / (400 * mtab->size); |
206 | |
207 | add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len); |
208 | } |
209 | |
210 | static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist, |
211 | int ch, float *out, float gain, |
212 | enum TwinVQFrameType ftype) |
213 | { |
214 | const TwinVQModeTab *mtab = tctx->mtab; |
215 | int i, j; |
216 | float *hist = tctx->bark_hist[ftype][ch]; |
217 | float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype]; |
218 | int bark_n_coef = mtab->fmode[ftype].bark_n_coef; |
219 | int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef; |
220 | int idx = 0; |
221 | |
222 | for (i = 0; i < fw_cb_len; i++) |
223 | for (j = 0; j < bark_n_coef; j++, idx++) { |
224 | float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] * |
225 | (1.0 / 4096); |
226 | float st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0 |
227 | : tmp2 + 1.0; |
228 | |
229 | hist[idx] = tmp2; |
230 | if (st < -1.0) |
231 | st = 1.0; |
232 | |
233 | twinvq_memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]); |
234 | out += mtab->fmode[ftype].bark_tab[idx]; |
235 | } |
236 | } |
237 | |
238 | static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb, |
239 | uint8_t *dst, enum TwinVQFrameType ftype) |
240 | { |
241 | int i; |
242 | |
243 | for (i = 0; i < tctx->n_div[ftype]; i++) { |
244 | int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]); |
245 | |
246 | *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]); |
247 | *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]); |
248 | } |
249 | } |
250 | |
251 | static int twinvq_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx, |
252 | const uint8_t *buf, int buf_size) |
253 | { |
254 | TwinVQFrameData *bits = &tctx->bits[0]; |
255 | const TwinVQModeTab *mtab = tctx->mtab; |
256 | int channels = tctx->avctx->channels; |
257 | int sub; |
258 | GetBitContext gb; |
259 | int i, j, k, ret; |
260 | |
261 | if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0) |
262 | return ret; |
263 | skip_bits(&gb, get_bits(&gb, 8)); |
264 | |
265 | bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS); |
266 | |
267 | if (bits->window_type > 8) { |
268 | av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n"); |
269 | return AVERROR_INVALIDDATA; |
270 | } |
271 | |
272 | bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[0].window_type]; |
273 | |
274 | sub = mtab->fmode[bits->ftype].sub; |
275 | |
276 | read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype); |
277 | |
278 | for (i = 0; i < channels; i++) |
279 | for (j = 0; j < sub; j++) |
280 | for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++) |
281 | bits->bark1[i][j][k] = |
282 | get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit); |
283 | |
284 | for (i = 0; i < channels; i++) |
285 | for (j = 0; j < sub; j++) |
286 | bits->bark_use_hist[i][j] = get_bits1(&gb); |
287 | |
288 | if (bits->ftype == TWINVQ_FT_LONG) { |
289 | for (i = 0; i < channels; i++) |
290 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
291 | } else { |
292 | for (i = 0; i < channels; i++) { |
293 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
294 | for (j = 0; j < sub; j++) |
295 | bits->sub_gain_bits[i * sub + j] = get_bits(&gb, |
296 | TWINVQ_SUB_GAIN_BITS); |
297 | } |
298 | } |
299 | |
300 | for (i = 0; i < channels; i++) { |
301 | bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0); |
302 | bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1); |
303 | |
304 | for (j = 0; j < mtab->lsp_split; j++) |
305 | bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2); |
306 | } |
307 | |
308 | if (bits->ftype == TWINVQ_FT_LONG) { |
309 | read_cb_data(tctx, &gb, bits->ppc_coeffs, 3); |
310 | for (i = 0; i < channels; i++) { |
311 | bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit); |
312 | bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit); |
313 | } |
314 | } |
315 | |
316 | return (get_bits_count(&gb) + 7) / 8; |
317 | } |
318 | |
319 | static av_cold int twinvq_decode_init(AVCodecContext *avctx) |
320 | { |
321 | int isampf, ibps; |
322 | TwinVQContext *tctx = avctx->priv_data; |
323 | |
324 | if (!avctx->extradata || avctx->extradata_size < 12) { |
325 | av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n"); |
326 | return AVERROR_INVALIDDATA; |
327 | } |
328 | avctx->channels = AV_RB32(avctx->extradata) + 1; |
329 | avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000; |
330 | isampf = AV_RB32(avctx->extradata + 8); |
331 | |
332 | if (isampf < 8 || isampf > 44) { |
333 | av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n"); |
334 | return AVERROR_INVALIDDATA; |
335 | } |
336 | switch (isampf) { |
337 | case 44: |
338 | avctx->sample_rate = 44100; |
339 | break; |
340 | case 22: |
341 | avctx->sample_rate = 22050; |
342 | break; |
343 | case 11: |
344 | avctx->sample_rate = 11025; |
345 | break; |
346 | default: |
347 | avctx->sample_rate = isampf * 1000; |
348 | break; |
349 | } |
350 | |
351 | if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) { |
352 | av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n", |
353 | avctx->channels); |
354 | return -1; |
355 | } |
356 | avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO |
357 | : AV_CH_LAYOUT_STEREO; |
358 | |
359 | ibps = avctx->bit_rate / (1000 * avctx->channels); |
360 | if (ibps < 8 || ibps > 48) { |
361 | av_log(avctx, AV_LOG_ERROR, "Bad bitrate per channel value %d\n", ibps); |
362 | return AVERROR_INVALIDDATA; |
363 | } |
364 | |
365 | switch ((isampf << 8) + ibps) { |
366 | case (8 << 8) + 8: |
367 | tctx->mtab = &mode_08_08; |
368 | break; |
369 | case (11 << 8) + 8: |
370 | tctx->mtab = &mode_11_08; |
371 | break; |
372 | case (11 << 8) + 10: |
373 | tctx->mtab = &mode_11_10; |
374 | break; |
375 | case (16 << 8) + 16: |
376 | tctx->mtab = &mode_16_16; |
377 | break; |
378 | case (22 << 8) + 20: |
379 | tctx->mtab = &mode_22_20; |
380 | break; |
381 | case (22 << 8) + 24: |
382 | tctx->mtab = &mode_22_24; |
383 | break; |
384 | case (22 << 8) + 32: |
385 | tctx->mtab = &mode_22_32; |
386 | break; |
387 | case (44 << 8) + 40: |
388 | tctx->mtab = &mode_44_40; |
389 | break; |
390 | case (44 << 8) + 48: |
391 | tctx->mtab = &mode_44_48; |
392 | break; |
393 | default: |
394 | av_log(avctx, AV_LOG_ERROR, |
395 | "This version does not support %d kHz - %d kbit/s/ch mode.\n", |
396 | isampf, isampf); |
397 | return -1; |
398 | } |
399 | |
400 | tctx->codec = TWINVQ_CODEC_VQF; |
401 | tctx->read_bitstream = twinvq_read_bitstream; |
402 | tctx->dec_bark_env = dec_bark_env; |
403 | tctx->decode_ppc = decode_ppc; |
404 | tctx->frame_size = avctx->bit_rate * tctx->mtab->size |
405 | / avctx->sample_rate + 8; |
406 | tctx->is_6kbps = 0; |
407 | if (avctx->block_align && avctx->block_align * 8 / tctx->frame_size > 1) { |
408 | av_log(avctx, AV_LOG_ERROR, |
409 | "VQF TwinVQ should have only one frame per packet\n"); |
410 | return AVERROR_INVALIDDATA; |
411 | } |
412 | |
413 | return ff_twinvq_decode_init(avctx); |
414 | } |
415 | |
416 | AVCodec ff_twinvq_decoder = { |
417 | .name = "twinvq", |
418 | .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"), |
419 | .type = AVMEDIA_TYPE_AUDIO, |
420 | .id = AV_CODEC_ID_TWINVQ, |
421 | .priv_data_size = sizeof(TwinVQContext), |
422 | .init = twinvq_decode_init, |
423 | .close = ff_twinvq_decode_close, |
424 | .decode = ff_twinvq_decode_frame, |
425 | .capabilities = AV_CODEC_CAP_DR1, |
426 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
427 | AV_SAMPLE_FMT_NONE }, |
428 | }; |
429 |