blob: 87cd7cb2110c11789002637d3d67072da38b3c61
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 | #include "libavutil/channel_layout.h" |
29 | #include "libavutil/float_dsp.h" |
30 | |
31 | #define BITSTREAM_READER_LE |
32 | #include "avcodec.h" |
33 | #include "fft.h" |
34 | #include "get_bits.h" |
35 | #include "internal.h" |
36 | #include "lsp.h" |
37 | #include "sinewin.h" |
38 | |
39 | #include "twinvq.h" |
40 | #include "metasound_data.h" |
41 | |
42 | static void add_peak(float period, int width, const float *shape, |
43 | float ppc_gain, float *speech, int len) |
44 | { |
45 | int i, j, center; |
46 | const float *shape_end = shape + len; |
47 | |
48 | // First peak centered around zero |
49 | for (i = 0; i < width / 2; i++) |
50 | speech[i] += ppc_gain * *shape++; |
51 | |
52 | for (i = 1; i < ROUNDED_DIV(len, width); i++) { |
53 | center = (int)(i * period + 0.5); |
54 | for (j = -width / 2; j < (width + 1) / 2; j++) |
55 | speech[j + center] += ppc_gain * *shape++; |
56 | } |
57 | |
58 | // For the last block, be careful not to go beyond the end of the buffer |
59 | center = (int)(i * period + 0.5); |
60 | for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++) |
61 | speech[j + center] += ppc_gain * *shape++; |
62 | } |
63 | |
64 | static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef, |
65 | const float *shape, float *speech) |
66 | { |
67 | const TwinVQModeTab *mtab = tctx->mtab; |
68 | int isampf = tctx->avctx->sample_rate / 1000; |
69 | int ibps = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels); |
70 | int width; |
71 | |
72 | float ratio = (float)mtab->size / isampf; |
73 | float min_period, max_period, period_range, period; |
74 | float some_mult; |
75 | |
76 | float pgain_base, pgain_step, ppc_gain; |
77 | |
78 | if (tctx->avctx->channels == 1) { |
79 | min_period = log2(ratio * 0.2); |
80 | max_period = min_period + log2(6); |
81 | } else { |
82 | min_period = (int)(ratio * 0.2 * 400 + 0.5) / 400.0; |
83 | max_period = (int)(ratio * 0.2 * 400 * 6 + 0.5) / 400.0; |
84 | } |
85 | period_range = max_period - min_period; |
86 | period = min_period + period_coef * period_range / |
87 | ((1 << mtab->ppc_period_bit) - 1); |
88 | if (tctx->avctx->channels == 1) |
89 | period = powf(2.0, period); |
90 | else |
91 | period = (int)(period * 400 + 0.5) / 400.0; |
92 | |
93 | switch (isampf) { |
94 | case 8: some_mult = 2.0; break; |
95 | case 11: some_mult = 3.0; break; |
96 | case 16: some_mult = 3.0; break; |
97 | case 22: some_mult = ibps == 32 ? 2.0 : 4.0; break; |
98 | case 44: some_mult = 8.0; break; |
99 | default: some_mult = 4.0; |
100 | } |
101 | |
102 | width = (int)(some_mult / (mtab->size / period) * mtab->ppc_shape_len); |
103 | if (isampf == 22 && ibps == 32) |
104 | width = (int)((2.0 / period + 1) * width + 0.5); |
105 | |
106 | pgain_base = tctx->avctx->channels == 2 ? 25000.0 : 20000.0; |
107 | pgain_step = pgain_base / ((1 << mtab->pgain_bit) - 1); |
108 | ppc_gain = 1.0 / 8192 * |
109 | twinvq_mulawinv(pgain_step * g_coef + pgain_step / 2, |
110 | pgain_base, TWINVQ_PGAIN_MU); |
111 | |
112 | add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len); |
113 | } |
114 | |
115 | static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist, |
116 | int ch, float *out, float gain, |
117 | enum TwinVQFrameType ftype) |
118 | { |
119 | const TwinVQModeTab *mtab = tctx->mtab; |
120 | int i, j; |
121 | float *hist = tctx->bark_hist[ftype][ch]; |
122 | float val = ((const float []) { 0.4, 0.35, 0.28 })[ftype]; |
123 | int bark_n_coef = mtab->fmode[ftype].bark_n_coef; |
124 | int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef; |
125 | int idx = 0; |
126 | |
127 | if (tctx->avctx->channels == 1) |
128 | val = 0.5; |
129 | for (i = 0; i < fw_cb_len; i++) |
130 | for (j = 0; j < bark_n_coef; j++, idx++) { |
131 | float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] * |
132 | (1.0 / 2048); |
133 | float st; |
134 | |
135 | if (tctx->avctx->channels == 1) |
136 | st = use_hist ? |
137 | tmp2 + val * hist[idx] + 1.0 : tmp2 + 1.0; |
138 | else |
139 | st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0 |
140 | : tmp2 + 1.0; |
141 | |
142 | hist[idx] = tmp2; |
143 | if (st < 0.1) |
144 | st = 0.1; |
145 | |
146 | twinvq_memset_float(out, st * gain, |
147 | mtab->fmode[ftype].bark_tab[idx]); |
148 | out += mtab->fmode[ftype].bark_tab[idx]; |
149 | } |
150 | } |
151 | |
152 | static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb, |
153 | uint8_t *dst, enum TwinVQFrameType ftype) |
154 | { |
155 | int i; |
156 | |
157 | for (i = 0; i < tctx->n_div[ftype]; i++) { |
158 | int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]); |
159 | |
160 | *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]); |
161 | *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]); |
162 | } |
163 | } |
164 | |
165 | static int metasound_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx, |
166 | const uint8_t *buf, int buf_size) |
167 | { |
168 | TwinVQFrameData *bits; |
169 | const TwinVQModeTab *mtab = tctx->mtab; |
170 | int channels = tctx->avctx->channels; |
171 | int sub; |
172 | GetBitContext gb; |
173 | int i, j, k, ret; |
174 | |
175 | if ((ret = init_get_bits8(&gb, buf, buf_size)) < 0) |
176 | return ret; |
177 | |
178 | for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet; |
179 | tctx->cur_frame++) { |
180 | bits = tctx->bits + tctx->cur_frame; |
181 | |
182 | bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS); |
183 | |
184 | if (bits->window_type > 8) { |
185 | av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n"); |
186 | return AVERROR_INVALIDDATA; |
187 | } |
188 | |
189 | bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[tctx->cur_frame].window_type]; |
190 | |
191 | sub = mtab->fmode[bits->ftype].sub; |
192 | |
193 | if (bits->ftype != TWINVQ_FT_SHORT && !tctx->is_6kbps) |
194 | get_bits(&gb, 2); |
195 | |
196 | read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype); |
197 | |
198 | for (i = 0; i < channels; i++) |
199 | for (j = 0; j < sub; j++) |
200 | for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++) |
201 | bits->bark1[i][j][k] = |
202 | get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit); |
203 | |
204 | for (i = 0; i < channels; i++) |
205 | for (j = 0; j < sub; j++) |
206 | bits->bark_use_hist[i][j] = get_bits1(&gb); |
207 | |
208 | if (bits->ftype == TWINVQ_FT_LONG) { |
209 | for (i = 0; i < channels; i++) |
210 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
211 | } else { |
212 | for (i = 0; i < channels; i++) { |
213 | bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS); |
214 | for (j = 0; j < sub; j++) |
215 | bits->sub_gain_bits[i * sub + j] = |
216 | get_bits(&gb, TWINVQ_SUB_GAIN_BITS); |
217 | } |
218 | } |
219 | |
220 | for (i = 0; i < channels; i++) { |
221 | bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0); |
222 | bits->lpc_idx1[i] = get_bits(&gb, mtab->lsp_bit1); |
223 | |
224 | for (j = 0; j < mtab->lsp_split; j++) |
225 | bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2); |
226 | } |
227 | |
228 | if (bits->ftype == TWINVQ_FT_LONG) { |
229 | read_cb_data(tctx, &gb, bits->ppc_coeffs, 3); |
230 | for (i = 0; i < channels; i++) { |
231 | bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit); |
232 | bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit); |
233 | } |
234 | } |
235 | |
236 | // subframes are aligned to nibbles |
237 | if (get_bits_count(&gb) & 3) |
238 | skip_bits(&gb, 4 - (get_bits_count(&gb) & 3)); |
239 | } |
240 | |
241 | return (get_bits_count(&gb) + 7) / 8; |
242 | } |
243 | |
244 | typedef struct MetasoundProps { |
245 | uint32_t tag; |
246 | int bit_rate; |
247 | int channels; |
248 | int sample_rate; |
249 | } MetasoundProps; |
250 | |
251 | static const MetasoundProps codec_props[] = { |
252 | { MKTAG('V','X','0','3'), 6, 1, 8000 }, |
253 | { MKTAG('V','X','0','4'), 12, 2, 8000 }, |
254 | |
255 | { MKTAG('V','O','X','i'), 8, 1, 8000 }, |
256 | { MKTAG('V','O','X','j'), 10, 1, 11025 }, |
257 | { MKTAG('V','O','X','k'), 16, 1, 16000 }, |
258 | { MKTAG('V','O','X','L'), 24, 1, 22050 }, |
259 | { MKTAG('V','O','X','q'), 32, 1, 44100 }, |
260 | { MKTAG('V','O','X','r'), 40, 1, 44100 }, |
261 | { MKTAG('V','O','X','s'), 48, 1, 44100 }, |
262 | { MKTAG('V','O','X','t'), 16, 2, 8000 }, |
263 | { MKTAG('V','O','X','u'), 20, 2, 11025 }, |
264 | { MKTAG('V','O','X','v'), 32, 2, 16000 }, |
265 | { MKTAG('V','O','X','w'), 48, 2, 22050 }, |
266 | { MKTAG('V','O','X','x'), 64, 2, 44100 }, |
267 | { MKTAG('V','O','X','y'), 80, 2, 44100 }, |
268 | { MKTAG('V','O','X','z'), 96, 2, 44100 }, |
269 | |
270 | { 0, 0, 0, 0 } |
271 | }; |
272 | |
273 | static av_cold int metasound_decode_init(AVCodecContext *avctx) |
274 | { |
275 | int isampf, ibps; |
276 | TwinVQContext *tctx = avctx->priv_data; |
277 | uint32_t tag; |
278 | const MetasoundProps *props = codec_props; |
279 | |
280 | if (!avctx->extradata || avctx->extradata_size < 16) { |
281 | av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n"); |
282 | return AVERROR_INVALIDDATA; |
283 | } |
284 | |
285 | tag = AV_RL32(avctx->extradata + 12); |
286 | |
287 | for (;;) { |
288 | if (!props->tag) { |
289 | av_log(avctx, AV_LOG_ERROR, "Could not find tag %08"PRIX32"\n", tag); |
290 | return AVERROR_INVALIDDATA; |
291 | } |
292 | if (props->tag == tag) { |
293 | avctx->sample_rate = props->sample_rate; |
294 | avctx->channels = props->channels; |
295 | avctx->bit_rate = props->bit_rate * 1000; |
296 | isampf = avctx->sample_rate / 1000; |
297 | break; |
298 | } |
299 | props++; |
300 | } |
301 | |
302 | if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) { |
303 | av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n", |
304 | avctx->channels); |
305 | return AVERROR_INVALIDDATA; |
306 | } |
307 | avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO |
308 | : AV_CH_LAYOUT_STEREO; |
309 | |
310 | ibps = avctx->bit_rate / (1000 * avctx->channels); |
311 | |
312 | switch ((avctx->channels << 16) + (isampf << 8) + ibps) { |
313 | case (1 << 16) + ( 8 << 8) + 6: |
314 | tctx->mtab = &ff_metasound_mode0806; |
315 | break; |
316 | case (2 << 16) + ( 8 << 8) + 6: |
317 | tctx->mtab = &ff_metasound_mode0806s; |
318 | break; |
319 | case (1 << 16) + ( 8 << 8) + 8: |
320 | tctx->mtab = &ff_metasound_mode0808; |
321 | break; |
322 | case (2 << 16) + ( 8 << 8) + 8: |
323 | tctx->mtab = &ff_metasound_mode0808s; |
324 | break; |
325 | case (1 << 16) + (11 << 8) + 10: |
326 | tctx->mtab = &ff_metasound_mode1110; |
327 | break; |
328 | case (2 << 16) + (11 << 8) + 10: |
329 | tctx->mtab = &ff_metasound_mode1110s; |
330 | break; |
331 | case (1 << 16) + (16 << 8) + 16: |
332 | tctx->mtab = &ff_metasound_mode1616; |
333 | break; |
334 | case (2 << 16) + (16 << 8) + 16: |
335 | tctx->mtab = &ff_metasound_mode1616s; |
336 | break; |
337 | case (1 << 16) + (22 << 8) + 24: |
338 | tctx->mtab = &ff_metasound_mode2224; |
339 | break; |
340 | case (2 << 16) + (22 << 8) + 24: |
341 | tctx->mtab = &ff_metasound_mode2224s; |
342 | break; |
343 | case (1 << 16) + (44 << 8) + 32: |
344 | tctx->mtab = &ff_metasound_mode4432; |
345 | break; |
346 | case (2 << 16) + (44 << 8) + 32: |
347 | tctx->mtab = &ff_metasound_mode4432s; |
348 | break; |
349 | case (1 << 16) + (44 << 8) + 40: |
350 | tctx->mtab = &ff_metasound_mode4440; |
351 | break; |
352 | case (2 << 16) + (44 << 8) + 40: |
353 | tctx->mtab = &ff_metasound_mode4440s; |
354 | break; |
355 | case (1 << 16) + (44 << 8) + 48: |
356 | tctx->mtab = &ff_metasound_mode4448; |
357 | break; |
358 | case (2 << 16) + (44 << 8) + 48: |
359 | tctx->mtab = &ff_metasound_mode4448s; |
360 | break; |
361 | default: |
362 | av_log(avctx, AV_LOG_ERROR, |
363 | "This version does not support %d kHz - %d kbit/s/ch mode.\n", |
364 | isampf, ibps); |
365 | return AVERROR(ENOSYS); |
366 | } |
367 | |
368 | tctx->codec = TWINVQ_CODEC_METASOUND; |
369 | tctx->read_bitstream = metasound_read_bitstream; |
370 | tctx->dec_bark_env = dec_bark_env; |
371 | tctx->decode_ppc = decode_ppc; |
372 | tctx->frame_size = avctx->bit_rate * tctx->mtab->size |
373 | / avctx->sample_rate; |
374 | tctx->is_6kbps = ibps == 6; |
375 | |
376 | return ff_twinvq_decode_init(avctx); |
377 | } |
378 | |
379 | AVCodec ff_metasound_decoder = { |
380 | .name = "metasound", |
381 | .long_name = NULL_IF_CONFIG_SMALL("Voxware MetaSound"), |
382 | .type = AVMEDIA_TYPE_AUDIO, |
383 | .id = AV_CODEC_ID_METASOUND, |
384 | .priv_data_size = sizeof(TwinVQContext), |
385 | .init = metasound_decode_init, |
386 | .close = ff_twinvq_decode_close, |
387 | .decode = ff_twinvq_decode_frame, |
388 | .capabilities = AV_CODEC_CAP_DR1, |
389 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
390 | AV_SAMPLE_FMT_NONE }, |
391 | }; |
392 |