2 * Voxware MetaSound decoder
3 * Copyright (c) 2013 Konstantin Shishkov
4 * based on TwinVQ decoder
5 * Copyright (c) 2009 Vitor Sessak
7 * This file is part of FFmpeg.
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.
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.
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
27 #define BITSTREAM_READER_LE
28 #include "libavutil/channel_layout.h"
29 #include "libavutil/float_dsp.h"
38 #include "metasound_data.h"
40 static void add_peak(float period, int width, const float *shape,
41 float ppc_gain, float *speech, int len)
44 const float *shape_end = shape + len;
46 // First peak centered around zero
47 for (i = 0; i < width / 2; i++)
48 speech[i] += ppc_gain * *shape++;
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++;
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++;
62 static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef,
63 const float *shape, float *speech)
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);
70 float ratio = (float)mtab->size / isampf;
71 float min_period, max_period, period_range, period;
74 float pgain_base, pgain_step, ppc_gain;
76 if (tctx->avctx->channels == 1) {
77 min_period = log2(ratio * 0.2);
78 max_period = min_period + log2(6);
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;
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);
89 period = (int)(period * 400 + 0.5) / 400.0;
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;
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);
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);
110 add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
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)
117 const TwinVQModeTab *mtab = tctx->mtab;
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;
125 if (tctx->avctx->channels == 1)
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] *
133 if (tctx->avctx->channels == 1)
135 tmp2 + val * hist[idx] + 1.0 : tmp2 + 1.0;
137 st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
144 twinvq_memset_float(out, st * gain,
145 mtab->fmode[ftype].bark_tab[idx]);
146 out += mtab->fmode[ftype].bark_tab[idx];
150 static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb,
151 uint8_t *dst, enum TwinVQFrameType ftype)
155 for (i = 0; i < tctx->n_div[ftype]; i++) {
156 int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);
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]);
163 static int metasound_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
164 const uint8_t *buf, int buf_size)
166 TwinVQFrameData *bits;
167 const TwinVQModeTab *mtab = tctx->mtab;
168 int channels = tctx->avctx->channels;
173 init_get_bits(&gb, buf, buf_size * 8);
175 for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet;
177 bits = tctx->bits + tctx->cur_frame;
179 bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
181 if (bits->window_type > 8) {
182 av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
183 return AVERROR_INVALIDDATA;
186 bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[tctx->cur_frame].window_type];
188 sub = mtab->fmode[bits->ftype].sub;
190 if (bits->ftype != TWINVQ_FT_SHORT && !tctx->is_6kbps)
193 read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
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);
201 for (i = 0; i < channels; i++)
202 for (j = 0; j < sub; j++)
203 bits->bark_use_hist[i][j] = get_bits1(&gb);
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);
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] =
213 get_bits(&gb, TWINVQ_SUB_GAIN_BITS);
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);
221 for (j = 0; j < mtab->lsp_split; j++)
222 bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
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);
233 // subframes are aligned to nibbles
234 if (get_bits_count(&gb) & 3)
235 skip_bits(&gb, 4 - (get_bits_count(&gb) & 3));
238 return (get_bits_count(&gb) + 7) / 8;
241 typedef struct MetasoundProps {
248 static const MetasoundProps codec_props[] = {
249 { MKTAG('V','X','0','3'), 6, 1, 8000 },
250 { MKTAG('V','X','0','4'), 12, 2, 8000 },
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 },
270 static av_cold int metasound_decode_init(AVCodecContext *avctx)
273 TwinVQContext *tctx = avctx->priv_data;
275 const MetasoundProps *props = codec_props;
277 if (!avctx->extradata || avctx->extradata_size < 16) {
278 av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
279 return AVERROR_INVALIDDATA;
282 tag = AV_RL32(avctx->extradata + 12);
286 av_log(avctx, AV_LOG_ERROR, "Could not find tag %08X\n", tag);
287 return AVERROR_INVALIDDATA;
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;
299 if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
300 av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
302 return AVERROR_INVALIDDATA;
304 avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
305 : AV_CH_LAYOUT_STEREO;
307 ibps = avctx->bit_rate / (1000 * avctx->channels);
309 switch ((avctx->channels << 16) + (isampf << 8) + ibps) {
310 case (1 << 16) + ( 8 << 8) + 6:
311 tctx->mtab = &ff_metasound_mode0806;
313 case (2 << 16) + ( 8 << 8) + 6:
314 tctx->mtab = &ff_metasound_mode0806s;
316 case (1 << 16) + ( 8 << 8) + 8:
317 tctx->mtab = &ff_metasound_mode0808;
319 case (2 << 16) + ( 8 << 8) + 8:
320 tctx->mtab = &ff_metasound_mode0808s;
322 case (1 << 16) + (11 << 8) + 10:
323 tctx->mtab = &ff_metasound_mode1110;
325 case (2 << 16) + (11 << 8) + 10:
326 tctx->mtab = &ff_metasound_mode1110s;
328 case (1 << 16) + (16 << 8) + 16:
329 tctx->mtab = &ff_metasound_mode1616;
331 case (2 << 16) + (16 << 8) + 16:
332 tctx->mtab = &ff_metasound_mode1616s;
334 case (1 << 16) + (22 << 8) + 24:
335 tctx->mtab = &ff_metasound_mode2224;
337 case (2 << 16) + (22 << 8) + 24:
338 tctx->mtab = &ff_metasound_mode2224s;
340 case (1 << 16) + (44 << 8) + 32:
341 tctx->mtab = &ff_metasound_mode4432;
343 case (2 << 16) + (44 << 8) + 32:
344 tctx->mtab = &ff_metasound_mode4432s;
346 case (1 << 16) + (44 << 8) + 40:
347 tctx->mtab = &ff_metasound_mode4440;
349 case (2 << 16) + (44 << 8) + 40:
350 tctx->mtab = &ff_metasound_mode4440s;
352 case (1 << 16) + (44 << 8) + 48:
353 tctx->mtab = &ff_metasound_mode4448;
355 case (2 << 16) + (44 << 8) + 48:
356 tctx->mtab = &ff_metasound_mode4448s;
359 av_log(avctx, AV_LOG_ERROR,
360 "This version does not support %d kHz - %d kbit/s/ch mode.\n",
362 return AVERROR(ENOSYS);
365 tctx->codec = TWINVQ_CODEC_METASOUND;
366 tctx->read_bitstream = metasound_read_bitstream;
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;
373 return ff_twinvq_decode_init(avctx);
376 AVCodec ff_metasound_decoder = {
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),
382 .init = metasound_decode_init,
383 .close = ff_twinvq_decode_close,
384 .decode = ff_twinvq_decode_frame,
385 .capabilities = CODEC_CAP_DR1,
386 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
387 AV_SAMPLE_FMT_NONE },